[YARN-3933] FairScheduler: Multiple calls to completedContainer are not safe

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.7.1
• Fix Version/s: 2.8.0, 3.0.0-alpha4
• Component/s: fairscheduler
• Labels: oct16-medium
• Hadoop Flags: Reviewed
• Assignee: Shiwei Guo
• Reporter: Lavkesh Lahngir
• Created: 17/Jul/15 05:55
• Updated: 12/Oct/23 04:33
• Resolved: 13/Feb/17 19:43

Description

In our cluster we are seeing available memory and cores being negative.
Initial inspection:
Scenario no. 1:
In capacity scheduler the method allocateContainersToNode() checks if
there are excess reservation of containers for an application, and they are no longer needed then it calls queue.completedContainer() which causes resources being negative. And they were never assigned in the first place.

I am still looking through the code. Can somebody suggest how to simulate excess containers assignments ?

Issue Links

• is duplicated by: YARN-4045 Negative avaialbleMB is being reported for root queue.
• is duplicated by: YARN-4067 available resource could be set negative
• is duplicated by: YARN-4089 Race condition when calling AbstractYarnScheduler.completedContainer.
• is duplicated by: YARN-6158 FairScheduler: app usage can go to negative
• is duplicated by: YARN-6212 NodeManager metrics returning wrong negative values
• is duplicated by: YARN-10317 RM returns a negative value when TEZ AM requests resources
• is duplicated by: YARN-4698 Negative value in RM UI counters due to double container release
• is related to: YARN-4809 De-duplicate container completion across schedulers
• relates to: YARN-4089 Race condition when calling AbstractYarnScheduler.completedContainer.