RELEASED: Public: October 18, 2023
New features
- Display the load per replica summary in the High Availability diagnostic page (BEE-35368)
-
The High Availability diagnostic page now displays a list of replicas and some load information, such as running builds and online agents.
Feature enhancements
- Display queue size in the High Availability (HA) summary (BEE-40738)
-
Now, the queue size on each cluster member is displayed in the High Availability (HA) summary.
- Load-aware build adoption in High Availability (HA) (BEE-40958)
-
Previously, when a High Availability (HA) controller terminated, its builds were adopted at random by other replicas.
Now, the adoption is preferentially directed toward replicas that currently have the least load according to a simple metric.
Resolved issues
-
Internal changes to streamline development. Not user visible. (BEE-35583)(BEE-38770)::
- Retain queue ID when adopting queue items in High Availability (HA)(BEE-33892)
-
If you schedule a build using a High Availability (HA) controller, but it does not start before the replica shuts down, the corresponding adopted queue item in another replica now uses the same ID.
This allows external tools to track the build when it does start (or the item is cancelled).
- Send a warning if the
managed-master-hibernationplugin is installed on a High Availability High Availability (HA)controller (BEE-39349) -
Added an administrative monitor warning to administrators who run an High Availability High Availability (HA) controller that the
managed-master-hibernationplugin should be uninstalled as it is incompatible with it. - Excessive HTTP client threads created (BEE-39352)
-
Certain operations, such as contacting the operations center for user lookup, could create excessive Java threads.
This issue has now been resolved.
- More reliable cross-replica build loading for High Availability (HA) controllers (BEE-39952)
-
Previously, High Availability (HA) controller replicas could partially load build records even when the builds were in progress on other controllers, leading to errors.
Now, a replica only loads a build record when the build is complete or when it is adopting a build in progress.
- Log recorders synchronization does not work for creation and deletion (BEE-40044)
-
The log recorders are now correctly synchronized between replicas of a controller in High Availability (HA) mode.
- Optimized handling of job next build number in HA controllers (BEE-40338)
-
Synchronizing the next build number of jobs can be a bottleneck in HA controllers on slow network filesystems.
This process is now improved and the issue has been resolved.
- Builds are incorrectly adopted from a replica that was in quiet down mode (BEE-40816)
-
If a controller replica was set to quiet down mode and left there for several minutes without exiting, its builds might be adopted by other replicas.
If the original owner was brought out of quiet down mode, or some steps ran in the build on the original replica and printed the log output, the metadata may be corrupted. This issue has been resolved.
- Incompatible HA agent administrative monitor is displayed even if empty (BEE-40907)
-
In some cases, the administrative monitor that displays incompatible High Availability (HA) agents can be displayed, even when an incompatible agent is not found.
This issue has been resolved and the administrative monitor is no longer displayed.