The Windows service will run as the user account LocalSystem and be configured for automatic start at machine reboot. You can change these values using the Windows MMC Administrative Tools Services snap-in. The daemon will run the default configured OpenSplice domain. There is an issue with log files when running under a windows service. See the Known Issues
New DDS Examples have been added for enterprise platforms (see the examples directory)
So far the OpenSplice DDS durability service was not able to handle the so-called "split-brain syndrome". A split-brain syndrome can be described as the situation in which two different nodes (possibly) have a different perception on (part of) the set of historical data. This split-brain occurs when two nodes or two sets of nodes (i.e. two systems) that are participating in the same DDS domain have been running separately for some time and suddenly get connected to each other. Equally this syndrome arises when nodes re-connect after being disconnected for some time. Applications on these nodes may have been publishing information for the same topic in the same partition without this information reaching the other party. Therefore their perception on the set of data will be different. Because of the new emerging environments in which OpenSplice DDS is deployed, the durability service has been extended with additional functionality to support the alignment of historical data when two nodes get connected. Of course, the basic use case of a newly started node joining an existing system is supported, but in contradiction to that situation there is no universal truth in determining who has the best (or the right) information when two already running nodes (re)connect. When this situation occurs, the durability service now provides the possibility to configure its behaviour in this situation in the form of so-called merge policies. The following options are supported:
The RunTime System Monitor (RTSM) is a commandline tool that allows monitoring of metrics and statistics provided by OpenSplice. It can show delta's on for example reader-statistics and reset them. The tool is currently in a beta-stage and is only delivered with Linux-based builds. The tool can be driven through an interactive menu, or by passing options/commands through the command line or a file.
The tool contains explanation of the available options, which can
be displayed by calling 'rtsm -h'. If you have a domain running
with the name "OpenSpliceWithStats", where statistics for readers
are enabled, you can issue the following command:
> rtsm -domain OpenSpliceWithStats -count 5 -readerSort -exit
It will first set the domain to "OpenSpliceWithStats". It will set
the number of iterations to show statistics to 5, and then run a
readerSort. When finished it will exit.
The same can be achieved in the interactive mode by calling:
> rtsm
Then type '4
Then type '17
Then type the domain-name: 'OpenSpliceWithStats
Then type '13
Then type the count: '5
Then type '0
Then type '1
Then, when finished, type '0
Interactive and commandline-mode can also be combined.
When rebuilding the CORBA Java interface, PERC may be used instead of the Sun JDK. Source the percenv.sh script and set SPLICE_JDK=perc before running the OpenSplice release.com. This will also cause the example build and run scripts to use PERC.
The durability service now has the capability to discover namespaces from other nodes at runtime. This can significantly reduce configuration effort in large-scale systems, while also enabling OpenSplice to run in more dynamic environments.
For C#, there is now a signed and versioned Assembly called dcpssacsAssembly.dll, created from the the dcpssacs.dll module, which is added to the Global Assembly cache. In this release, as part of the installation, this is added to the global assembly cache.
See the OpenSplice Documentation Set for detailed information.
Implementation available for the coherent access PRESENTATION QoS policy for INSTANCE (existing)
and TOPIC (new) access_scopes
See the OpenSplice Documentation Set for detailed information.
This functionality allows an application to dispose of all of the instances
for a particular topic without the overhead of using an individual
dispose call for each instance individually. Its effect is equivalent to
invoking a separate dispose operation for each individual instance
on the DataWriter that owns it
See the OpenSplice Documentation Set for detailed information.
Added Windows 7 32 bit support with Microsoft Visual Studio 2008.
OpenFusion Real Time ORB Java Edition is now tested with OpenSplice DDS. Additional examples have been provided to allow users to get started quickly. In a future release documentation updates will be made to the manuals.
With Linux or Unix platforms, it is possible to specify the symbolic name of a networking interface to be used, like eth0, eth1, wlan0 etc. In this release this concept is now supported on Windows using "friendly names", which are unique symbolic names tied to each interface.
OpenSplice DDS V5.x is compiled with JDK1.6, but can run on a Java 5 JRE as the compile flag -source 1.5 is set
OpenSplice DDS can be configured to use zlib compression www.zlib.net to compress network packets. See the OpenSplice Documentation set for further information.
OpenSplice DDS can be configured to use an external reporting library. See the OpenSplice Documentation set for further information. This functionality is not available on the embedded platforms (VxWorks or Integrity)
A DomainId consists of a string that represents either a URI to the location of the configuration file (e.g. 'file:///projects/DDS/ospl.xml') or the Domain name as specified in the configuration file. The actual value returned is dependent of the value used when creating the DomainParticipant, also see the DomainParticipantFactory create_participant operation. If a DomainParticipant is created using the Domain name then it will also return the Domain name and visa-versa when created using a URI then the URI will be returned by this operation. The configuration file, identified by the URI, specifies all configuration details of the Domain
In the V4 series OpenSplice DDS used system wide environment variables. With OpenSplice DDS V5.1, these are not set. Note: Users will require OpenSplice DDS PowerTools 2.3 to cope with this change.
The existing topology-discovery mechanism of OpenSplice's Networking Service is extended with a dynamic capability to discover and maintain all relevant nodes in large-scale (WAN) systems. In such large-scale systems, static-unicast node pre-definition is not practical nor sufficient and dynamic-multicast based discovery is typically not supported or even allowed. The new dynamic discovery is based on overlaying the physical network with a notion of 'Roles' and 'Scopes' that represent 'area-of-interest regions' within the WAN domain such that only nodes within the defined scope-of-interest will be automatically discovered and their state maintained. In this way, the scalability of the dynamic DDS-system is ensured whilst minimizing the impact on the overhead of the required discovery process.
For more information see the NetworkService elements Role, Discovery, Scope and ProbeList in the Deployment Guide.
Report ID. | Description |
---|---|
OSPL-1023 |
Service failure actions can be taken multiple times When multiple services are known to the service framework, a failure action can be taken multiple times. Solution: The defect in the service failure action algorithm is now fixed and the action will only be done once. |
OSPL-1537/ 11089 OSPL-1966/ 11443 |
spliced may crash after a service with the "systemhalt" failure action dies If a service dies it may not have performed a detach from the OpenSplice kernel/database and so the attached services count may be incorrect. That could lead to a crash of a database thread because the spliced may get detached too early. Solution: When spliced detects that a service has died, it ensures that attached services count is correctly maintained, so such a crash cannot occur. |
OSPL-991/ 10735 OSPL-1771/ 11395 |
take() and take_w_condition() do not have the same behaviour/random
crash on take next instance The dispose_all_data operation on the topic was not treated identically to sending a separate dispose message for every entity. This manifested itself especially in the way the disposed data was delivered to late joiners (which sometimes couldn't see that the data was actually disposed) and in the way disposed data ignored the cleanup delay specified on the durability service. Solution: The new implementation of dispose_all_data is much more inline with sending separate dispose messages for every instance, and thus late joiners will always see the correct instance state. Furthermore the dispose messages obey the same cleanup delays as normal dispose messages. However, the dispose_all_data function will still be me more efficient with respect to the utilization of network network bandwitdth, CPU cycles and memory than the manual transmission of separate dispose messages. |
Report ID. | Description |
---|---|
OSPL-1341 / 10914 OSPL-1930 / 11473 |
The reliable network communication may not operate correctly when the
first messages of a sending node arrive out of order. When reconnection is enabled and when the first messages of another starting node arrive out of order and also the discovery heartbeats arrive later then the first message the reliable network channel will not operate correctly. Solution: The notification that a node has become alive by the discovery protocol should not reinitialize the reliable channel administration associated with that node when the reliable channel had already detected that the node was alive. |
OSPL-1923 / 11472 |
Application not detached from shm after deletion of all the
participants. When an application creates 2 or more participants and all these participants are deleted using the delete_participant function, the application still hold a reference to the shared memory. Solution: The defect is now fixed and when all participants from an application are gone the connection to the shared memory is also gone. |
OSPL-1936 / 11509 |
Reporting/tracing of networking service could be improved In case the networking service is configured to allow re-connections, it should report when a remote node re-connects. In addition, the networking service should also report the channel when reporting a missing packet in its trace (if configured) as well as report when and if the missing packet is received to be able to find out if the service recovered from the missing packet or not. Solution: The reporting and tracing extensions requested above have been added to the networking service. |
OSPL-1946 / 11512 |
With compression or encryption enabled lost packets may not be resent In some situations networking would try to access the compressed and/or encrypted content of a packet in its resend-administration, causing packets to not be re-transmitted. Solution: All information needed for (re-)sending of a packet are read from the packet-buffer before compression and/or encryption. |
Report ID. | Description |
---|---|
OSPL-1557/ 10734 |
Crash of durability during initial alignment Unregistrations that are aligned by the durability service are stored without any data to reduce footprint. These unregistrations can only be re-published locally when an instance handle is provided by the durability service while doing so. In some scenario's, the durability service did not use an instance handle while doing so, which made the service crash while locally republishing an aligned unregistration. Solution: The durability service now ensures an instance handle is created in any case while also making sure a registration is created in the instance for the writer that originally wrote the aligned sample. |
OSPL-1682/ 11291 |
crash of spliced In certain scenario's the reader purgeList was doing invalid memory reads by trying to access already deleted purgeItems, thereby causing potential memory corruption of totally unrelated objects. Solution: The purge algorithm has been modified to prevent this situation, thus preventing the memory from becoming corrupted and improving the overall stability of the system. |
OSPL-1852 |
Topic disappearance when topic created in parallel In a scenario where a specific topic was created for the first time in the system, but for which a duplicate was created by another application before the original could enable its topic, a refCount to the resulting topic got dropped and that topic could suddenly disappear while still being used by the system. Solution: A change in the refCounting algorithm has now solved this issue. |
OSPL-1867/ 11443 |
System termination fallback mechanism When a service crashes and system termination is set into progress, a safe system termination is not always guaranteed i.e. the service could end up in a deadlock stalling system termination. This is not acceptable and the system should always terminate. Solution: A new service termination thread is spawned when the system state of a process is set to terminating. This thread waits for 5 seconds to see if the process goes from the terminating state to the terminated state. If this has not happened after 5 seconds the process is being killed by means of _exit. |
OSPL-1882 |
Race condition in parallel demarshalling When the number of threads used for parallel demarshalling is changed, a deadlock condition can occur. Solution: The failing thread synchronization has been fixed. |
Report ID. | Description |
---|---|
OSPL-1491/ 8957 |
In the event of an application receiving a SIGSEGV, listener threads
can spin and write repetitive messages to the ospl-error.log When the listener is accessing the internal waitset representation of its list of events, it was not checking to see whether the waitset failed because the DDS database is being detached. That led to the listener reporting an error but then continuing in its while loop. Solution: The listener thread now correctly checks the status return from the waitset operation and will terminate its while loop in the case that it is in a detaching state. This will mean that the listener thread will terminate elegantly when a SIGSEGV is received within the application. |
OSPL-1620 |
Generated code crashes when sequences are not set correctly. When a sequence is set to have a non-zero length, but it's buffer-field is not set, the generated code would crash. Solution: The generated code will now verify that for non-zero length sequences, an allocated buffer is present. |
OSPL-1669 |
Merged historical data is not delivered to active datareaders. When configured for merging after re-connecting, the durability service does not deliver the aligned samples to existing data-readers. Only newly created data-readers from that point forward will get the data. This is caused by some internal optimisation mechanism between the transient store and existing data-readers. The connection between transient store and existing data-readers can be closed in some situation where remote nodes disconnect and the durability service does not re-instate this connection when the node reconnects. Solution: The connection between transient store and existing readers is now checked and re-instated in case a node reconnects. |
OSPL-1690 |
Memory leakage when reading a topic without keys With circular read, a bug was introduced which would quickly leak large amounts of memory and would hang the process when reading a topic without keys. Solution: The memory leak has been fixed. |
Report ID. | Description |
---|---|
OSPL-1166/ 10823 |
Terminating the "ospl -f start" operation may not kill all services The algorithm in ospl that performs the shutdown of the splice daemon and its child services in the case of blocking mode did not correctly detect whether the splice daemon had been terminated. This meant that it was possible that the splice daemon and its services may not necessarily have been terminated when "ospl -f start" returns. Solution: The shutdown algorithm in ospl has been improved to correctly detect the termination status of the splice daemon in both normal and blocking modes. |
OSPL-1172/ 10827 |
Performance difference between the datareader listener and subscriber listener In certain scenarios the subscriber listener handling is faster than the datareader listener handling. Solution: The handling algorithm of the datareader listener is improved to match the subscriber handling. |
OSPL-1384/ 10907 |
OpenSplice logs errors when the XML configuration file contains DOCTYPE descriptors Solution: The code that logged the errors has been removed. |
OSPL-1409/ 10785 |
Durability sometimes reported "No matching policy for nameSpace AutoBuiltinPartition" The message "No matching policy for nameSpace AutoBuiltinPartition" sometimes appeared in the durability logging. This happened because durability created a temporary namespace object which had no associated policy. Functionally there was no issue, but the log message is confusing. Solution: Durability won't lookup a policy anymore when the temporary namespace object is created. |
OSPL-1478/ 11042 |
BOUNDS_CHECK does not report the name of the incorrect member when the member is null. The copyin routines for C and C++ did not check for null-values in string fields. Solution: Check for null-values in string-fields is added. |
OSPL-1558 |
Durability sometimes creates conflicting namespace for __BUILT-IN
PARTITION__ when partitionTopic is used in namespace-definition. Durability could automatically create conflicting namespace for __BUILT-IN PARTITION__ when partitionTopic is used. Solution: Durability now only creates a namespace in which the builtin-topics are matched, instead of matching the whole builtin-partition. |
Report ID. | Description |
---|---|
OSPL-1076/ 10787 |
Compression in networking is not configurable other than on/off. Data compression typically involves a trade-off between CPU usage and the amount of compression achievable. The utility of the compression feature would be increased by allowing more flexibility in terms of this trade-off. Solution: The compression "level" of the existing zlib compressor may now be configured. Other compression algorithms may also be used. |
OSPL-1486 |
Java API copy routines were non-optimal for arrays and sequences The Java copy routines used intermediate copies which were not always necessary. Solution: Copying of arrays and sequences has been changed to do direct copies where possible. In case direct copies were not possible due to memory layout differences, the specialized copy-routines have been optimized as well, providing a considerable performance gain on copy-performance. |
Report ID. | Description |
---|---|
OSPL-1509/ 11063 |
A service generates a trail of defunct processes when the restart FailureAction
is configured. In the situation when a service crashes on a posix system and this service is configurated with the FailureAction set to restart then a trail of defunct processes will be generated of this service. Solution: The fault in the restart mechanism is now fixed and will not occur anymore. |
Report ID. | Description |
---|---|
OSPL-1116/ 10796 |
When compression is enabled in networking the reliable channel may not operate correctly. When compression is enabled then the resend messages are incorrect which causes the reliable communication to fail. Solution: When sending a compressed message the compressed message and the correct message size is now saved on the resend list. |
Report ID. | Description |
---|---|
OSPL-631 |
Using read or take with max_samples limit set can cause some key-values to be never read. The read and take operations return instances starting with lower key-values. If not all available data is read at once (e.g., when having set the max_samples limit) and the lower key-values keep receiving updates, a subsequently performed read operation will return the updated instances, which may prevent the higher key-values to be read. Solution: The read and take operations are changed to provide data circularly from a cursor. This means that these operations will 'resume' a read as if read_next_instance was succesively called. This way all instances can be read even when lower key-values get updated between two read-operations with a max_samples limit. |
Report ID. | Description |
---|---|
OSPL-1148 |
The end_coherent_changes() operation may crash if the publisher contains a
writer that did not write any samples in that coherent update There was an issue where calling end_coherent_changes() on a publisher may cause a crash if that publisher contained a data writer that had not written any samples as part of that coherent update. The algorithm assumed that every writer belonging to a coherent publisher would write samples within each coherent update. Solution: The algorithm in end_coherent_changes() has been fixed to also support data writers that were not active during that coherent update |
Report ID. | Description |
---|---|
OSPL-1049 /107678 |
Java application crash on type register Randomly, on type register, a java application can crash with the following notice: malloc(): memory corruption. Solution: The type register memory allocation algorithm for java has been fixed and will no longer crash. |
OSPL-1073 /10785 |
Persistency did not function correctly with PartitionTopic elements
in namespace configuration When a PartitionTopic element was specified for a namespace, the durability persistent store ignored the topic-part of the expression. This caused all data to be stored persistently for a partition rather than only the specified partition-topic combinations. Solution: An additional filter is added which takes into account topics to make sure that only data matched by the expressions as specified in the durability namespace configuration will be persisted. |
OSPL-1110 /10793 |
Coherent Updates performance issues when re-using instances When a sample was written to a pre-existing instance of a coherent update, it was leading to a decrease in performance on the data reader side and eventually possible deadlock, depending on the sample rate. Solution: The algorithm when an end coherent changes marker is received at the data reader side has been improved to clear the trigger status meaning that inserting a later sample into the instance will not necessarily cause a re-evaluation of the query condition. This means that samples can be written to the data reader's queue at a consistent rate. |
Report ID. | Description |
---|---|
OSPL-1013/ 10750 |
Coherent access with TOPIC presentation not working properly Applications are not notified of new data when the complete set of samples that belong to a transaction have been received when using a listener or waitset mechanism. Also, transactions were not properly stored in historical data and could therefore not be provided to late-joining readers. Solution: The triggering mechanism for waitsets and listeners has been adapted to demonstrate correct behaviour concerning notifications of end of transactions. Furthermore, the reader plus durability transient and persistent store algorithms have been modified to store transactions properly. |
OSPL-1017/ 10755 |
Durability crashes when alignment data contains only unregister messages In the scenario where there are only unregister messages aligned by the durability service for a given partition-topic combination, the service would crash because an instance lookup would fail. Solution: Durability is made robust against this kind of scenario. |
Report ID. | Description |
---|---|
OSPL-371/ 10193 |
The durability 'PartitionTopic' configuration was not correctly communicated. Because of a bug in the way partition-topic combinations were communicated between durability services, communication was sometimes not established because durability namespaces did not seem to match. Solution: The durability service now correctly communicates the partition-topics for a namespace. |
Report ID. | Description |
---|---|
OSPL-371/ 10193 |
Durability needs to support topic granularity in namespace configuration. In previous versions a user was able to express which partitions belonged to a namespace. It appears that there are usecases where a user needs topic-level control here. Solution: Durability now allows specifying partition-topic combinations which allow for a finer degree of control. Look for the 'PartitionTopic' element in the Durability/NameSpaces/NameSpace configuration. |
OSPL-706/ 10272 |
On a failing boundscheck the corresponding member is not reported. When during a write a member is not in conformance with the bounds specified by the type of that member, the middleware reported an error. However, the corresponding member was not logged, which makes debugging of these kinds of errors hard. Solution: The member for which the boundscheck fails is now logged. |
OSPL-892/ 10091 |
DDSI2 can respond with incorrect set fragments to retransmission requests The DDSI protocol allows reliable readers to request retransmission of individual fragments of large samples. DDSI2 could respond with fragments other than the ones requested, which could generally be masked by an eventual request for the full sample to be retransmitted, but would cause the communication to come to a complete standstill if a particular fragment always got lost. For example, when the rapid (re)transmission of a large amount of data systematically lead to a switch throwing away a whole range of packets. Solution: DDSI2 now always responds with the requested fragments. |
Report ID. | Description |
---|---|
OSPL-646 |
Implement Java API performance improvements to reduce JNI boundary crosses The copy functions that are used to copy data from the shared memory representation into their Java language representation and vice-versa were not optimal under all conditions. Especially the combination of enums/unions and an external IDL compiler (in CORBA cohabitation mode for Java) resulted in additional JNI transitions due to the fact that the copy functions had no a-priory knowledge about their actual implementation. In these cases the copy functions had to revert to invoking the IDL mandated manipulation operations instead directly setting the underlying attributes, and each time a Java operation needed to be invoked added the overhead of one additional JNI transition. Solution: The copy functions have now been adapted to use, besides the a-priori knowledge about the idlpp output in StandAlone mode, also a-priori knowledge about the exact layout for enums/unions generated by the JacORB V2.3.x compiler. Therefore there is not need to make additional JNI transitions when manipulating types that match one of aforementioned layout, thereby reducing the time required to read and write these types. CopyCache administration is now completely separated for the a-priory supported IDL layouts (idlpp native output and JacORB V2.3 output) and the non-recognized IDL layouts. In the first case it will store jfieldIDs, in the latter jmethodIDs. Both layouts now have their own copy functions. |
Report ID. | Description |
---|---|
9963/ dds3426 |
DDSI2 used incorrect encoding for fragment data message headers DDSI2 incorrectly used to generate and interpret fragmented data message headers as if they were slightly extended versions of the non-fragmented data message headers. This caused DDSI2 to be non-compliant with respect to the standard and to fail to interoperate with other vendors' implementations for large samples. Solution: the setting and interpretation has been corrected. This breaks backwards compatibility, but because DDSI2 is still in beta, this does not constitute a change of policy. For those exceptional cases where backwards compatibility is currently an issue, a setting Unsupported/LegacyFragmentation has been introduced, which may be set to true to continue using and interpreting the old message format. |
Report ID. | Description |
---|---|
dds3525 |
Performance enhancement: Number of select-calls performed by RT-networking reduced
Solution: The number of select calls performed by RT-networking has been reduced when data is received, increasing the performance in many cases. Specifically platforms on which selects are expensive and/or deployments where multiple networking services share the same NIC (e.g. Solaris Zones) benefit from this enhancement. |
Report ID. | Description |
---|---|
10299 / dds3492, dds3500 |
The durability MMF persistency implementation leaks shared memory The durability service MMF persistency implementation leaked shared memory in both the MMF store itself as well as in the 'regular' shared memory on the node. Furthermore it synchronised the MMF store to disk during the alignment phase unnecessarily. Solution: The memory leaks in the MMF implementation have been resolved and during the alignment phase the durability service does not synchronise the data to disk any more as it will switch back to the back-up in a consecutive run anyway when alignment does not complete successfully. |
dds3501 |
Thread creation hangs when trying to use real-time scheduling without
appropriate permissions When OpenSplice is started with a configuration that specifies real-time scheduling for one or more threads by a user that does not have operating system permissions to start real-time threads,thread creation fails and OpenSplice hangs after reporting the error. Solution: OpenSplice now reverts to time-sharing scheduling and continues after reporting a warning to indicate missing permissions. |
Report ID. | Description |
---|---|
10239 / dds3261 |
Communication error with VxWorks 6.x Byte order was reversed, messages were not processed hence no communication. Solution: Correct _BYTE_ORDER check used. |
10180 / dds3418 |
Optimisation for empty Strings in Java Solution: The functions used to translate incoming samples into the Java language binding have been optimized to handle empty strings. These functions used to allocate a separate Java String object for each empty string they encountered, but now they will share a single and global reference to a default empty String object that has been pre-allocated during the start of the Java application. Reusing a single empty string object to represent all empty string will reduce the overall Java memory footprint and reduce the number of Java object allocations, thus reducing execution times as well. This optimization will especially benefit applications that communicate large amounts of empty strings. |
9963 / dds3478 |
DDSI2 fragment size is now configurable DDSI2 never creates data messages containing a payload larger than the FragmentSize, any sample larger than the FragmentSize gets split into multiple fragments of FragmentSize each. These fragments are then transported independently (but may yet be merged into larger UDP datagrams). Solution: This size is now configurable using Unsupported/FragmentSize, with a default of 1280 bytes. Values below 1025 bytes violate the DDSI2 specification, above approximately 65000 bytes it (probably) won't fit inside a single UDP datagram. Increasing the size will shift more fragmenting and reassembling to the IP stack, which is generally more efficient because it is done inside the network stack, but which is incapable of retransmitting individual lost fragments. Increasing it may also allow operating without any fragmenting at the DDSI level, which may help avoid interoperability issues. |
10271 / dds3474 |
Storage of partition-topic definitions to MMF store takes a long time during alignment The durability service synchronised the MMF store to disk after storing each and every partition-topic definition. Even though this is not incorrect, the synchronisation may take a long time, especially in a system where a lot of other software is running in parallel. In such a situation synchronisation may take up to several minutes. Solution: The durability service now no longer synchronises the partition-topic definitions to disk during initial alignment, because if not all of them are present, the store cannot be used as a source in a consecutive run anyway. |
10270 / dds3476 |
SA non root user cannot use Real-Time scheduling When configuring Real-Time scheduling class for a process or thread, they were only applied if the process/thread was started by the root user even though other users may have the ability to use the Real-Time scheduling classes as well. Solution: The restriction of needing a root user for Real-Time scheduling has been removed. |
Report ID. | Description |
---|---|
dds3217/9881 dds3430 |
Spliced shared memory leak on remote durability shutdown A memory leak was found whenever a remote node was started and then stopped while durability was also running. Solution: When durability detected a remote node, it would write an update into the system (and thus the shared memory on the local node). The logic for this write wrongly performed a double string creation for the same string. The second string creation overwrote the pointer to the first created string, which resulted in the first created string to be never freed when the remote node disconnected. The solution was to remove the first string creation, as it was superfluous. |
dds3399/ 10139 |
Incorrect network statistics. The networking service optionally keeps track of statistics that can be inspected by means of the Tuner tool. After inspection it turned out that the maxNumberOfPacketsResentToOneNode and maxNumberOfBytesResentToOneNode statistics are showing wrong values. Solution: The maxNumberOfPacketsResentToOneNode and maxNumberOfBytesResentToOneNode statistics are now showing the correct values. |
dds3420/ 10187 |
Applications SEGFAULT when the domain is not started. In the case no DDS domain is running but there is still a dirty shared memory segement present from an old instance of Spliced. It can happen that when an application is started, the application can crash with a SEGFAULT. Solution: The defect is fixed and the application will no longer crash because of this. |
dds3455 |
The durability service should improve the memory used during alignment. The durability service temporarily caches received alignment data until the set for a specific partition-topic combination is complete. The algorithm implemented there could be improved to reduce the amount of memory used during this phase. Solution: The durability service now stores unregistrations is a much more memory-efficient way reducing the memory overhead for alignment to a minimum. |
dds3457/10241 dds3458/10242 |
Crash of networking and/or durability during due to memory exhaustion. The networking and durability services could crash when the shared memory was exhausted. Solution: The services now check the available memory threshold and does not claim more memory when the threshold has been reached. Furthermore they will terminate when no more memory becomes available within a few seconds. |
Report ID. | Description |
---|---|
dds3248/ 9908 |
Partition expressions not matched properly. In case an application publishes/subscribes to multiple partitions and the list of partitions has names that are sub-strings of one of the others in the list, some partitions may be ignored. Even though communication within a node and communication over the network using the native networking service still works correctly, this issue causes communication over the network to fail when using the ddsi networking service and it also causes listeners not to be triggered on matched subscriptions and publications. Solution: The error in the pattern matching algorithm in the kernel has been repaired. |
dds3251/ 9912 |
Alignment of non-volatile data temporary requires an additional shared memory size of all unregistration messages that will be aligned. During the alignment of a set of data, the durability service temporarily allocated a complete sample for each unregistration message that needs to be aligned in order to keep track of the instance. In cases where a lot of instances exists and samples are large, this had a large impact on the amount of shared memory that needed to be available. Solution: The durability service now allocates one message for an unregistration at the time and frees it before allocating the next one. This ensures only the size of one extra sample needs to be available in shared memory, independent of the number of unregistration messages to align. |
dds3431/ 10188 |
ospl -f always waits worst-case terminate period. When 'ospl -f' receives a TERM signal, it will stop spliced and wait a configured amount of time (4s + Lease/ExpiryTime + ServiceTerminatePeriod) for that to finish. If spliced doesn't normally finish during that period, ospl will forcedly stop the processes and cleanup the resources. The issue is caused by a deadlock in the 'ospl -f'. When ospl decides to shut-down the spliced due to the reception of a signal, the (single-threaded) tool is waiting for the 'system' call to return (reading the exit-status from the process) while in the meantime checking the existence of the process by means of 'kill(pid, 0) != -1'. That check does however not yield the desirec outcome since kill will only return -1 if pid does not exist (and note that an existing process might be a zombie, a process which already committed termination, but has not yet been wait()ed for). Solution: The solution is to call waitpid in the signalhandler of ospl, before kill(pid,0) is used to check if the spliced process is gone. Additionally, the signalhandler sets the exitstatus returned by waitpid, because this status is no longer correctly returned by the system function(the process has already been wait()ed for). |
Report ID. | Description |
---|---|
dds3240 |
RTNetworking did not correctly check eventmask for NEW_GROUP events Instead of checking if the V_EVENT_NEW_GROUP was set in the eventmask of an observer, networking checked if the mask was equal to V_EVENT_NEW_GROUP, and this test failed when multiple events had occurred before networking (or any other service) had the chance to react. Solution: The check is modified, so networking only checks for the related event-bit. |
Report ID. | Description |
---|---|
dds3393/ 10133 |
StoreSessionTime not respected in all cases The StoreSessionTime configuration item was ignored for SMP stores, which could cause data to be synchronised very often. Some use-cases require less synchronisations to be performed in order to improve throughput. Solution: In order to allow for delayed flushing, the StoreSessionTime configuration is now also respected by the MMF persistence implementation. |
Report ID. | Description |
---|---|
dds3155/ 9755 |
Delayed initial durability alignment Previously, when a system had a late-joining node with persistent data, this persistent data was not injected. This behavior was considered valid, since it can be disruptive to inject historical data in a running system. However, in a system where no data has been written, it can be desirable to do a delayed injection of persistent data from a late-joining node. Solution: A new option has been implemented where, in the case where no writer has been created in any of the partitions in a durability namespace, persistent data from a late-joining node can still be aligned. This functionality has to be explicitly configured on a namespace policy, using the delayedAlignment option. See the configurator tool (osplconf) for more details. |
dds3302 |
New tool mmfd available A new tool has been added called 'mmfd' that can be used to dump the contents of an 'MMF' type persistent file. You must pass the tool the name of the persistent file and its mapping address, and then it allows you to browse through the contents of the mmf file and to monitor its memory statistics. |
dds3409 |
mmstat tool extended mmstat has been extended with an option to order its object references (in case of the 't' or '-T' mode) by objectCount/objectSize/TotalSize. This helps you to focus on those object types that have the most impact on the consumption of shared memory, by listing the biggest objectCount/objectSize/TotalSize on the top. You can also limit the the amount of items being displayed to only the top 'n' items, where 'n' is a command-line parameter. For this purpose 2 new command-line parameters have been added: -o <C/S/T> : displays an ordered list sorted by objectCount/objectSize/TotalSize. -n <n> : displays only the top 'n' items in this list. |
dds3410 |
Performance and memory enhancement for data models with empty strings Currently the OpenSplice database allocates separate empty strings as separate chunks of memory, each having their own '\0' terminator and headers. In applicationss with huge amounts of empty strings, a lot of memory could be saved if the database allocates 1 empty string by default, and then always points to the location of this default string (increasing its reference count accordingly) when an empty string needs to be expressed. Solution: Empty strings in the database now always refer to the same chunk of memory (containing the '\0' terminator), instead of allocating each empty string separately. In case of data models that use huge numbers of empty strings, this can save a lot of performance and memory. |
Report ID. | Description |
---|---|
dds3376/10114 |
A dispose_all_data() applied consecutively on several topics does
not remove all data A regular DataWriter and DataReader are used to transport the dispose_all_data command. However, both DataWriter and DataReader have their HistoryQosPolicy set to KEEP_LAST with a depth of 1. Since each splice deamon has its own private instance, consecutive dispose_all_data commands that originate from a single source will overwrite previous commands in its DataWriter (in case a resend is required) or in one of more splice daemon DataReaders (in case they haven't processed the command before the next command arrives). This may lead to missed dispose_all_data commands resulting in not disposing the data nor notifying the applications. Solution: Both internal DataWriter and DataReader now use a KEEP_ALL history policy to ensure all commands are sent and received. |
dds3385 |
Durability tracing level NONE missing in configurator The configurator tool did not allow configuring a value of NONE for the durability service tracing verbosity that is supported by the service. Solution: The NONE verbosity level has been added to the configurator tool. |
Report ID. | Description |
---|---|
dds3200/9845 |
The 'ospl list' command does not list all running domains In case an OSPL_URI environment variable was set, the 'ospl list' command only showed the domain associated with that URI (in case that was running). Solution: The 'ospl list' command no longer takes the OSPL_URI environment variable into account and simply lists all running domains. |
dds3211/9859 |
Networking crash when receiving a TERM signal Terminating the real time networking service by sending it a TERM signal is not supported, but doing it sometimes resulted in a segmentation-fault. Solution: The termination request is now fully ignored and the networking service no longer crashes. |
dds3274 |
Umask was not correctly applied on SHM and keyfile The umask was interpreted wrongly, potentially causing unexpected permissions on the shared-memory segment and keyfile. Solution: The umask is interpreted as expected now. In case only read or write permission are specified for an access category, that category will be granted both read and write permissions on the domain. |
dds3315/10030 |
Durability service fails to inject persistent topic definitions in case of dependency conflicts In some scenarios the mmf store failed to inject its persistent topic definitions when started from a clean Domain. In the error log it would write messages stating that the Deserialize failed, and later on that it failed to register the topic type.This issue was caused by the fact that the persistent store recorded the wrong entry-point for its persistent datatypes: instead of pointing directly to the structured datatype definition it was pointing to the outer-most module of this datatype. In case other datatypes were specified in the same module, the deserializer would try to deserialize all datatypes from that module at the same time. Solution: Those issues have now been fixed by making sure that the a clone on a type definition always returns the same entry point as the original, which in this case is the structured datatype itself. This way the mmf store should be able to inject its persistent topic definitions without any further problems, even if their datatypes are defined in the same module. |
dds3316/10031 |
Persistent store sometimes contains duplicate instances Instances that have a string as their key (either an IDL string or an IDL char array that was transformed by a #pragma cats) were duplicated in the mmf type persistency store when updates are applied to them. Solution: A fix has been applied to mmf store to avoid duplication. |
Report ID. | Description |
---|---|
dds3156/9753 |
Reliablility problem in networking when running multiple nodes. When running real time networking and nodes are communicating over a reliable channel, it is possible that when a node is shutdown other nodes in the network start to lose communication with each other due to a corruption in the network administration. Solution: The defect in the network administation algorithm is now fixed. |
dds3290 |
Native networking retransmits to the wrong IP address Native networking should retransmit to the IP address of the destination machine, but uses the global partition's destination address instead. The result is that all machines receive the retransmissions. When combined with network partitions, machines that never received the original data may get the retransmissions destined for another node, and the network floods. Solution: The defect in native networking is now fixed. |
Report ID. | Description |
---|---|
dds3236 |
Durability MMF store destroys database on exit and snapshot The database of an MMF store was destroyed in case durability detached the store from memory, effectively corrupting the data. This behavior is related to the introduction of the new memory manager, which performs processing that has to be explicitly stopped before database detachment. Stopping this processing implied destroying the database, which caused the described issue. Solution: A suspend\resume mechanism for the memory manager now allows durability to explicitly start\stop memory manager processing without destroying the database." |
Report ID. | Description |
---|---|
9127/ dds2821 |
OpenSplice DDS should not start if ReportPlugin configuration is invalid or fails to initialize If a ReportPlugin configuration is specified and that ReportPlugin contains invalid data or fails to intialize OpenSplice DDS still starts. If the ReportPlugin contains invalid data or fails to initialize OpenSplice DDS should fail to start. Solution: The solution was to report any invalid data found in a ReportPlugin or any failure to initialize a ReportPlugin to the default error log and for OpenSplice DDS to fail to start in these circumstances. |
Report ID. | Description |
---|---|
9682/ dds3137 |
networking service crash with synchronous write acknowledgments The native networking service may crash during synchronous write operations. This was because the management of a table of WaitList objects was found not to be thread safe. Solution: The solution was to add appropriate mutex locking within the synchronous write administration in order to make the mechanism thread safe. |
9298/ dds2913 |
Intermittent crashing of OpenSplice Tuner The tuner crashed when an invalid entity was passed from the control and monitoring API, while trying to report that the entity could not be claimed. Solution: A check is added to the function reporting the error so it doesn't crash on an invalid entity. An error-report is added to the control and monitoring API which reports the xml-string when it contains an invalid entity. |
dds2998 |
OpenSplice can fail to start as a Windows service but report that is has started successfully. If the OpenSplice daemon failed to start under certain circumstances when installed to run as a windows service, the service control manager would not detect this and would report that the service was running. The circumstances included if the machine had not been rebooted after installation and if the system path or other environment variable values were incorrectly set. Solution: If the OpenSplice daemon fails to start when installed as a windows service in these circumstances this is detected and an error message is logged to the Windows EventLog. The windows service will exit and indicate an error to the Service Control Manager. This allows the user to configure recovery actions for failure of spliced. See http://technet.microsoft.com/en-us/library/cc738230%28WS.10%29.aspx for details of this. |
dds3078 |
OpenSplice emits spurious error message regarding a failure accessing temporary directories on Windows. OpenSplice emitted spurious error messages regarding a failure accessing temporary directories on Windows. The messages were typically of the form "Description : Unable to create directory 'C:' within path ... Errorcode: 5" Solution: OpenSplice no longer emits spurious error message regarding a failure accessing temporary directories on Windows. |
3089 |
Networking service crashed on startup when no network interfaces found. If the OpenSplice native networking service was started before any network interfaces were available from the operating system, it would lead to undefined behaviour. This has been corrected and the behaviour now is that the networking service will write a message to the ospl-error.log indicating that no interfaces could be found and networking will then terminate. Solution: The networking service will now exit gracefully and not crash if it cannot find any network interfaces it can bind to. It is a requirement that OpenSplice can only be started once the operating system is fully active and a network interface is available to allow the networking service to run. |
Report ID. | Description |
---|---|
7255/dds2239 7653/dds2405 |
Memory thresholds not always obeyed by services When memory thresholds were defined, some services would not check the threshold status before allocating shared resources. Solution: The routines of services that allocate shared memory now obey the rules regarding memory-thresholds. |
8957/dds2697 |
Unexpected signal handling on SIGABORT or SIGSEGV The implemented signal handling for POSIX platforms makes it impossible to generate application core files because the clean-up in ospl relies on exit() being called. Solution: A new signal handling mechanism has been implemented for POSIX-compliant platforms. The reference manuals have not been updated properly to reflect the changes, which is marked in the DCPS Known Issues section as dds3080 that also describes the new behaviour. |
9377/dds2993 |
Problem using dds in namespace The code that is generated by the OpenSplice IDL pre-processor for C++ does not use the correct scope for the DDS module (DDS::<X> instead of ::DDS::<X>). In case a DDS sub-module is used for application-specific types, the types are incorrectly resolved. Solution: The IDL pre-processor has been adapted to generate the absolute scope for internal DDS types. |
9388/dds3027 |
Incorrect dereferencing in idlpp-generated code for bounded strings The copyIn routine for (array of) bounded strings, generated by idlpp, contained a bug related to incorrect dereferencing. The code in question is only used if it is compiled with the OSPL_BOUNDS_CHECK flag. Solution: Corrected the idlpp code generation. |
9470/dds3039 |
64bit support for MMF store The 4GB limitation on MMF store size is removed. On 64bit platforms it will now be possible to specify sizes above 4GB. Solution: Adjusted the configuration and operating system abstraction to allow for passing of 64-bit values. |
9476/dds3047 |
Wireshark plugin is out of date and does not compile The plug-in referenced old documentation that made it difficult to compile manually. Also, the latest networking protocol changes were not applied. Solution: Updated the documentation and protocol decoding for OpenSplice Networking. |
dds3049 |
C# API is crashing when reading samples from the DCPSPublication and DCPSSubscription topics Caused by incorrect idlpp output where the gapi representation of the struct containing the builtin topics is not referring to the gapi representations of its embedded QosPolicies, but to the C# representation of these QosPolicies. In most cases these representations have the same footprint, but in some cases (PresentationQosPolicy) it does not. That may cause wrong pointer evaluations in the CopyOut function. Solution: idlpp output corrected. |
9459/dds3060 |
The 'c_bind' and 'c_select' symbols conflict between Eiffel and OpenSplice One of the ospl libraries exports the 'c_bind' and 'c_select' symbols. These symbols conflict with symbols used in Eiffel Solution: The 'c_bind' and 'c_select' symbols have been renamed to respectively 'ospl_c_bind' and 'ospl_c_select' |
dds3072 |
Networking crash when using "reconnection_allowed" in multinode In some circumstances the networking service can crash when the "reconnection_allowed" option is used due to the fact that the administration for a disconnected remote node is not cleaned up properly. Solution: The administration of the remote node is properly cleaned up. |
Report ID. | Description |
---|---|
dds3006 |
Errors are logged for destruction of mutexes during shutdown on AIX Errors were logged because of destruction of unused mutexes structs which had not been initialised. Solution: Cleanup of mutex destruction/removal of unused mutex. |
dds3021 |
mmstat cannot be piped to a log file. The mmstat tool exits immediately when stdin is not a tty. Solution: Updated mmstat to continue if stdout or stdin is not a tty. |
Report ID. | Description |
---|---|
dds2723 / |
Creation of DataReader with a Topic resulting from a find_topic call Within OpenSpliceDDS a problem existed surrounding the creating of a DataReader with a Topic that was the result of a 'find_topic' call. The typeSupport that is needed by a DataReader must be registered for the same type name as used to create the DataReader's Topic. A Topic resolved by 'find_topic' call is implicitly created by using the default type name (the IDL type name). So if the DomainParticipant has registered the TypeSupport using an alias name that is different than the default name then OpenSpliceDDS will not be able to find the TypeSupport by its default name. Solution: The solution is that whenever a TypeSupport is registered it is now also always registered by its default name. |
dds2787-1 / 9112 |
Exit-hook in JVM could be executed during a callback causing a deadlock
on shutdown of a Java application The exit-hook ensuring proper clean-up of entities created in Java could be called and executed during a listener-callback. This would cause a deadlock. The tutorial-code (e.g. the MessageBoard) could trigger this issue, but it could also occur when the application was terminated by issuing CTRL-C. Solution: The exit-hook has been wrapped in a separate thread, ensuring parallel execution of the clean-up code preventing the deadlock. |
dds2787-2 / 9112 |
Freeing of process-space (user-layer) entities could cause spurious logs,
hangs or crashes During clean-up of entities in process-space (user-layer), invalid memory may be accessed. This could result in logs about for example os_mutexUnlock-fails, but could also result in a crash or hang of the application, preventing proper shutdown. Solution: The offending code has been fixed to properly adhere to the locking and freeing sequence of user-layer entities. |
dds2810 / 9112 |
DDS_RETCODE_TIMEOUT return-code of writes ambiguous when resource-limits
are set on a synchronous writer When resource-limits were set on a synchronous writer and DDS_RETCODE_TIMEOUT was returned, it was impossible to detect whether the timeout occurred due to the resource-limits or due to synchronisation with subscribed synchronous readers. Solution: On a synchronous writer a call which in the normal case would block on resource-limits now immediately returns DDS_RETCODE_OUT_OF_RESOURCES. A synchronous writer now only blocks (for maximally the time specified in the reliability.max_blocking_time) when data is not yet delivered to all subscribed synchronous readers. |
dds2883 / 9009 |
On Windows, OpenSplice failed to run as a windows service if a 'Log on as'
value of a non-elevated privilege user account is specified. Global named objects could only be accessed from sessions that were interactively logged-in, or using accounts that corresponded to SID type Local System or Domain Administrator. Solution: The access control list has been relaxed & OpenSplice runs as a windows service if a 'Log on as' value of a non-elevated privilege user account is specified |
dds2968 / 9343 |
Variable assignment in release.com is not Korn shell compatible Checking an environment variable within the release.com where it has not been set fails when using ksh. Solution: Change variable usage in release.com to ${VAR:=} which guarantees nothing is returned if the variable has not been set. |
dds2975 |
Reading/Taking invalid samples Reading or Taking invalid samples could in certain scenario's result in a crash, or in samples that seemed to have a corrupted state Solution: Invalid samples will always be guaranteed to have legitimate key-values. The non-keyfields of an invalid sample (for which the valid_data field in the corresponding SampleInfo is marked FALSE) are NULLIFIED and should not be interpreted by an application, since this is the only situation in which null pointer strings can be encountered. |
dds2980-1 |
Default ExpiryTime and ExpiryTime@update_factor for Lease element of the Splicedaemon modified The defaults for the Lease element of the Splicedaemon configuration were incorrect. Solution: The defaults have been modified. The default ExpiryTime is now 10s, the default ExpiryTime@update_factor is 0.2, resulting in heartbeats being sent out every 2s. For more information or modifying the configuration files the configurator tool or deployment manual should be consulted. |
dds2980-2 |
The default value for the standards conformance attribute of DDSI2 has been changed The default for DDSI2 for its standards conformance was 'strict', which would cause high CPU and network-load when communicating with RTI, because RTI doesn't adhere to the specification. Solution: The default of DDSI2's standards conformance has been relaxed. The new default is 'lax', providing the smoothest possible interoperability by default with vendors not adhering to the specification. |
dds2988 |
Deadlock involving the waitset and status conditions When a waitset is triggered by a statuscondition, then it will first lock the status condition, then the entity within the status condition. However if the entity is also deleted at the same time the locking strategy is reversed. The entity is locked first and then the condition is locked. This creates a deadlock scenario. Solution: The deadlock scenario has been removed by no longer locking the entity once the condition is locked. This is safe as the entity remains in consistent as long as the condition is not deleted and the condition can not be deleted without claiming the lock of the condition. |
dds2991 |
Define log output directory on installation when OpenSplcie is installed as a windows service When OpenSplice is installed as a windows service, the environment is configured such that OpenSplice will write log files into a directory within the current users profile. On some Windows installations the default service user account LocalSystem does not have a profile directory and logs cannot be created. Solution: The OpenSplice installer prompts for the user to supply a system wide directory path for all OpenSplice logs when OpenSplice is installed as a windows service. |
Report ID. | Description |
---|---|
dds2984 |
Building examples on AIX fails The -qobjmodel=ibm flag had been removed from the build but not example makefiles Solution: The -qobjmodel=ibm has been removed from the example makefiles Java based tools require Sun JVM which is not available on AIX The osplconf and ospltun tools will not start without a Sun JVM. Solution: The JVM vendor check has been removed |
Report ID. | Description |
---|---|
dds2940 / 9322 |
Connection problem on solaris 10 With the implementation of IPv6 support a bug was introduced that caused connection problems on Solaris 10 platforms, when using the networking service. This resulted in inability to communicate with other nodes and the following ERROR was reported: "Description : sendto returned errno 22 (Invalid argument)". The root-cause was a wrong sockaddr size that was supplied to the sendto() call. Solution: The issue is now fixed by supplying the correct size, based on the used addressing-method ( IPv4 or IPv6 ). |
Report ID. | Description |
---|---|
dds2267 / 7285 |
ospl stop does not clean properly On 'ospl stop', OpenSplice services do not terminate when the splice-daemon has already terminated, causing the key file as well as the services to remain active until all services have detected the termination of the splice-daemon. This detection time is determined by the //OpenSplice/Domain/Lease/ExpiryTime setting. Solution: On start-up the splice-daemon now runs in its own process group and this group includes all services it starts. On ospl stop, this group is killed in case the splice-daemon has already terminated and the other services have not detected it yet. |
dds2449 / 7946 |
Durability and networking still running after spliced death Other services keep running If the splice-deamon is killed. Solution: All services are now actively monitoring whether the splice-daemon updates its lease and terminate when they detect this is no longer happening. This detection time is determined by the //OpenSplice/Domain/Lease/ExpiryTime setting. To ensure immediate termination of the services, an 'ospl stop' can be issued by the user. This will actively kill the remaining services and clean up the shared memory. |
dds2497 / dds2868 9169 |
Primitive type array may get corrupted when fragmented on the network. When a data sample containing an array of primitive types is transported by the networking service between nodes, corruption of the content can occur when the sample is fragmented over multiple network packets by the networking service. Solution: The bug in the defragmentation/deserialisation algorithm that caused this behaviour has been fixed. |
dds2732 |
On Windows, OpenSplice shuts down when any interactive user session logs off. OpenSplice is shutting down when it received the Windows control event CTRL_LOGOFF_EVENT. Solution: OpenSplice now ignores the Windows control event CTRL_LOGOFF_EVENT and does not shutdown. |
dds2761/ 9058 |
Copy of idl-union sets the m__d_set flag making it different from the original The idlpp tool generates incorrect code for making a copy of an idl-union, which results in a copy that is different from the original. Solution: The idlpp tool has been modified to generate the correct code for copying unions. |
dds2818 / 9123 |
DataReader no more notified after a dispose_all_data Disposes of empty instances end up invisible to the application due to the fact the DataReader instance administration is not updated properly for empty instances. Furthermore, the Tuner tool is not able to cope with dispose messages that were triggered by dispose_all calls due to a missing check in the Tuner code. Solution: The DataReader administration is updated properly for empty instances as well and the Tuner now is able to handle all types of dispose messages. |
dds2823 |
write_w_timestamp uses wrong base time for computing blocking duration write_w_timestamp determines when to return with a timeout by adding max_blocking_time to the time-stamp supplied by the application, instead of adding it to the current time. The application-supplied time stamp may be long in the past, causing spurious timeouts. Solution: The blocking time is now based on the current time instead of the supplied time. |
dds2841 |
DDSI configuration for compatibility mode can't be set. The configurator tool does not allow configuring the already supported compatibility mode for ddsi. This mode determines how strict ddsi spec compliance is enforced to allow communication with other less-compliant ddsi implementations. Solution: The configuration parameter has been added to the configurator. |
dds2911 |
Tuner crashes when creating a shared reader Creation of a shared DataReader is only possible in case the Subscriber is also a shared Subscriber. The Tuner never creates shared Subscribers in case a shared DataReader needs to be created. Creation of a shared DataReader for a non-shared Subscriber results in a crash. Solution: The Tuner now explicitly checks whether a shared DataReader is to be created. If so, if will make sure to create a shared Subscriber for it and if the user tries to create a shared DataReader for a non-shared Subscriber, it will report an error and refuse to create the DataReader. |
dds2923 / 9314 |
The same sample is received several times with DDSI2 With DDSI2 time stamps may be off by 1ns and a single sample may be received twice with different time stamps DDSI2 performs conversion between DDS time stamps (seconds + nanoseconds) and DDSI time stamps (64-bit fixed-point with the binary point to the left of bit 31, i.e., seconds + the fraction of a second as a 32 bit unsigned integer). The conversion routines contained a rounding error that caused the result of a round-trip through the conversion routines to be 1ns before the original. Because of this a data reader on the same node as the data writer would often observe a different source time stamp than a data reader on a remote node connected via DDSI2 did. This is part one of the issue. Part two of the issue, receiving a sample multiple times with different source time stamps was caused by an interaction with the durability service alignment procedure. The durability services interact among themselves via a special encoding of the data, which does not suffer from this rounding error. Hence, If a durability service was receiving data from its peer on the node where the data was originally published, while the writer was also sending data directly, some samples would arrive twice, with different time stamps. Usually the latency in data forwarding through the durability service is longer than that of the normal data transmission path, and this, combined with the rounding causing the timestamp to be 1ns early, caused the copy received via the durability services to be accepted as a valid new sample. Solution: The conversion routines have been fixed. The DDSI standard does not define the conversion routines in sufficient detail to guarantee that all implementations will interpret time stamps the same way. Similar issues with time stamps may therefore appear in a system environment in which multiple vendors' DDS implementations are interoperating. |
Report ID. | Description |
---|---|
dds2224 |
Usability and minor error corrections with the OpenSplice Tuner. Minor usability enhancements and edge case minor errors were identified with the OpenSplice Tuner. Solution: Implemented small fixes to improve usability and address edge case errors. |
8237 / dds2470 / CR_342 |
Daemon configuration scoped differs for different attributes Locking the daemon in memory on Linux must be specified outside the scope of < Domain > to be effective while the attributes for setting the scheduling priorities of the threads on VxWorks must be specified within the < Domain > to be effective. This should be consistent for all attributes Solution: The daemon is now locked within the < Domain > to be consistent will all other attributes. |
dds2491 |
Ref counted object compilation errors with TAO Ref counted objects on recent version of TAO will cause compilation errors Solution: Update ORB abstraction for DOC group TAO ccpp API. ::CORBA::LocalObject is refcounted from TAO 1.6.4/5 onwards |
dds2574 |
Durability fails to align starting node under heavy load When a system was experiencing heavy load, durability sometimes lost nodes because heartbeats were not received on time. Because of an optimization in the durability protocol, the status of groups partition topic combination) was not properly aligned when rediscovered. This caused problems with the alignment of other durability nodes, which never left the alignment state. Solution: After a node is rediscovered, durability sends requests for all groups that are not complete. This ensures that the durability node always has the last state. |
dds2593 |
Windows - CLASSPATH environment variable changed after compiling Java DLRL example After compiling the Java DLRL Tutorial the CLASSPATH environment variable is changed from the default after running the release.bat script. Solution: Corrected the Java DLRL Tutorial batch file and also found this issue when running the ospltun.bat script so fixed that aswell. |
dds2650 |
Issues with using the domain name instead of URI on windows Using the domain name, instead of URI, in a create_participant() call would fail on Windows if the proxy between application and OpenSplice kernel was not initialized yet. Also a non-default shared memory mapping address, supplied in the configuration file, was ignored. Solution: The proxy between application and OpenSplice daemon is now properly initialized in the case where the first create_participant call of an application passes the domain name instead of URI. The mapping address is now stored in the shared memory keyfile so it can be retrieved from the keyfile instead of config file URI. |
dds2656 |
Added missing networking parameters to the configurator The following parameters are not documented in the configurator but are in the deployment manual: - NetworkService/Discovery/Receiving/DeathDetectionCount - NetworkService/Discovery/Sending/Interval Solution: Added to configurator. |
dds2661 |
Durability service tracing log corruption Log messages used to be written to the tracing log in two steps: first the header containing a timestamp, state and source of the message, then the actual message. This caused incidental corruptions of the log file when headers and messages are written in incorrect order. Solution: Commit the complete message to the log file in a single action |
dds2693 |
Durability merge policy not applied if fellow node has not confirmed master The merge policy mechanism would check if a potential merge fellow had confirmed its master. An unconfirmed master was a reason to not merge, since it was assumed that merging with an inconsistent node would unnecessarily cause a network load while the same data would be available from a different fellow that did meet all merge conditions, or the node would cause a new merge trigger as soon as it did confirm its master. There are however situations in systems with multiple nodes and roles, where data can only be retrieved by merging with a fellow even if that did not confirm its master yet. Solution: The 'confirmed master' is no longer a condition for a merge action. Instead the fellow state is checked. A merge action with a fellow that has an unconfirmed master is only carried out if that fellow is past the initial alignment state. |
dds2696 |
Incorrect error reporting of durability error Randomly, at system start up, durability logs errors "Topic not (yet) known". Solution: After investigation, there was no error and therefore no need for this error message to be reported. It has been removed. |
9035 / dds2745 |
VC++ 2005 Compiler Error C2026 : line too big with idlpp idlpp can generate code that is too big for the cl.exe compiler. Solution: Generated code is now blocked into approx 100 characters to avoid exceeding the microsoft limit. The blocks vary depending on where the split occurs (between 2 xml tags) to ensure its user friendly to read. |
8986 / dds2749 |
Fix for Memory leak in C and C++ TypeSupport constructor In code generated by the IDL preprocessor various memory leaks were found, which caused strings to leak. In the generated code, strings were allocated and passed to the parent constructor of the TypeSupport constructor, but never freed. The other leak was in TypeSupport::register_type() where a string was allocated when parameter type_name was NULL, and never freed. Solution: In the generated code the strings are no longer allocated on heap, but are hard-coded and returned as 'const char *'. Another leak has been fixed where a string was not freed after a call on TypeSupport::register_type() with parameter type_name = NULL |
dds2753 |
Durability service gets stuck in infinite loop during master selection After a short period of one-way traffic between two hosts running the durability service, one of the hosts may fail to process a namespaces request sent by the durability service of the other host. Since two-way communication is restored shortly after, the reliability protocol cannot detect this. Thus the other host waited indefinitely on a reply to the namespaces request. Solution: The master selection algorithm now keeps a record of all nodes that haven't replied to a namespaces request in a timely manner. If, after a short while, these nodes are still alive, they apparently missed the namespace request and a new request is sent. Precautions are taken so that this behaviour will not cause network spikes. |
dds2786 |
Many Instance Handles causes performance degradation on Windows platforms The integrity of each InstanceHandle is protected by a mutex that is shared by all DDS processes. On Windows platforms, the construction and destruction of these shared mutexes leads to a significant performance degradation when a large number of instances are created and/or deleted. Solution: The mutex has been removed from the instance handle and replaced with a mechanism that uses a counter in combination with atomic increment and decrement operations. |
Report ID. | Description |
---|---|
dds3108 |
Problems starting spliced on VxWorks 5.5 with the mvme5100 target. Problems starting spliced on VxWorks 5.5 with the mvme5100 target due to an issue with how our logging determined thread name. Solution: Fix for the logging systems routine to determind thread id, for cases where we have no stored thread id. |
Report ID. | Description |
---|---|
dds3084 |
Problems loading libddscore.so on Vxworks5.5 ppc7448 due to relocation offset to large. Problems loading libddscore.so on Vxworks5.5 ppc7448 due to relocation offset to large. Solution: We have added the -mlongcall compiler option. |
Report ID. | Description |
---|---|
dds2747 |
SIGSEGV signal coming from Java Garbage Collector crashes the app. There are some bad interactions happening between Java's signal handling and signal handling set up by OpenSplice code. The JVM normally installs a signal handler for some signals, because there are situations where Java uses it internally for instance to allocate memory or wake-up a thread. OpenSplice blocks signals for a specific thread when accessing shared resources. Blocking of these signals prevents proper propagation to the JVM resulting in a crash. Solution: The blocking of signals is no longer necessary in OpenSplice due to a new approach for cleaning up shared resources with a dedicated thread and therefore threads are now no longer blocked. By pre-loading the jsig library that is delivered with each JVM, signal handlers of OpenSplice and the JVM are chained correctly and signals are handled by the correct signal handler. |
dds2755 |
Error in serializer w.r.t. module-less IDL-types. When serializing a type without any module, the type-name turns out to be an empty string causing the full type-name to be ":::: Solution: Types without a module are now serialized correctly by skipping the module without a name that is used internally for types without a module. |
9063 / dds2762 / SPCR1609 |
Finalise hook on the report plug-in not called during shutdown. The finalise hook of the report plug-in is not called when detaching from a domain. Solution: The missing calls to the finalise hook have been added to the implementation. |
Report ID. | Description |
---|---|
7738 / dds2440
8286 / dds2476 |
DDS keeps working after spliced daemon has stopped. When the splice daemon is stopped while applications are still running, the applications continue to function and use DDS. This is considered unsafe, as the splice daemon can no longer manage various DDS related tasks. When the splice daemon is gone, the applications should no longer be allowed to use DDS. Solution: When the splice daemon stops the applications will now automatically detect this and be prevented from using DDS. The application will receive the ERROR return code when using DDS functions. The detection time of the applications is based on the configured lease expiry time, in the default configuration this means that applications will detect the splice daemon being gone within 60 seconds. |
7886 / dds2445 |
Durability alignment not robust. When some of the first reliable packets that a node sends are lost on the network, the reliable communication between the nodes concerned may under some circumstances be blocked indefinitely in one or two directions for that channel. The visible result for the users is that the duribility services cannot allign, as they need reliable communication to work. Solution: The mechanism that handles joining of a new node in a reliable communication channel been extended, so that it can handle missing packets in any situation. |
8826 / dds2578 |
A QueryCondition that was not attached to a WaitSet failed to evaluate incoming samples. A QueryCondition uses a cache to prevent multiple evaluations of the same sample. When a sample is inserted into a Reader that has a corresponding QueryCondition attached to a WaitSet, then the sample must be evaluated to determine whether or not to trigger the WaitSet. This evaluation result is then cached so that when a user reads/takes from the query, it does not need to re-evaluate the full reader content. A flag determines whether the cache is considered incomplete (i.e. the reader still contains samples that need to be evaluated) or whether the contents of the cache reflect all matching samples. Due to an error in the query caching algorithm, the cache was only taking into account incoming samples when the query was attached to a WaitSet. This meant that the query cache could indicate that is was complete, while in fact the query still needed to evaluate some samples to substantiate that claim. Solution: The query cache is now marked incomplete by raising its flag when data is inserted while the query is not attached to a WaitSet. This causes the query to evaluate the reader content as well. |
8813 / dds2563 |
Method set_query_parameters on query conditions of dataviews did not function properly. Parameters in a query were not correctly resolved, which lead to failing calls on set_query_parameters on query conditions of dataviews. Solution: Internal prefixing has been added so that parameters are correctly resolved. |
dds2581 |
Java API lookup_topicdescription() function tries to find topic when lookup fails. On the Java API : When lookup_topicdescription() is called and the lookup fails, OpenSplice tries to find the topic as if find_topic() was performed. This conflicts with the DDS spec. If the lookup fails, the call should return NULL. Solution: Using Java API, When lookup_topicdescription() is called and the lookup fails, NULL is returned. |
dds2582 |
When calling find_topic, and the corresponding TypeSupport is not found, OpenSplice creates the TypeSupport. When calling find_topic on a DomainParticipant, and the corresponding TypeSupport is not found, OpenSplice creates the TypeSupport. This conflicts with what is specified in the Tutorial and the Reference manual, and also can cause problems if the user then registers the same TypeSupport explicitly. Solution: When calling find_topic, no TypeSupport is created/registered as specified in the manual. |
dds2595 |
Bounded Strings as character arrays See 5.4.2 release highlights for more info |
dds2629 |
Improved performance for MMF persistence See 5.4.2 release highlights for more info |
dds2633 |
Unable to configure scheduling properties of the main splice deamon thread. There are no configuration options available to adjust the scheduling properties of the main splice deamon thread. This thread is responsible for the renewal of the liveliness lease of the splice deamon. By the inability to configure the scheduling properties there can be problems when the lease renewal time is set to a very small period. The splice deamon thread will then potentially not be able to renew the lease in time and thus will be considered dead to the services, whom will then terminate. Solution: The scheduling properties of the splice deamon thread can now be configured by setting the parameters in Domain/Watchdog/Scheduling. |
dds2640 |
Deadline mechanism misbehaves when changing the deadline duration to a new value. In OpenSpliceDDS 5.4.1 a bug was introduced in the deadline mechanism. This bug occurred in the scenario where an existing deadline duration for a datareader or datawriter was updated to a new, for example shorter, duration. The new (shorter) deadline duration did not come into effect immediately, but only after the previous (longer) deadline had expired or if a read or write occurred before the deadline expired. Afterwards the new, shorter, duration would be in effect as normal. The same issue occurred for longer durations being set, i.e., the deadline would expire based on the previous, shorter, duration. Due to this bug OpenSpliceDDS would not properly indicate the number of deadlines missed in specific scenarios. It could be too low or too high. Solution: The bug in OpenSpliceDDS has been resolved and new durations being set are now immediately taken into account when determining the deadline missed status. |
dds2660 |
Java based floating license Hostname limitations when using java based floating licenses Solution: Removed java based licensing and returned to C based licensing. |
dds2661 |
Durability service tracing log corruption Log messages used to be written to the tracing log in two steps: first the header containing a timestamp, state and source of the message, then the actual message. This caused incidental corruptions of the log file when headers and messages are written in incorrect order. Solution: Commit the complete message to the log file in a single action. |
dds2666 |
size of LongDoublePair causing issues in idlpp When idlpp source generated for one platform is compiled for a different target instead then it could cause memory corruption. Solution: The idlpp generated code now determines structure sizes at compile time instead of code generation time. Issue affected C & C++ language bindings. |
dds2673 |
Cannot specify shared memory sizes of more than 4GB The implementation of friendly names ticket in 5.4.1 (dds2241) has caused any shared memory database sizes over 4GB to be set to size required - 4GB, i.e. 5GB would be set to 1GB. The size specified in a config file was read in to a 32 bit integer (an unsigned long). Secondly, two additional size parameters (Domain->Listeners->StackSize and DurabilityService->Persistent->MemoryMappedFileStore->Size) of the config file were not user-friendly interpretable (e.g. 10M instead of 10485760 could not be used). Solution: The integer where the size is read into is now platform independent. This means that on 64bit platforms the size is read into a 64bit integer and on a 32bit platform into a 32bit integer. Secondly the types of the two parameters have been converted from type 'long' to type 'size'. |
dds2685 |
C&M module not included in the OpenSplice Community Edition The C&M module was not included in the OpenSplice Community Edition. This meant that OpenSplice Community Edition users could not use the C&M module. Solution: The C&M module is included in the OpenSplice Community Edition. |
dds2704 |
Splicedaemon heartbeat check no longer performed The splicedaemon is responsible for checking whether heartbeats from remote splicedaemons are received to be able to detect a node disconnection. When this happens, it is responsible to clean up all resources from any entity from that node. This means that all instances that are currently registered by any DataWriter from the disconnected node need to be unregistered. As a result of that, instance states potentially change and the same goes for the ownership of (some of) the instances. This heartbeat checking mechanism has been implemented in the splicedaemon by means of registering a lease with the lease manager. Every time the lease expires, the service checks whether the connection with one or more nodes has been lost. Due to some re-design the lease to check for heartbeats has been replaced by a lease to send heartbeats. As a result, the splicedaemon now sends twice the amount of heartbeats it should and is no longer checking incoming heartbeats. As a result instance states can become incorrect, ownership is not taken over when it should and memory is not cleaned up properly. Solution: The faulty lease registration of heartbeat sending has been replaced by a heartbeat checking lease registration to make sure the mechanism works as before. |
dds2705 |
The DDSI2 service (beta) may use incorrect type names for topics. The DDSI2 service (beta) failed to include module names in the topic type names. Because of this, interoperability with other DDSI implementations was limited to topics of which the type was not declared in a module. Solution: The module names inside which the type is declared are now prefixed to produce a fully qualified type name. |
dds2721 |
32-bit installers are generated for 64-bit platforms Even though the product is properly compiled as 64-bit, the installer that is generated for 64-bit platforms turns out to be a 32-bit executable. This installer will therefore not run on 64-bit platforms that have no 32-bit libraries installed. Solution: The installer itself now is 64-bit when the product is generated for a 64-bit platform. |
Report ID. | Description |
---|---|
dds2644 / dds2660 |
Java based floating license Java license fixed to work with floating licenses. |
Report ID. | Description |
---|---|
dds2241 |
User-friendly size-specification The OpenSplice shared memory database size can now be specified in M (megabytes) or G (gigabytes) instead of just bytes. e.g. 256M, 2G. |
7592/ dds2367 |
Report plug-in suppress ospl_error and ospl_info log output When using a custom report plugin, it is not possible to suppress the output sent to ospl-error.log and ospl-info.log files. Solution: It is now possible to configure whether or not the output to ospl-error and ospl-info files need to be suppressed using the //OpenSplice/Domain/ReportPlugin/SuppressDefaultLogs tag. |
dds2396 / bugzilla 37 |
write_dispose() returning RETCODE_PRECONDITION_NOT_MET Invoking write_dispose() passing HANDLE_NIL as second parameter returns RETCODE_PRECONDITION_NOT_MET which conflicts with HANDLE_NIL (no handle) as stated in the manual. Solution: An additional check in an internal algorithm in the DataWriter has been added to ensure proper behaviour in this case. |
7946 dds2449 |
OpenSplice services continue to run after the daemon process is terminated Services do not detect when the splice deamon has crashed and keep running. When the splice deamon crashes abruptly the services would never terminate and had to be killed manually. Solution: Services will now properly detect when the splice deamon is no longer available in any situation. This is accomplished by means of a lease mechanism, where the splice deamon periodically renews a lease and services monitor this lease and react once they detect it has not been renewed and thus expired. When the expired lease of the splice deamon is detected it will result in the services shutting down in a normal manner. This does imply that services at worst wait the maximum duration of the lease, which by default has a duration of 60 seconds. If a quicker detection time is desired the duration of the lease should be configured to a shorter time value. This lease duration can be configured in the XML configuration at Domain/Lease using the OpenSplice DDS configurator tool. |
8246/ dds2471 |
DBMS Connect: Mapping with float types Whilst not entirely sure as to why MySQL reports the column to be of type FLOAT, when it has been specified as REAL (and the sql-mode of the connection/server contains REAL_AS_FLOAT). Solution: Extended DBMSConnect to be able to map both types on the internal float-type. |
dds2513 |
TimeToLive not being set on multicast messages Only the default operating system TTL value was being used for multicast, but the user defined value for unicast and broadcast was being respected. Solution: Now propogate correctly the TTL value for multicast messages, with the TTL applying at the network partition level. |
dds2521 |
Error/success reporting on socket creation/binding incorrect in nw_socket_new() The reporting of error/success in nw_socket_new(), was incorrect, which would lead to incorrect "success" message in the network tracing, also not all errors result in a proper error message in the error-log if creation or initialisation of a socket fails. Solution: Reporting corrected. |
dds2524 |
java IDL generated files do not include the super.finalize() in finalize method The output of idlpp did not include a call from finalize() to super.finalize() which is detected by static code analysis files. Solution: finalize() method now calls super.finalize(). |
8650 dds2526 |
Creation of query on dataReaderView fails Creation of query on dataReaderView fails when the query addresses attributes within a record. Solution:Prefixing of the internal name of attributes has been moved to the correct internal component. |
dds2535 |
Robustness: Improve memory depletion detection so it accounts for free 'reusable' memory as well. In the 5.4 release of OpenSplice DDS a new threshold value for the shared memory usage was introduced. When the available shared memory became less then the value of the threshold then OpenSplice DDS would prevent further allocations of data within shared memory. In the 5.4 version there was no recovery possible. Even if applications terminated and freed up memory and enough shared memory would be available, OpenSplice DDS would still not allow further allocations. In effect once the threshold was reached it meant that the deamon had to be restarted. Solution: In this release the algorithm for detecting shared memory depletion has been improved to include freed memory in its calculations. |
dds2540 |
Support for resource_limits.max_instances on DataWriter. Functionality added to provide OMG specification compliance. |
8813/ dds2563 |
DataReaderViewQuery created by MW08 sim is invalid, set_arg fail An invalid DataReaderViewQuery is constructed from a valid expression provided by the application. After creation it is not possible to use it to read data or set its parameters due to a wrong translation from the application attributes to the internal representation of these attributes. Solution: The error in the mapping from application to internal attributes has been fixed. |
dds2570 |
Report plugin to fail on invalid configuration In v5.3 the report plugin was introduced, but on invalid configuration, would just continue and ignore the report plugin. Solution:The behaviour has been changed to fail when invalid report plugin configuration is used. |
dds2598 |
Networking: MaxBurstSize is calculated using uncompressed sizes if compression is enabled If compression is enabled for a networkchannel, the MaxBurstSize limitatiuon for that channel still uses the uncompressed size of the messages. This results in far less data being out on the wire than that is allowed by the "MaxBurstSize" setting. If compression is enabled, the MaxBurstSize limitation should be calculated over the compressed size. Solution: "MaxBurstSize" computations moved to lower level in code to take account of compression. |
dds2627 |
The Tuner always displays the synchronous_write ReliabilityQoSPolicy attribute as false Due to an error in the parsing of the synchronous_write ReliabilityQosPolicy attribute, the Tuner displayed the attribute as being false in all cases. Solution: The parsing algorithm for synchronous_write in the Tuner has been fixed. |
Report ID. | Description |
---|---|
8686/ dds2556 |
Durability service fails in an 8 node system Due to a concurrency issue in the durability service, two threads could modify/read a specific variable at the same time. If that situation occurred it lead to a segmentation fault. The signal was caught and the service tried to clean up and terminate, but went into a spinning loop as a result of that, which was the second issue. Solution: Access to the variable is now protected to make it thread-safe and also the service will no longer get into a spinning loop when an error occurs when the service terminates because of a signal. |
Report ID. | Description |
---|---|
dds398 |
Possible out of order delivery in case of ORDER_BY_SOURCE_TIMESTAMP. In case of ORDER_BY_SOURCE_TIMESTAMP samples could be delivered out of order. Solution: In case of KEEP_LAST this is now avoided by discarding samples that have a timestamp that is older than the lastest consumed sample. Since KEEP_ALL is not allowed to discard samples, it can still receive samples out of order. |
Bugzilla#34 dds478 / dds2353 |
Susbcribing with an illegal topic name cause core dump The OMG specification states in Annex A.2 regarding the TOPICNAME that a topic name is an identifier for a topic, and is defined as any series of characters ?a?, ..., ?z?, ?A?,..., ?Z?, ?0?, ..., ?9?, ?-? but may not start with a digit. OpenSplice will not accept - or /. Solution: Ensured OpenSplice DDS is compliant with the OMG specification and added a better error message where invalid topic names are used. |
dds1504 |
Windows builds put dll's in /lib dll files should be placed in the bin directory. Solution: Changed the packaging of the dll files and they now appear in the bin directory. |
5218 / 7254 / 7255 dds665 / dds2235 / dds2338 / dds2339 |
Exhaustion of shared memory When OpenSplice shared memory limit was exceeded then a crash would occur. Solution: Additional checking and configuration has been added to stop the crash from occurring and also notifying the application developer with messages in the info log that shared memory is reaching pre-configured limits. See the 5.4 release highlights above |
6925 dds1948 / dds2054 |
Gracefully shutdown OpenSplice Where OpenSplice has shutdown unexpectedly then it would leave invalid processes and temporary files on windows. Solution: On startup of OpenSplice, invalid process and temporary files are detected and removed. |
dds2277 |
Crash in create_datareader on ContentFilteredTopic that was created with incorrect filter This was caused by an invalid user filter being added. Solution: The invalid user filter is now handled correctly and error message reported. |
dds2281 |
The Durability Service ExpiryTime had a maximum value of 20 seconds Solution: Removed the a maximum expiry time, so it is now unlimited. |
7319 dds2285 |
The dispose_all command leads to a crash of the splice deamon on
remote nodes where the topic/partition combination is not known. When Node A issues a dispose_all command for a specific topic/partition combination and this dispose all command is received by remote nodes where this topic/partition combination is not known, then a crash in the splice daemon is observed when the dispose_all command is processed and the topic listeners are being retrieved. Solution: When the splice daemon processes a dispose_all command it received and it has arrived at the point the topic listeners need to be received, it will now verify if the partition/topic combination is known. If the partition/topic combination is not known, it will not try to send a notification to the topic listener and thus avoid the crash. |
7466 dds2357 |
DDS and ethernet connection problems on a multi ports computers In OpenSplice DDSv5.2 when the default Ethernet card is not available the first available multicast Ethernet card is selected, previous to 5.2 it was the first available broadcast address. However, this doesn't allow users to block all traffic if the default adapter is not available. Solution: Two new configuration options have been added OpenSplice/NetworkService/General/NetworkInterfaceAddress and OpenSplice/NetworkService/General/NetworkInterfaceAddress[@forced]. For OpenSplice/NetworkService/General/NetworkInterfaceAddress[@forced] it specifies whether only the selected NetworkInterfaceAddress should be used or others can be used too. |
dds2404 |
Errors in Ping Pong examples The issue was caused by the pong implementation doing a "take" of some data, and then writing this straight back to ping. The pong was receiving samples that were invalid at the end of the run due to the ping's writer implicitly unregistering the instances on shutdown (if there are no more samples in the datareader for that instance, an invalid sample is generated). These invalid samples are meant to be ignored by the user, instead pong was forwarding on the samples. Solution: Corrected the problem by adding checks of the form if(infoList.value[j].valid_data) after a take is made and before it is forwarded. |
dds2425 |
idlpp error with typedef of an array with struct in a typedef of array of array with struct Solution: Fixed idlpp to recursively resolve the actual type of a typedef to assure typedefs of typedefs are processed correctly |
dds2426 |
Value of the "MaxBurstSize" configuration of a networkchannel (200KB
per resolutiontick) causes problems when running typical evaluation
tests on an out-of-the-box configuration Solution: Change made so default configuration is "unlimited" for all practical networks in the foreseeable future. |
8686 dds2541 |
Crash on restart of application or daemon When using synchronous write, the splicedaemon or the application itself could crash when terminating. This was caused by the fact that an internal clean-up algorithm sometimes freed an object twice causing the reference count of the object ending up to low. The crash occurred when the reference count reached 0 and the object was accessed after that again. Solution: Fixed the internal algorithm to prevent double frees of the object. |
Report ID. | Description |
---|---|
7121/ dds2129 |
Queried meta descriptor incorrect The capability of querying for the XML metadescriptor string for a specific application type did not return the correct representation in some cases where cyclic dependencies between types in modules existed. Solution: Fixed the internal algorithm that builds up the XML metadescriptor representation. |
7373/ dds2312 |
Crash in the tuner when using an enum The Tuner tool internally using a fixed size buffer to store a key definition of a Topic. In case a key definition exceeded this buffer, the tool crashed. Solution: Changed the fixed size buffer in a dynamically allocated buffer based on the actual size of the key definition. |
7438/ dds2333 |
Performance is too slow The query optimiser failed to optimise queries for key fields due to some internal key name changes introduced by the implementation of DataReader instance key optimisations. This implied that all query read/take actions were linear searches. Solution: The DataReader Query constructor has been made aware of the name translation and now implements these translations during optimisations. |
dds2378 |
MMF-store configuration should not be mandatory The mapping-address and size configuration parameters for the MMF based persistency implementation should be optional, reducing the complexity of out-of-the-box configuration. Solution: The mapping-address now by default lies directly after the mapped database in virtual memory and is by default twice the size of the database to make sure all persistent data and potentially one backup fit in there. |
dds2417 |
DLRL Java Doc from windows start menu gives incorrect doc For the Windows platform not all DLRL documentation was in the distribution and the link in the start menu did not point to the correct page. Solution: Added the missing documentation to the distribution and updated the link in the start menu. |
Report ID. | Description |
---|---|
dds2362 |
Deadlock possible with shutdown of splice daemon Clean up code in the splice daemon could become locked due to a race condition. Solution: Re-factored the clean up process to eliminate the race condition. |
5097/ dds729 |
Splice daemon may consume 100% CPU. When using multiple domains on Windows, services and application processes sometimes requested access to a domain from the wrong splice daemon. This caused unpredictable behaviour in both services and user applications. Solution: As there is only one splice daemon per domain the domain name has been made part of the name of the pipe making the pipe name unique. Furthermore, an illegal request from the Tuner tool to the splice daemon pipe on initialisation has been removed. |
7162/ dds2164 |
OSPL takes a long time to stop. Update on previous fix to allow handling of 0.0 instead of just 0 for time argument. Solution: Corrected parameter handling. |
7319/ dds2285 |
Using dispose_all_data can cause the splice daemon to crash When a remote node calls a dispose on a local topic instance that does not exist then splice daemon will crash Solution: Check to ensure local topic instance exists and if not then queue the dispose request until it does. |
Report ID. | Description |
---|---|
7444/ dds2384 |
Incorrect memory allocation causing ospl deamon to stop immediately after starting The implementation of the code to externally monitor OSPL instances on local/remote node used strlen incorrectly causing memory corruption. This can have the effect of causing the ospl daemon to stop immediately after starting. Solution: Remove the use of strlen and use MAX_HOST_NAME_LENGTH variable. |
dds2180 |
Windows cannot start and stop 2 separate spliceds with separate domains An error occurred starting 2 separate spliceds and domains on the same windows machine. This was because the _ospl_servicePipeName parameter is not unique: When the second spliced is started is uses the same pipe name (and as the max instances param is set to 8) CreateNamedPipe fails because it has exceeded the max instances. Solution: As there is only one splice daemon per domain the domain name has been made part of the name of the pipe making the pipe name unique. |
Report ID. | Description |
---|---|
7444/ dds2367 |
Problems with OSPL Log Report Plugin The initialize and finalize methods of the report plugin were being called twice. This was because the report plugin registration and unregister had been implemented in 2 different places but were actually only needed in 1 place. This also caused the report method of the report plugin to be called after the finalize method had been called. Solution: Changed implementation to only call report plugin register and unregister once. |
Report ID. | Description |
---|---|
6494 dds1506 |
Datareaderlistener is still being triggered while the datareader
and all related objects have already been freed When a datareader is being deleted at the same time it is being triggered asynchronously via on_data_available() then the application may crash. This is caused by an incorrect check that validates whether or not a listener thread should be terminated. Solution: The error in the checking algorithm has been repaired. |
dds1629 |
More helpful error message when using Tuner When using the Tuner, an error message "Supplied URI is not available" would be returned, but this could occur if the URI is not available or the shared memory segment could not be mapped. Solution: The error message has been updated. |
dds2054 |
Gracefully shutdown OpenSplice on machine shutdown If a machine is shutdown during the operation of OpenSplice, it is possible that OpenSplice will not adequately clean up all temporary resource files. This can be particularly an issue on Windows because the shared memory is represented by a mapping to a physical file and can cause problems restarting the domain. Solution: On Windows, "ospl start" has been improved to check for and remove resource files that do not relate to running processes, prior to starting the domain. In future releases this will be improved to remove OpenSplice processes that may be running but are associated with invalid resources file. |
7051 dds2071 |
-n option added to idlpp to override the suffix for the ORB header file include -n <include-suffix> Overrides the suffix that is used to identify the ORB dependent header file (specifying the data model) that needs to be included. Normally the name of this include file is derived from the IDL file name and followed by an ORB dependent suffix (e.g. 'C.h' for ACE-TAO based ORBs). This option is only supported in Corba cohabitation mode for C++. In all other cases it is simply ignored. Example usage: -e .stub.hpp (For a file named 'foo.idl' this will include 'foo.stub.hpp' instead of 'fooC.h', which is the default expectation for ACE-TAO.) |
dds2081 |
Issues with compliance of tooling over different platforms The tooling delivered with OpenSpliceDDS (like mmstat) did not have the same command-line options on all supported platforms. Solution: The options have been made consistent for all supported platforms. |
7068 dds2083 |
Memory leak for arrays When an array was freed, and thus the references contained within that array, the size was incorrectly determined, which caused a memory leak as not all references were freed. Solution: So now when the collection is an array, the size is determined by the maximum size field of its type, which is always set, since an array is always fixed length. |
dds2128 |
Wireshark plug-in not compliant anymore The Wireshark plug-in is no longer showing correct information with the updates in the network format. Solution: Updated the Wireshark plug-in to correctly interpret the newest networking format. |
7162 dds2164 |
OpenSplice takes a long time to stop when using the ospl -f option OpenSplice at least takes multiple seconds to terminate when sending a TERM signal. Solution: Setting the configuration of the already existing ServiceTerminatePeriod to "0.0" will now be interpreted by all OpenSplice services as immediate request to terminate without cleaning up. As a result, all services terminate immediately and do not clean-up anything, except for the shared memory segment that is still destroyed. |
dds2228 |
Registering a type compiled with idlpp -j option causes JVM crash Calling register_type on TypeSupport classes that have been generated using the "-j" option made the application crash. This was caused by the fact that an internal algorithm did not substitute the name of the type with the by idlpp generated type name. Solution: Extended the internal algorithm to substitute the name. |
dds2247 |
Code generated by idlpp (C++) for multidimensional array does
not compile The IDL pre-processor did not generate correct code to copy the contents of multidimensional arrays. Solution: Updated idlpp to correctly handle multidimensional arrays. |
7272 dds2254 |
Error message not displayed The error message "Description : shmat failed for 1 with errno 22 ()" is unclear. Solution: Updated the error message to show the string representation of the error between parentheses or show "Unknown error" otherwise. |
dds2274 |
Network communication on Windows does not work in case of multiple network partitions When multiple network partitions were configured for the networking service, communication on Windows did not occur due to a bug in the algorithm that matches partition-topic combinations to a multicast address. Solution: The error in the matching algorithm has been repaired. |
dds2276 |
Network partition-name conversion to hash results in always the same hash The networking service calculates a hash value for every configured network-partition. Due to an error in the hash algorithm the same hash was calculated for all network-partitions. Solution: The error in the hashing algorithm has been repaired. |
dds2278 |
Not able to view all DDS nodes When discovery for networking is enabled, the data that is published before the other node has been detected is not forwarded by the networking service even though it does see the information. Solution: During the start-up of the networking service, it will now initially assume there is a remote node available and send the data over the network, before actually starting to rely on discovery. |
7319 dds2285 |
Using dispose_all_data can cause the splice daemon to crash Sometimes internal transaction messages were accidently stored as actual application samples after they were handled. Solution: The algorithm now checks whether a message is a transaction message and if so, will not store it anymore. |
7329 dds2294 |
IDL Sequences in C# Generated C# classes for a sequence of structures failed to write a sample. Solution: The marshallers for sequence of structures was incorrect and has now been fixed. |
7371 dds2304 |
Memory leak when continually registering and unregistering the same instance of a sample. Some piece of administration in shared memory was not freed. Solution: Made sure the administration is freed. |
7373 dds2305 |
mmstat does not display valid values for 64-bit architecture The mmstat tool was not able to display memory sizes above the 32-bit boundary, because a 32-bit type was used to store the size. Solution: Updated the type of the field to be 64-bit on 64-bit platforms. |
Bugzilla#34/ dds2309 |
Memory growth (leak) in ospl when sending messages with strings in sequences The memory used by ospl as reported by mmstat grows each time a message is sent when the message contains a sequence of strings or a sequence of structures that contain strings. The internal algorithm that generically frees data in shared memory uses the available meta-data to know how to free the data. Due to an error in the meta-data, the algorithm could omit freeing a structure, causing memory leakage. Solution: Fixed the error in the meta-data for sequences. |
dds2313 |
Application may crash during execution when using sequences The internal algorithm that generically frees data in shared memory uses the available meta-data to know how to free the data. Due to an error in the meta-data, the algorithm could free a structure in the wrong way, causing the application to crash. Solution: Fixed the error in the meta-data for sequences. |
7444/ dds2335 |
Multiple instances with the same key Character array fields can be marked to be stored as strings internally (using the pragma cats). When a sample is written in Java, the internal string had rubbish attached to the end of the string. This was caused by the algorithm that copies samples from heap into shared memory since it did not add a '\0' terminator to the string in case the source is a character array causing the middleware to interpret the string up to the first '\0' in memory after the start of the string. Solution: Extended the copy algorithm to add a '\0' terminator at the end of the string. |
Report ID. | Description | dds2462 |
OpenSplice for Windows CE now uses the CeSetThreadPriority and CeGetThreadPriotiry functions for thread priority administration. This replaces the usage of the legacy SetThreadPriority and GetThreadPriotiry functions. A bug has also been fixed where some threads started within the OpenSplice daemon had the priority value of 0 rather than CE_THREAD_PRIO_256_NORMAL which maps to 251. |
---|
Report ID. | Description |
---|---|
7444/ dds2367 |
Problems with OSPL Log Report Plugin The initialize and finalize methods of the report plugin were being called twice. This was because the report plugin registration and unregister had been implemented in 2 different places but were actually only needed in 1 place. This also caused the report method of the report plugin to be called after the finalize method had been called. Solution: Changed implementation to only call report plugin register and unregister once. |
Report ID. | Description |
---|---|
7157/ dds2158 |
Bad parameter passing in reportPlugin initialize. The report plug-in function definitions and calls did not comply with API defined in deployment guide Solution: The function definitions have been compliant with the deployment guide. |
7270/ dds2251 |
Unable to publish an InternalFlight. Java application crashes during the writing of a sample. The issue was that a size marker used in creating the Java meta data was of type "short", which is too small. Solution: A bug has been fixed in the Java implementation of OpenSplice which restricted the size of the meta model of an IDL component (as signified by a #pragma keylist) to 2^16 : 65536. This has been rectified and the maximum size of an IDL component's meta model is now 2^32 : 4294967296. |
Report ID. | Description |
---|---|
7247/ dds2229 |
The spliced process goes to 100% load When running on a 64-bit platform, the spliced goes to 100% load in some situations when accessing a virtual memory address above the 32-bit range, because of a wrong cast being done in an internal algorithm. Solution: The wrong cast has been corrected. |
7250/ dds2237 |
OSPL unable to create a 4GB shared memory segment When specifying a database size over 2GB in the configuration file, OpenSplice could not be started. This was caused by two things:
Solution: The size is now an unsigned long long for 64-bit platform to solved the first issue. To solved the second issue, a default address of 0x140000000 has been configured for 64bit platforms, but the suitability of this value will vary depending on the size of the database that is required, and also with the process properties at the operating system level. If the address is not suitable, the error log file will describe this. On linux the "pmap" utility when ran with the "-x" option can be used to view the mapping addresses in use by a particular process. A mapping address value is suitable if the addresses that immediately follow it (up to the size of the required database) are not in use. Refer to the OpenSplice_Deployment guide for details of the OpenSplice/Domain/Database/Address attribute. |
dds2243 |
Topic name for Read/Write access QoS policy interpreted as type name The XML configuration allows for topic access elements ('OpenSplice/Domain/TopicAccess') to be defined which determine whether reader and/or writers are allowed to be created for a specific topic expression (by means of the 'topic_expression' attribute). A bug existed in the implementation of this feature where the topic expression was compared to the type name instead of the topic name. In effect no proper access control could be defined for topics. Solution: The implementation has been changed so that it now compares the topic expression with the topic name instead of the type name. |
Report ID. | Description |
---|---|
6488 dds1513 |
OpenSplice can be configured to allow a single application to operate in multiple domains |
6457 dds1542 |
Generation of additional inappropriate files from idlpp If A.idl depends on B.idl, and A.idl defines a "pragma keylist", then we need to apply idlpp on B.idl too so that BSplDcps.h is generated. However idlpp will also generate BDcps.idl which only contains namespaces and no struct. Solution: idlpp amended to generate only the appropriate files. |
dds1619 |
DDS shall provide a way to dispose all the instances of a topic in one call
and to notify the receivers in one notification See new features in 5.2 above. |
dds1929 |
Dynamic namespaces for durability
Solution: OpenSplice DDS now supports more dynamic environments by introducing dynamic namespaces. With a dynamic namespaces, the durability service can determine at runtime if a previously unknown namespace discovered from another node needs to be maintained and\or aligned locally. For more information see the related documentation. |
dds1958 |
Warning when opensplice is close to exceeding its amount of configured shared memory
Solution: OpenSplice DDS will now print a warning to the ospl-info.log file if the available, non fragmented shared memory is below 10% of the total available shared memory size. This warning will only be printed once. |
dds2099 |
Process not terminating if several listeners attached to serveral entities Using a domain participant in combination with a listener caused a deadlock situation to occur within OpenSplice DDS in situations where the listener was processing events while the DomainParticipant was already being deleted. The deletion operation would claim a mutex and perform various clean functions including stopping the listener thread and then waiting for it to terminate. Meanwhile the listener thread could be processing some events and require the mutex lock in use by the participant delete function. Solution: Fix the delete function of the participant to release the mutex while waiting for the listener thread to exit and reclaim it once the exit function returned. |
7122 dds2131 |
Unclear error reports when osplconf updates the configuration Configuration not parsed properly. Solution: Fixed an error in an XPath expression that is used internally to parse part of the configuration and improved the warning messages that appear in the info-log file. |
dds2145 |
Developers needed to add dcpssacs.dll to the PATH to use C#
Solution: For C#, there is now a signed and versioned Assembly called dcpssacsAssembly.dll, created from the the dcpssacs.dll module, which is added to the Global Assembly cache. In this release, as part of the installation, this is added to the Global Assembly cache. |
dds2154 |
SPLICE_ORB environment variable not set In order to build the CORBA C++ example you require the SPLICE_ORB to be set. This is not done in release.com for Unix systems. Solution: Added the variable to release.com. |
7131/ dds2157 |
OpenSplice release.com loads jar files at the end of CLASSPATH Solution: script fixed. |
7157/ dds2158 |
Incorrect parameter passing in report plugin. Parameter passing in the Report Plugin intialise function was incorrect. Solution: Small code error fixed. |
7156/ dds2159 |
Missing report configuration in osplconf. osplconf tool was not up to date with the deployment guide and did not allow some of the configuration items to be set. Solution: Updated the configuration to match the deployment guide. |
7158/ dds2160 |
spliced crashes when adding a report plugin Solution: Small code error fixed. |
dds2163 |
When using the pragma 'cats' in combination with multiple listed fields,
the IDLP Pre-processor would crash and not generate any code There was a bug within the IDL Pre-Processor idlpp which occured when the 'cats' pragma directive was being evaluated and incorrectly take 2 fields from a list. This caused a null pointer to be retrieved from the list in the next iteration, where a valid field was expected, and idl would crash when the pointer was dereferenced. Solution: The double take from the list has been removed. |
dds2167 |
Example C#Visual Studio Solution and Project files should be valid for both VS2005 and VS2008 In 5.1, the C# example projects would only support Visual Studio 2008 Solution: Projects are now created with Visual Studio 2008. |
dds2173 |
Coherent updates See new features in 5.2 above. |
dds2188 |
Builtin topic default QoS incorrect in relation to reliability.max_blocking_time The specification specifies that the max_blocking_time of the reliability QoS should be 100ms. Currently it isset to 0ms. Solution: Updated code to set QoS to 100ms. |
7190/ dds2192 |
Instances are unregistered on the DataWriter when its liveliness expires All instances are unregistered on the DataWriter when its liveliness expires, but this is wrong behaviour. It causes write calls with an instance-handle to return PRECONDITION_NOT_MET after liveliness expires. Solution: A DataWriter no longer unregisters its instances when its liveliness expires. A DataReader does reset the ownership of an instance when liveliness of DataWriter expires in case of exclusive ownership. |
7186/ dds2204 |
Communication between different subnetworks not supported To be able to communicate between different subnetworks the DontRoute networking service configuration option needs to be set to false for all channels including discovery. Even though the networking service supports this configuration, the configurator tool did not offer this option for discovery. The same goes for other send/receive options that are available for channels. Solution: The configurator has been extended to support all send/receive options for networking discovery as for the channels. |
Report ID. | Description |
---|---|
dds2078 |
Disabled factories default QoS settings not returned. Default Qos settings for factories are not returned if the factory is in a disabled state. If the caller did not initialize the Qos parameter (assuming get_default_qos would do that), then an uninitialized QoS would be returned. Solution: Removed the check for disabled factories, so that they are handled like enabled factories. |
7104 dds2108 |
make_hde will not copy host libs for linux cross compile If the host and target operating systems are the same, in this case Linux, but the target architectures are different the make_hde will not copy the host lib's. Solution: Changed the check in the make_hde script. |
dds2133 |
PATH environment variable corrupted by checkconf script on Windows build
when using Visual Studio 2008 The ospli/bin/checkconf script, which is included by configure, is responsible for checking build requirements and prepares the environment for building OpenSplice. The part of the script that checks the Microsoft toolchain contains an error when Visual Studio 2008 is detected. It uses 'cygpath -d' (line: 186) to translate a path but this returns the windows representation of the path (including c:\). The result is then appended to the Unix PATH environment variable, which uses ':' as a field seperator. This corrupts the PATH variable which causes the build to fail. Solution: Script amended to handle windows paths correctly. |
dds2165 |
Set default QoS behaviour not compliant with the OMG specification. The factory operations set_default_xxx_qos() (where xxx represents an entity type, i.e. topic, publisher, subscriber, datawriter and datareader) do not work according to the DDS spec when their corresponding XXX_QOS_DEFAULT label is passed as parameter. According to the DDS specification, the set_default_xxx_qos operation should in that case: "The special value XXX_QOS_DEFAULT may be passed to this operation to indicate that the default QoS should be reset back to the initial values the factory would use, that is the values that would be used if the set_default_xxx_qos operation had never been called." In OpenSplice DDS these operations either do nothing and return RETCODE_OK (e.g. set_default_publisher_qos), or return a RETCODE_BAD_PARAMETER (e.g. set_default_subscriber_qos). Solution: Updated the functionality to behave as per the specification. |
dds2177 |
PartitionID changed to be byte swapped before being sent out over the network |
Report ID. | Description |
---|---|
7053 dds2046 |
When an application attach listeners, then it does not terminate and just hangs after the main thread finished. Java 'hangs' on exit when not all DDS entities have been deleted. Solution: A shutdown hook is now installed on the JVM. The shutdown hook is installed during the first successful call on the DomainParticipantFactory.get_instance() operation. This shutdown hook will ensure all contained entities of the DomainParticipantFactory are deleted before the final shutdown of the JVM commences. Before this fix the contained entities where deleted during/after the final shutdown of the JVM which caused various JVM resources to be accessed by OpenSplice while they were no longer available. |
7141 dds2143 |
Code generation with OSPL IDL pre-processor fails for specific key types The IDL pre-processor rejects usage of character arrays as key fields in IDL. This occurs even though the character array has been marked to be interpreted as a string internally by the middleware via a '#pragma cats' definition. The IDL pre-processor fails to recognize the '#pragma cats' definition during key field validation if it is located in a different module and/or in an included IDL file. Furthermore the IDL pre-processor incorrectly validates character arrays when they are used in combination with typedefs. Solution: The IDL pre-processor now is able to recognize the '#pragma cats' definition regardless of it's location and will also correctly resolve any used typedefs in determining whether or not a key field is a character array. |
7140 dds2144 |
TIMEOUT when writing a topic with synchronous reliability activated When activating the synchronous reliability on DataReaders and DataWriters, the write returns an error code 10 (TIMEOUT) even with large timeout (5 secs). Solution: A fix has been implemented to prevent synchronous writers from blocking on ACKs from non-matching readers (i.e. subscribed on other topics). This was because when the writer was created, *all* synchronous readers were injected into its administration. But as the non-matching readers never received messages from this writer, they would never acknowledge and the writer would wait infinitely (until timeout). |
Report ID. | Description |
---|---|
4838 dds505 |
Shared DataReaders needed for efficiency Applications need facilities that allow them to share one DataReader over multiple processes on the same node. This allows reducing the memory footprint on a node significantly in case a lot of applications are interested in the same data Solution:A new 'share' QoS policy has been introduced in the SubscriberQos and DataReaderQos that allow applications to share these entities. |
dds1011 |
Not all .bat files in the 'bin' directory of the release use Windows line endings Some .bat files in the 'bin' directory of the Windows release use UNIX-style line endings. Solution:Changed the line-endings of the .bat files to Windows-style line-endings. |
6569 dds1541 |
OSPL does not generate default entry of switch of unions in C++ According to the IDL to C++ mapping, the IDL pre-processor should generate a default entry for in case no all the possible values of the switch are defined as cases and no default case is modelled in IDL. Because this is not done, compiling the generated code will show warnings. Solution:A default entry is now generated by the IDL pre-processor to make sure compiling the generated code does not show any warnings. |
6556 dds1551 |
User application conflict with cfg_parse operation OpenSplice has a 'cfg_parse' function that is also used a lot in other products. This leads to a name collision when OpenSplice is used in combination with these other products. Solution:The 'cfg_parse' function has been renamed to 'cfg_parse_ospl' to prevent collisions. |
dds1552 |
Support for abstract domain identification in create_domain_participant OpenSplice V4 and before only support URI as domain ID Solution:A DomainId consists of a string that represents either a URI to the location of the configuration file (e.g. file:///projects/DDS/ospl.xml) or the Domain name as specified in the configuration file. The actual value returned is dependent of the value used when creating the DomainParticipant, also see the DomainParticipantFactory create_participant operation. If a DomainParticipant is created using the Domain name then it will also return the Domain name and visa-versa when created using a URI then the URI will be returned by this operation. The configuration file, identified by the URI, specifies all configuration details of the Domain. |
dds1677 |
os_threadCreate failed because of wrong stack size If on the system the PTHREAD_STACK_MIN is smaller than OS_STACKSIZE_DEFAULT the call to pthread_attr_setstacksize in os_threadCreate (POSIX) will fail. Solution:PTHREAD_STACK_MIN is now used instead. |
6815 dds1829 |
Persistent data can get lost Due to the durability alignment strategy, persistent data can get 'lost'. The durability when not master for a name-space sets aside the locally stored persistent data set. Then it aligns with the master. When during this process, the host is reset (actively or power-fail), the persistent data-set may not be complete. Solution:The durability service now stores information about having a complete set or not on disk. If it detects on re-start that the set is not complete, the 'old' set is restored and injected instead to at least have a complete set of data. |
dds1836 |
On VxWorks - ospl reports warning sched_getscheduler failed with error 3 in cmsoap service This occurs because the sched_getscheduler requires a task id in vxworks6. Solution:Properly handle the case when os_procAttrGetClass is called by adding a vxworks6.x specific version of os_procAttrGetClass that passes the current task id rather than process id to sched_getscheduler |
dds1844 |
Addition of networking statistics A large number of statistics for the networking service have been added and these are now available in the OpenSplice Tuner. By default, statistics are switched off. The enable them, the following needs to be added to the configuration file under //OpenSplice/Domain: <Statistics> <Category name="networking" enabled="true"/> </Statistics> |
dds1847 |
Support multiple installations of OpenSplice DDS on Windows platforms OpenSplice DDS uses system variables and thus installing a new version of OpenSplice will overwrite the variables for the previous installation. Solution:OpenSplice does not use system variables anymore. A release.bat script is available in the installation directory to configure your environment. |
6792 dds1865 |
The IDL pre-processor must report unsupported key types The IDL pre-processor does not report illegal key fields in a type when the attributes are there, but the type of the attributes is not supported as a key field. Besides that, it also still generated code, causing problems with this code only at run-time. Solution:The IDL pre-processor now reports unsupported key fields and does not generate any code in that situation. |
dds1894 |
durability: Two nodes select each other as master, blocking alignment Due to an error in the algorithm that determines the master-ship of a durability service, multiple masters can appear, causing the alignment of non-volatile data to fail. Solution:The error in determination of the master has been repaired. |
dds1899 |
Asymmetric NetworkPartition configuration Right now networkPartitions are characterized by an ID that's determined by the order in which they're defined in the config-file. This does NOT allow for asymmetric configuration. Solution:A name-hash is calculated to create an ID that allows asymmetric configuration of networkPartitions over nodes. |
6861 dds1903 |
Network partition expression more limited than described and needed The network partition.topic expression is more limited than described in the deployment manual regarding the usage of wild-cards ('*'). Solution:Wild-cards are now supported on all levels for network-partitions. |
dds1914 |
Improve reporting of non-matching topics Currently when an application creates a topic that already exists in the system it will be compared against the existing one and the creation will fail if the definition is not equal. The error reports that are generated are not useful as they often do not report the topic name. It would be very helpful if users know which topic creation fails and what the cause is, preferable also report conflicting values. Solution:The type name of the topic is now reported when the user tries to create a topic with the same name but a different type. |
6879 dds1921 |
Illegal object reported followed by segmentation faults Applications sometimes report detecting an illegal object after which they crash. This is caused by an inconsistency in the kernel instance administration. Solution:The error in the internal administration of instances in the kernel has been corrected. |
dds1926 |
Setting parameters of a query is too expensive Setting a parameter on a simple query (using only '=' i.c.w. AND) is too expensive and results in congestion on the memory manager. Solution:Instead of fully reconstructing a query, only the parameters are now changed for simple queries. |
dds1932 |
State the paths of the ospl-info and ospl-error files immediately after running ospl start |
6904 dds1938 |
Set query parameters should reset the query cache Setting the query parameters should reset the query cache because the query when executed over the reader database might pass samples with the new arguments. This causes the reader not to trigger the application when new data arrives and also can lead to the situation in which an application cannot read any data from the reader, even though there is data. Solution:The cache is now re-evaluated when a query is modified and also when historical data is delivered. |
dds1946 |
Corrupted caching instance pipeline The instance pipeline in the kernel is formed by caching destination instances at source instances. The design supports disconnecting sources but disconnecting destination instances is poorly implemented. This may lead to corrupted instance caches and in some cases temporary memory leakage. Solution:The pipeline implementation has been modified to allow disconnecting destination instances also. |
dds1947 |
Default JDK is now Java 6 OpenSplice DDS V5 is compiled with JDK6, but is compatible with JDK5 as we use the -source 1.5 flag |
dds1965 |
Persistent data-store completeness needs to be communicated to late-joining nodes When nodes are aligned simultaneously, it is possible for them to detect whether they did receive a consistent (complete) persistent data-set or not. In this scenario, these nodes can take action to prevent injection of incomplete persistent data the next time the node starts. However, for late joining nodes there is no way to detect if a currently available persistent set is complete/consistent. Therefore, if a system would be restarted, and a previous late joining node from a system where incomplete persistent data was injected would become master for this set, injection of incomplete data would happen. Solution:Late joining nodes get information about persistent data completeness of other nodes. |
dds1986 |
Problem with persistent topic having exclusive ownership strength
Because exclusive persistent/transient samples were not stored
correctly, writers that no longer existed (for example from a
persistent store) could in some cases prevent new writers from
writing new data. |
6989 dds1988 |
Linux 64 bit crashes when using a shared memory address above the 4GB range When using a shared memory address which is not aligned on 8192 bytes, OpenSplice DDS would crash. Solution:Added functionality to print a warning in the info log when the configured shared memory address is not aligned on 8192 bytes on 64-bit platforms. DDS will also automatically select the next correctly aligned address and use that instead. |
dds1998 |
Problem with unions in durability in debug mode Due to an invalid assertion, the durability service crashed in debug mode when using specific constructions within a union. Solution:The invalid assertion has been removed. |
dds1999 |
Problem with durability when publishing a union with switch set to a value without corresponding case Due to an invalid assumption in the serialization routines, the durability service is not able to handle unions correctly when the switch is set to a value without a corresponding case. Solution:The serialization routines have been updated to handle these unions correctly. |
dds2007 |
Durability crashes when tracing is enabled and needs to print '%' character The durability service crashes while tracing information when there's a % in the path, because 'fprintf' takes it as format. Solution:The information to print using 'fprintf' is done using a format with the string as argument instead of directly printing the string. |
dds2013 |
idlpp doesn't allow typedef of interface Even though no code is generated for (typedefs of) interfaces by the IDL pre-processor, it should still allow these constructions to be modelled in IDL. Solution:The IDL pre-processor has been modified to ignore typedefs of interfaces as well. |
7012 dds2015 |
idlpp truncates case label in case of enumeration for java If the enumeration label inclusive it's scope is larger than 100 characters, the label truncated to 99 characters. Solution:The length of an enum label is now dynamically determined. |
dds2022 |
cppgen for Windows can not handle multiple #includes unless
separated by more than a new-line In cpp (C Pre Processor) this occurs when you do not have a carriage return at the end of an included idl files and there is another idl file included after it. Solution:The cpp has been modified to handle this situation correctly. |
6974/7023 dds2025/dds2089 |
CTRL+C in Java has no effect, JVM must be shutdown using KILL -9 Solution:The exit handling in DDS has been tightened by installing a shutdown hook to ensure that all the DDS entities are cleaned up and therefore processes can terminate. |
6861 dds2029 |
u_queryNew accesses unset variable copy In the function u_queryNew the variable copy is accessed and used to dispose (q_dispose) while it's value has not been set. Solution:The variable is no longer disposed when it has not been set. |
dds2043 |
Allow bigger value of InitialDiscoveryPeriod
The InitialDiscoveryPeriod max value allowed is 10s. More is
required by customers. |
dds2045 |
PolySpace issues Static code analysis with PolySpace demonstrates some (potential) problems in the code-base. Solution:The detected problems in the code have been repaired. |
7062 dds2072 |
Deadlock in Opensplice callstack when calling write() and take()
at the same time Due to an error in the internal locking strategy in the OpenSplice kernel, an application can get into a deadlock when three threads are concurrently operating on the same set of objects. Solution:The error in the internal locking strategy has been repaired. |
Report ID. | Description |
---|---|
OSPL1078/ 10788 |
Support for parallel demarshalling should be available on Java an C++ Demarshalling data to a language-binding specific format may take considerable processing depending on the type of the data. Since demarshalling is done by the application thread performing the read/take, demarshalling occurs single-threadedly, limiting the throughput that can be achieved. Solution: The C++ and Java datareaders have been extended with support for demarshalling with multiple threads. The number of threads used can be controlled by using the new set_property(String, String) operation on C++ and Java datareaders. The property "parallelReadThreadCount" accepts a string representing a positive integer (e.g., "4") as value. If the call was successful, successive read/take operations on that datareader will use the provided number of threads for the demarshalling step of the respective operations. dr.set_property("parallelReadThreadCount", "4"); |
Report ID. | Description |
---|---|
dds2327 |
OpenSplice supports the get_discovered_xxx() APIs DomainParticipant; get_discovered_participants(), get_discovered_particpant_data(), get_discovered_topics() and get_discovered_topic_data() for C, C++ and Java. |
Report ID. | Description |
---|---|
dds2323 |
Implementation of new DataWriter and DataRead API DataWriter; get_matched_subscriptions(), get_matched_subscription_data() and get_publication_matched_status() DataReader; get_matched_publications(), get_matched_publication_data() and get_subscription_matched_status() DataWriterListener; on_publication_matched() DataReaderListener; on_subscription_matched() |
Report ID. | Description |
---|---|
dds845 |
DestinationOrderQosPolicy value is not listed in the language specific DCPSPublication topic The DDS 1.1 specification mistakenly forgot to mention the DestinationOrderQosPolicy in the DCPSPublication builtin topic, which should have been there since the policy is marked as being RxO. This omission was repaired in the DDS 1.2 specification. Although the DestinationOrderQospolicy is actually transmitted in the DCPSPublication builtin topic, it still misses in the language specific representation for C, C++ and Java. Solution: Added the C, C++ and Java representations for DestinationOrderQospolicy. |
dds2220 |
The get_key_value() method implementation on the dataReader is missing The DataReader misses the specified get_key_value() method. Solution: Implemented the missing get_key_value() method on the dataReader. |
Report ID. | Description |
---|---|
GEN_7 dds1577 |
Report Plug-In facility See the release highlights section |
DDS_4 dds1578 |
Network level compression See the release highlights section |
DDS_29 dds1617 |
Create a persistent snapshot The durability service allows an application to request a snapshot of the values of instances of a topic. A snapshot consists of a copy of the current persistent data in a new file, the name of which is provided by the application. |
dds1837 |
The C++ API does not support all Quality of Service The C++ API (sacpp & ccpp) did not support the following QoS;
Solution:QoS-ses are now implemented |
Bugzilla#32 |
IDL array of arrays generates code that dosn't compile with standalone C++ mappingidlpp / cppgen bug fixed such that the following IDL now generates compilable code: typedef unsigned short Matrix[1024][1024];
|