[SPARK-19755] Blacklist is always active for MesosCoarseGrainedSchedulerBackend. As result - scheduler cannot create an executor after some time.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Affects Version/s: 2.1.0
• Component/s: Mesos, Scheduler, Spark Core
• Labels: bulk-closed

Environment

mesos, marathon, docker - driver and executors are dockerized.

• Assignee: Unassigned
• Reporter: Timur Abakumov
• Created: 27/Feb/17 21:30
• Updated: 25/May/21 01:52
• Resolved: 25/May/21 01:42

Description

When for some reason task fails - MesosCoarseGrainedSchedulerBackend increased failure counter for a slave where that task was running.
When counter is >=2 (MAX_SLAVE_FAILURES) mesos slave is excluded.
Over time scheduler cannot create a new executor - every slave is is in the blacklist. Task failure not necessary related to host health- especially for long running stream apps.
If accepted as a bug: possible solution is to use: spark.blacklist.enabled to make that functionality optional and if it make sense MAX_SLAVE_FAILURES also can be configurable.

Issue Links

• causes: SPARK-24567 nodeBlacklist does not get updated if a spark executor fails to launch on a mesos node
• is duplicated by: SPARK-23423 Application declines any offers when killed+active executors rich spark.dynamicAllocation.maxExecutors
• is related to: SPARK-16630 Blacklist a node if executors won't launch on it.
• relates to: SPARK-23485 Kubernetes should support node blacklist
• links to: [Github] Pull Request #17619 (timout)
• links to: [Github] Pull Request #20640 (IgorBerman)
• links to: GitHub Pull Request #17619
• links to: GitHub Pull Request #20640