Akka Resilience Enhancements Release Notes

Version 1.1.14

Add sensitive config paths for cassandra, jdbc and couchbase Persistence plugins
Update link to credentials for dependencies

Added support for running under Scala 2.13.0
Improved documentation for Kubernetes Lease
Fixed false positive in config checker about “akka.actor.default-dispatcher.fork-join-executor.task-peeking-mode”

Kubernetes Lease now based on Akka Coordination Lease Compatible with Akka Version 2.5.22 and up.

New SBR strategy based on distributed lease (lock) implemented with Kubernetes API.
Handle edge cases in Keep Majority and Keep Oldest SBR strategies when there are membership changes at the same time as a network partition.
Static Quorum SBR strategy will down all nodes if number of members exceeds the quorum-size * 2 - 1 limit when a SBR decision is needed.
Enable Down all when unstable by default.
Fix issue in Config Checker when SBR dependency not on classpath.
Better version reporting and remove some false warnings in Diagnostics Recorder.
Fix issue in Starvation Detector when there are no threads to report.
Updated dependencies to Akka 2.5.21 (improved downing).

Updated dependencies to Akka 2.5.19 and akka-persistence-cassandra 0.92, which also updates the transitive dependency to Guava (older version has security vulnerability)

New SBR down-all strategy. This strategy can be a safe alternative if the network environment is highly unstable with unreachability observations that can’t be fully trusted, and including frequent occurrences of indirectly connected nodes. In such environments it can be better to shutdown all nodes and start up a new fresh cluster. See Down All documentation.
New SBR configuration property down-all-when-unstable to down all nodes if failure detector observations continue to change for too long. Down all when unstable documentation.
New SBR configuration property keep-one-indirectly-connected to allow downing of all indirectly connected nodes instead of keeping one, which can be a safer option in unstable environments. See Indirectly connected nodes documentation.
Highlight and clarify the scenarios that can lead to SBR decisions that result in forming separate clusters, i.e. the system will experience a split brain scenario.

New SBR algorithm to handle scenarios where nodes are marked as unreachable via some network links but they are still indirectly connected via other nodes, i.e. it’s not a clean network partition (or node crash). See Indirectly connected nodes documentation.

Fix issue that SBR might not be active on nodes with status WeaklyUp
More SBR logging
Clarify SBR stable-after and down-removal-margin
Additional hints by Thread Starvation Detector what can be the cause of the problems