HDFS-16432. Namenode block report add yield to avoid holding write lo… #3907
+91
−55
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
💔 -1 overall
This message was automatically generated. |
liubingxing
force-pushed
the
HDFS-16432
branch
from
0561a74 to
09824a5
Compare
|
💔 -1 overall
This message was automatically generated. |
|
@tasanuma @Hexiaoqiao Please take a look. |
liubingxing
force-pushed
the
HDFS-16432
branch
from
09824a5 to
2367054
Compare
|
💔 -1 overall
This message was automatically generated. |
|
Thanks for opening a PR. As for the block report taking time -- is it still a problem after removing the folded tree set structure (HDFS-13671)? |
|
Thanks @jojochuang for your comment. The block report taking time we reported in this PR was obtained after merging HDFS-13671. I think that the block report time will become longer after merging HDFS-13671 because FoldedTreeSet have a better performance in dealing FBR. |
|
I am also concerned about yielding the lock in the middle of processing a storage. I would be interested to understand where the time is spend doing the block report processing inside the namenode, to see if we can improve the performance of the code and avoid yielding the lock. Have you tried reproducing this problem on a small otherwise idle cluster and attaching the async profiler to the namenode process to see where the code is taking time during block reporting? I think that analysis would be valuable before we change the locking, as there may be something which can be improved around the performance. |
|
Hi @sodonnel, I have tried reproducing this problem on a test cluster and the cpu profile of block report process look like as follow. It looks like most of the time were taking by the #processReportedBlock and #moveBlockToHead, and the #moveBlockToHead is necessary after removing the folded tree set structure. |
|
@liubingxing Could you attach the flame chart svg file to the Jira so I can look at it in more detail please? |
|
@sodonnel I attached the flame chart svg file to the Jira. |
|
@liubingxing sorry to interrupt, Have you ever tried this patch in your production clusters? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
JIRA: HDFS-16432
In our cluster, namenode block report will held write lock for a long time if the storage block number more than 100000. So we want to add a yield mechanism in block reporting process to avoid holding write lock too long.