JobMaster occurs memory leak problems when running too many distributedLoad job #18635
Labels
type-bug
This issue is about a bug
Comments
|
After verifying on the test environment(alluxio.job.master.job.trace.retention.time=5m),the memory usage of mInfoMao is reduced |
alluxio-bot
pushed a commit
that referenced
this issue
Jul 8, 2024
fix:fix job-master leak memory when submitting a large number of distributed jobs(DIST_LOAD/DIST_CP/Persist jobs)
### What changes are proposed in this pull request?
Start a periodic thread to clear expired jobs information that cannot be trace by the client in CmdJobTracker.The default retention time is 1day,which is the same configuration as LoadV2.
### Why are the changes needed?
When many jobs are submitted,the job master finally will have an oom problem, we can find that the cmdJobTracker retains the residual job information and not cleaned regularly, resulting in memory leaks.
### Does this PR introduce any user facing changes?
Please list the user-facing changes introduced by your change, including
1.add Configuration:
alluxio.job.master.job.trace.retention.time=xx,the default value is 1d.
Related issue: #18635
pr-link: #18639
change-id: cid-d4e5853a1818a22c8a0411a27bfe1141c6f24ebd
alluxio-bot
pushed a commit
that referenced
this issue
Jul 8, 2024
fix:fix job-master leak memory when submitting a large number of distributed jobs(DIST_LOAD/DIST_CP/Persist jobs)
### What changes are proposed in this pull request?
Start a periodic thread to clear expired jobs information that cannot be trace by the client in CmdJobTracker.The default retention time is 1day,which is the same configuration as LoadV2.
### Why are the changes needed?
When many jobs are submitted,the job master finally will have an oom problem, we can find that the cmdJobTracker retains the residual job information and not cleaned regularly, resulting in memory leaks.
### Does this PR introduce any user facing changes?
Please list the user-facing changes introduced by your change, including
1.add Configuration:
alluxio.job.master.job.trace.retention.time=xx,the default value is 1d.
Related issue: #18635
pr-link: #18639
change-id: cid-d4e5853a1818a22c8a0411a27bfe1141c6f24ebd
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Alluxio Version:
2.9.3
Describe the bug
After submitting a large number of distributedLoad jobs in production environment, job master has a memory leak problem and finally cause OOM.
To Reproduce
1.Set up one alluxio cluster,1 master, 3 workers.
2.Mock a large number of small files in underFileSystem
3.Submit a large number of distributedLoad jobs.Notice:Take the batchsize=1 as the loading args.
4.Observe the memory changes and gc in JobMaster.
Expected behavior
The memory size continues to increase until the maximum memory size is reached,finally causing the OOM problem.
Urgency
yes
Are you planning to fix it
yes
Additional context
The cause of this bug is that the residual job information in mInfoMap is not deleted.
The text was updated successfully, but these errors were encountered: