Releases: AliceO2Group/Control
v0.68.1
v0.68.0
This release includes cgroups support for CPU and memory limits, a new feature that requires a Mesos agent configuration change and ensures a misbehaving task won't block the whole FLP.
- Resource limits:
- [core] Support Mesos task resource limits specification for task classes
- [core] Prevent crash in incomplete limits
- [core] Print limits to IL
- [core] Avoid triggering dead or inactive hooks on teardown
- [core] Proceed with task kill even if some cannot be killed
- [core] Explicit handling of executor/agent failed events
- [core] Only perform a STOP transition for ACTIVE tasks
- [core] Wait for 500ms for ERROR states to settle before GO_ERROR/STOP
v0.67.2
v0.67.1
v0.67.0
This release includes support for internal task error events being raised by tasks. Such an event immediately transitions the environment to the ERROR state.
- Task error events:
- [core] React to TASK_INTERNAL_ERROR with STOP_ACTIVITY attempt
- [core] Build TaskInternalError event
- [executor] Support TASK_INTERNAL_ERROR event
- [occ] Push TASK_INTERNAL_ERROR event
- [occ] Only emit task internal error event once
v0.66.0
This release includes crash fixes and improvements to integration plugins.
-
Integration:
- [apricot] PEDESTALS is now PEDESTAL
- [core] Add pdp_epn_shm_sizes param to ODC plugin
-
Bug fixes:
- [core] Prevent rare crash in resourceOffers tasksDeployed access
- [core] Bail early if a critical error occurs in a transition step
- [core] Prevent crash in Kafka plugin concurrent map access
v0.65.1
v0.65.0
This release brings important changes to the behavior of the scheduler component of the AliECS core. Specifically, a new UNDEPLOYABLE
status has been added for tasks that cannot be deployed due to cluster conditions, and the task scheduling algorithm has been reworked to fail early when possible.
-
AliECS core scheduler improvements:
- [core] Treat undeployable task separately from plain inactive
- [core] Complete product operation for UNDEPLOYABLE status
- [core] Incoming offers preprocessing for early failure
-
Miscellaneous:
- [executor] Log end of life
v0.64.2
The present release includes bug fixes for DCS integration, for the executor and for a race in the repository access layer.
- [core] Do not write back to varSpecMap coming from repos backend
- [core] Update run types enum in Bookkeeping client
- [core] Regenerate DCS protofile (AGD detector)
- [core] Inform user if offer includes multiple executors
- [executor] Do not cause executor disconnect on unprocessable MESSAGE