Aeron Versions Save

• [Java] Fix ClusterBackup getting stuck in the LIVE_LOG_REPLAY state.
• [Java] Validate that the egressChannel is set.
• [CI] Add Clang 18 to the build matrix.
• Upgrade to SBE 1.31.1.

Noteworthy Changes

• Response Channels
  • A new experimental feature that provides server/client behaviour to Aeron clients.
• Asynchronous DNS Name Resolution
  • Improve Publication/Subscription creation performance and avoid stalls by performing DNS resolution asynchronously with Aeron.
• Tethering Behaviour
  • Prevent premature disconnections by utilizing the maximum window size for tethering, even in congested networks.
• Adjust Overrun Behaviour
  • Make better use of buffers and reduce dropped consumers.
  • Enhance the performance of receivers in a max flow control use-case.
• Congestion Control Recovery
  • Significantly reduce recovery time in congested networks, without the need for any configuration changes.
• Loss Recovery Improvements
  • Reduce negative acknowledgements and message redelivery to improve the efficiency of message recovery.

Known issues

• ClusterBackup is stuck in the LIVE_LOG_REPLAY state as it trying to find recording using wrong archiveId. Fixed in 1.44.1.

Changelog

• [Java/C Driver] Support configuration flag for experimental features (#1577)
• [C Driver] Fix buffer overflow when updating counter labels.
• [Java ConsensusModule] Log cluster session failures.
• [Java/C/C++ Client/Driver] Add ability to set name on the Aeron client. (#1581)
• [Java Driver] Async DNS resolution. (#1564)
• [C Driver] async DNS resolution for the C media driver (#1566)
• [Java CluserBackup] Have cluster backup start from a specific position (#1579)
• [Java ConsensusModule] Introduce clustered service specific SnapshotDurationTracker (#1575)
• [Java Cluster] Allow setting ClusterClock via aeron.cluster.clock system property.
• [Java Driver] Call force on CnC file when doing clean shutdown and signalling ready.
• [C Driver] Add aeron_msync and use to flush CnC file contents upon start and shutdown.
• [Java Cluster/Archive] Add *MarkFile.force() method to flush any pending updates to disc and use it for signalling readiness and when closing the mark file.
• [C/Java Driver] Dissect frames and commands using key/value format.
• [C/Java Driver] Use nanosecond resolution for logging messages.
• [C Driver] Install pre-/post-interceptors logging interceptors if the incoming/outgoing chains are not NULL. Change logging order for outgoing events to log before executing command to match Java and log plain messages before encryption.
• [Java Driver] Fix FRAME_IN/FRAME_OUT event logging, i.e. allow redefinition of previously loaded classes.
• [C Driver] Scope logging functions with a "log" struct. (#1556)
• [C Driver] Add C driver logging for send_nak_message and resend (#1553)
• [C Driver] Track pending setups for cases when notified of unconnected messages. Only allow pending setups for unknown interest or existing images when in ACTIVE state. Handle unconnected streams by sending setup eliciting SMs.
• [C Driver] Fix potential leak in when adding stream interest to subscription by session.
• [Java Samples] Improve response channels samples.
• [C Driver] Align min flow control implementation with Java + allocate memory only if a receiver is added.
• [Java ConsensusModule] Use fixed 64 byte alignment that does not depend on the CACHE_LINE_LENGTH and is backwards compatible.
• [Java ConsensusModule] Fix IndexOutOfBoundsException when recording entry straddles page size.
• [Java ConsensusModule] Add leadership term id counter to ease debugging.
• [Java ConsensusModule] Add election counter.
• [Java ClusterTool] Ensure publication is connected before making cluster members query to avoid exception on checking result from ClusterTool.
• [Java] Add missing properties to the toString + print Archive/ConsensusModule/ClusteredServiceContainer configuration if aeron.print.configuration=true.
• [C/C++/Java Client] Re-assemble header for the fragmented message before calling user code.
• [Java Archive] Add Archive threading mode to the conductor duty cycle counter labels.
• [Java ClusteredServiceContainer] Trap and log snapshot exceptions while allowing cluster to resume (#1568)
• [Java ClusteredServiceContainer] Track ClusteredService lifecycle via a single field.
• [Java ConsensusModule] Remove retry logic from the ConsensusModuleProxy.
• [Java ConsensusModule] Retry sending cluster member query.
• [Java Cluster] Clarify Javadoc around clusterSessionId for offer/tryClaim methods.
• [C Samples] add locks around channel info map in response server sample app
• [Java Cluster Client] Close any partially created ingress publications and egress subscription if connect was not completed.
• [Java Client] Add Aeron#asyncRemoveSubscription.
• [Java ArchiveTool] Update Help section of Archive Tool to describe the different 'describe' commands. (#1573)
• [C++ Archive Client] Add archiveId API to the AeronArchive.
• [Java Archive Client] Use archiveId for looking up the RecordingPos counters. (#1571)
• [C++/Java Archive Client] Add getMaxRecordedPosition to AeronArchive to get the length of a recording if it is active or stopped.
• [Java/C++ Archive Client] Update Aeron Archive control protocol minor version.
• [C Driver] Align implementation of retransmit handler with Java driver.
• [Java Cluster] Set isLeader appropriately on cluster member when listing. Issue #1569
• [Java Driver] Remove unused sourceAddress from PublicationImage.
• [Java Driver] Allow the RetransmitHandler to prevent retransmissions where the term offsets don't match but NAK'd region is wholly subsumed by an existing retransmission.
• [C Driver] Per stream nak delay (#1570)
• [Java Driver] Add support for per channel configuration of nak-delay.
• [Java Driver] Add fixed loss interceptor for testing (#1551)
• [Java Driver] Suppress retransmits of the offset of the NAK is in the range of any active retransmits.
• [Java Driver] Clamp all retransmissions at the point where we create the RetransmitAction to ensure clamped length is accurately applied.
• [C/Java Driver] Change multicast NAK backoff and unicast delay linger to 10ms from 60ms.
• [C++ Wrapper Client] Prevent sanitiser errors when shutting down subscription and client in quick succession (#1549)
• [Java Archive] Validate replay completely before creating a replay Publication.
• [C Samples] add response_client/server sample apps (#1559)
• [C Driver] check for response parameters when adding destinations (#1565)
• [Java Driver] add validation when control-mode=response (#1563)
• [C Utility] Add support for PUT requests using aeron_http_util. (#1561)
• [Build] Set aeronmd rpath in the generated package. (#1557)
• Upgrade to Agrona 1.21.1.
• Upgrade to SBE 1.31.0.
• Upgrade to ByteBuddy 1.14.13.
• Upgrade to ASM 9.7.

• [Java/C] Allow publication image flow control under runs if they are heartbeats so a stream can be kept connected and to get a more accurate EOS when using MDS.
• [Java/C] Restrict the allowing of heartbeat under runs to only one term behind.
• [Java/C] Count under runs for heartbeats.
• [Java/C] Track EOS position per image connection for MDS.

• [Driver] Support specifying endpoints and tags in the same URI
• [C Driver] Add reference id to counters.
• [Driver] Set reference id of sub-pos to registration id of the image/publication.
• [C Client] Fix aeron_print_counters.
• [Java Driver] Touch ups for RetransmitHandler.
• [C Driver] Put enum events in order.
• [C Driver] Remove reliance of max event num in enum for driver agent.
• [C Driver] Fix signed comparisons.
• [C Driver] Declare void param.
• [C Driver] Keep loop types consistent.
• [Java Driver] Add a clamp on retransmit length.
• [Java Client] Fix Aeron.getSubscription(long) JavaDoc.
• [Java Driver] Encapsulate retransmit multiples constants.
• [Samples] Add sample code to support stress testing of MDC and Unicast connections.
• [Cluster] Mark services as active only after they have successfully started.
• [Driver] Improve error messages reported by clients if the Media Driver has been shutdown.
• [C Client] Populate the client image subscriber_position_id.
• [Cluster] Improve error message when cluster fails to connect.
• Add detailed version information to Aeron error counters for all components. Include release version and Git SHA.
• [C Driver] Fix capture of spy subscription channels so that image availability callbacks include the correct subscription URI.
• [C] Use -fno-omit-frame-pointer on Linux native builds so provide easy and accurate profiling with tools like perf.
• Add +guilty to version information if using a build that has changes that have not been committed to git.
• [C Client] Correctly handle the case where an image is used across multiple subscriptions. Including ensuring the unavailability callbacks delivered correctly.
• [Archive] Exclude invalid recordings from ArchiveTool describe and add a describe-all command which shows valid and invalid recordings.
• [C Driver] Introduce a parameters structure that can be used to carry transport configuration information when initialising a udp_channel_transport. Also pass MTU information through the bindings to allow for improved validation of within the DPDK bindings.
• [Driver] Only include timestamps on packets that carry the DATA_BEGIN_FLAG to prevent accidental corruption of packets when the timestamp is being carried within the payload.
• [Cluster] Remove the need for MDS subscription on cluster ingress by having a separate IPC ingress subscription when required. Plus await sockets closing on the log subscription in parallel with the other activities when preparing for an election.
• [Cluster] A leader should assert its leadership when a follower requests a vote and the leader has a more up-to-date log rather than re-initialising an election which could result in log truncation.
• [C Driver] Ensure loop iteration index is used when setting timestamps for publication.
• [Cluster] Log event for unknown session close in service container rather than throw and exception.
• [Cluster] Throw exception if a member id is not found in the cluster backup's configuration.
• [Cluster] Avoid list creation when throwing an exception.
• [Archive] Add ArchiveTool mark-valid, mark-invalid commands. These provide a way to patch the catalog in an emergency.
• [Cluster] Ensure that the ClusterMarkFile link is created using the parent directory from the MarkFile object.
• [Driver] Add stricter validation on control addresses, including: Require control address with control-mode=dynamic. Allow control addresses on non-MDC unicast publications. Add clearer validation on control-mode when using tags.
• [Driver] Introduce an enum to track the control mode specified on Channel URIs.
• [Driver] Update pub-pos when updating pub-lmt.
• [Java Driver] Check if CnC file has error log buffer allocated.
• [Driver] Print usable space when throwing StorageSpaceException or returning an error.
• [C Samples] Make AeronStat respect update interval option. AeronStat was effectively ignoring -u and incorrectly using -t instead.
• [Java] Fix NullPointerException in LogBuffers when construction fails with an exception.
• [Driver] Add counter to represent the number of bytes mapped by the Driver.
• [Driver] Pass in setup triggering SMs into flow control so that tags can be matched for controlling setup sender limiting.
• [C] Fix client image reference counting by use the appropriate literals to ensure that 64-bit parameters are used.
• [Cluster] Call unexpectedTermination on incompatible app versions in the ConsensusModuleAgent.
• [C Driver] Implement low storage warnings for the C driver.
• [Java Driver] Backoff from sending when a short send it detected by polling for status messages as a means to provide a little breathing room for underlying stack.
• [Cluster] Detect when log subscription in an election has an error, such as bind exception, throw an exception giving context and restart election.
• [Cluster] Close log subscription after prepareForNewLeadership.
• [Cluster] Introduce SnapshotDurationTracker to track the time taken for snapshots in the Cluster.
• [Java] Rename aeron-version module to aeron-annotations.
• [C Driver] add max_retransmission_length.
• [Java] Remove suffix from version.
• [Cluster] Stop log recording even if an image has not been picked up and no need to try stop the recording by identity if stopped by log subscription id.
• [Cluster] Increase visibility from private to package-protected for a number of ClusterTool utility methods.
• [Cluster] Add STANDBY as a component type to cluster mark file codec.
• [Cluster] Add cluster services directory to the cluster mark file and allow it to be set via the ConsensusModule.Context. This means that ClusterTool can resolve the location of the service container mark files even if they are not stored in the same directory.
• [C++] Use static constexpr for declaring constants.
• [C/C++] Use functions to return version information to enable runtime version checks against the media driver.
• [C++ Wrapper Client] Allow to access the underlying client of the C++ wrapper.
• [C Driver] Fix error reporting for aeron_port_manager.
• [C Driver] Implement timestamp capturing for the recvmsg path.
• [C Driver] Switch sender poller to poll for a single SM, i.e. do the recvmsg syscall.
• [C Driver] Add additional validation for frame_length on status messages.
• [C Driver] Fix error format strings when configuring sockets for the udp transport.
• [C/C++] Remove date/time from the version files for a reproducible build.
• [C] Fix line ordering bug and add missing return statements and error appending.
• [Driver] Allow publication image flow control under runs if they are heartbeats so a stream can be kept connected and to get a more accurate EOS when using MDS.
• [C Driver] Use correct type for array length calculation, fix #1539.
• [Driver] Track EOS position per image connection for MDS.
• [C++ Wrapper] Fix maxMessageLength.
• [C++ Wrapper] Fix FragmentAssembler.
• Upgrade to AsciiDoctorJ 2.5.10
• Upgrade to JGit 5.13.2.202306221912-r
• Upgrade to Versions Plugin 0.50.0
• Upgrade to ByteBuddy 1.14.10
• Upgrade to ASM 9.6.
• Upgrade to Agrona 1.20.0
• Upgrade to JUnit 5.10.1
• Upgrade to SBE 1.30.0

• [Java] A leader should assert its leadership when a follower requests a vote and the leader has a more up-to-date log rather than re-initialising an election which could result in log truncation.
• [Java] Change role to leader after winning election so leadership can be asserted during replay or replication.
• [Java] Log event for unknown session close in service container rather than throw and exception.

• [Java] Ensure that the LoggingErrorHandler instance is closed when wrapping a user-defined ErrorHandler.
• [Java] Invoke background work when idle and service is active so external connections from a cluster can be maintained.
• [Doc] JavaDoc clarifications on MediaDriver dirDeleteOnStart behavior.
• Upgrade to Agrona 1.19.2.
• Upgrade to ByteBuddy 1.14.7.

Register here for Aeron Community Meetups - 12 October London / 18 October NY / 16 November Chicago.

• Add port manager to C and Java drivers to allow for a limited range of ports that can be used when subscription endpoint addresses and publication control addresses specify a port of 0. This can make network access configuration simpler (Driver).
• Wait for end of stream event on the log instead of the counter becoming unavailable before entering into a new Election (Cluster).
• Close log publications before trying to stop recording to speed up graceful leader step down (Cluster).
• Ignore resting subscription positions when calculating join position or draining driver entities when tethering is being used (C Driver).
• Extensions points to support Standby Snapshots (Cluster).
• Ensure that the clusterDirectoryName field is synced with the clusterDir configuration parameter during conclude in ConsensusModule and ClusterBacker (Cluster).
• Cluster tutorial updates (Documentation).
• Extract invocation of background work in the service container.
• Do not fail if Aeron directory exists, but the CnC file does not (C Driver).
• Use aeron_errcode to get the latest error code when reporting problems creating the Aeron directory (C Driver).
• Don't let a bounded replay go beyond counter value to stop position so commit position can be used for replay (Archive).
• Add addition example Authentication and Authorisation services (Archive/Cluster).
• Add additional documentation to specify details when Authorisation is used (Archive/Cluster).
• Clarify logic about extending a replay while being bounded (Archive).
• Changes to host name resolution on start up (Driver).
  • Perform host name resolution once and track its execution time via the duty cycle tracker and the event log.
  • Use <unresolved> as a host name if it cannot be resolved instead of null.
  • Fix GCC 9 warning
  • Await initial counters being updated by the conductor thread before reading the values.
• Trigger graceful leader close election based on recording stop signal (Archive).
• Add end of stream position to Subscriptions (Client).
• Allow for relocatable mark files for the ConsensusModule, ClusteredServiceContainer, and ClusterBackup (Cluster).
• Move close operations on MDS transports on the conductor not the receiver thread (C Driver).
• Allow preparing for new leadership to be less serial to reduce election time (Cluster).
• Async removal of destinations as log adaptor closes when preparing for election (Cluster).
• Fix CMD_OUT_ERROR log format (C Driver).
• Fix aeron_err_clear() (C).
• C driver handle untethered subscriptions correctly, ensure that tethering behaviour matches Java driver (C Driver).
• Fix name in toString (Java Client).
• Improve error message (C++ Archive Client).
• Clean up Apple M1 compilation. Move loss reporter function into compilation unit. Add pragmas to account for unused but set variables (C).
• Add logging to track information when a ReplicationSession ends (Archive).
• Add deduplication step for aeron-agent jar.
• Add deduplication step for aeron-all jar.
• Add an annotation processor to generate a version class for a number of the Aeron packages.
• Include the git sha in the version C binaries.
• Extract untetheredWindowLimit method.
• Upgrade to SBE 1.29.0.
• Upgrade to JUnit 5.10.0.
• Upgrade to Agrona 1.19.1.

• [C] Fix potential memory leak when removing publication. Use correct action with publication notification.

Note: This release contains changes for the C media driver only. The Java implementation remains unchanged at version 1.39.0.

• Store Aeron.NULL_VALUE for lastActivityTimestamp in the snapshot for sessions as the value should be transient (Cluster).
• On snapshot of PendingServiceMessageTracker, check to see if the buffer is empty and the nextServiceSessionId is lower than what it should be. Correct this and log an error that follower may not have executed the service IPC logic deterministically (Cluster).
• Include the recording position in the debug log when replicate session state changes (Archive).
• Don't cool down an Image that is EOS (Driver)
• Reset eos_flagged when adding a receiver (C Driver)
• Filter images by session id when adding new network subscriptions (C Driver)
• Allow the session-id of the publication used during replication to be specified to avoid session-id clashes on replication retries (Archive).
• Allow the authorisation credentials to be specified on a replication request to support simple authentication between the destination and source archives (Archive).
• Use custom session-ids when replicating snapshots and logs (Cluster).
• SBE 1.28.3
• Agrona 1.18.2
• ByteBuddy 1.14.5
• Version 0.47.0.

• [Java] Add debug method for a replication completing withing Cluster.
• [Java] Let Archive take care of handling Aeron exceptions when client is embedded.
• [C] Prevent double free of the aeron_udp_channel_t when creating a send_channel_endpoint_t.
• [C] Upgrade to HDR Histogram 0.11.8
• [Java] Upgrade to JUnit 5.9.3 (platform console 1.9.3)
• [Java/C] Name resolution execution time tracking
• [Java] Simplify Client Close
• [Java] Perform Runtime.exit(int) on another thread in default error handler to avoid deadlocks if Aeron instances are used on JVM shutdown hooks.
• [C] Fix potential memory leak.  Use correct action with publication notification.
• [Java] Ensure segment file write of final byte for extension is successful