8000 HDFS-16855. Remove the redundant write lock in addBlockPool. by dingshun3016 · Pull Request #5170 · apache/hadoop · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

HDFS-16855. Remove the redundant write lock in addBlockPool. #5170

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

8000

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 1 commit into
base: trunk
Choose a base branch
from
Open

HDFS-16855. Remove the redundant write lock in addBlockPool. #5170

wants to merge 1 commit into from
+9 −11

Conversation

dingshun3016
Copy link
Contributor

When patching the datanode's fine-grained lock, we found that the datanode couldn't start，maybe happened deadlock，when addBlockPool, so we can remove it.

org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl#addBlockPool get writeLock

org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl#deepCopyReplica need readLock

because it is not the same thread, so the write lock cannot be downgraded to a read lock

@hadoop-yetus
Copy link

💔 -1 overall

Vote Subsystem Runtime Logfile Comment
+0 🆗 reexec 0m 50s Docker mode activated.
_ Prechecks _
+1 💚 dupname 0m 0s No case conflicting files found.
+0 🆗 codespell 0m 2s codespell was not available.
+0 🆗 detsecrets 0m 2s detect-secrets was not available.
+1 💚 @author 0m 0s The patch does not contain any @author tags.
-1 ❌ test4tests 0m 0s The patch doesn't appear to include any new or modified tests. Please justify why no new tests are needed for this patch. Also please list what manual steps were performed to verify this patch.
_ trunk Compile Tests _
+1 💚 mvninstall 44m 10s trunk passed
+1 💚 compile 1m 40s trunk passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 compile 1m 34s trunk passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 checkstyle 1m 28s trunk passed
+1 💚 mvnsite 2m 2s trunk passed
+1 💚 javadoc 1m 34s trunk passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javadoc 1m 48s trunk passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 spotbugs 4m 16s trunk passed
+1 💚 shadedclient 28m 43s branch has no errors when building and testing our client artifacts.
_ Patch Compile Tests _
+1 💚 mvninstall 1m 32s the patch passed
+1 💚 compile 1m 34s the patch passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javac 1m 34s the patch passed
+1 💚 compile 1m 19s the patch passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 javac 1m 19s the patch passed
+1 💚 blanks 0m 0s The patch has no blanks issues.
+1 💚 checkstyle 1m 0s the patch passed
+1 💚 mvnsite 1m 24s the patch passed
+1 💚 javadoc 0m 58s the patch passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javadoc 1m 30s the patch passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 spotbugs 3m 34s the patch passed
+1 💚 shadedclient 26m 18s patch has no errors when building and testing our client artifacts.
_ Other Tests _
-1 ❌ unit 388m 9s /patch-unit-hadoop-hdfs-project_hadoop-hdfs.txt hadoop-hdfs in the patch passed.
+1 💚 asflicense 0m 54s The patch does not generate ASF License warnings.
513m 3s
Reason Tests
Failed junit tests hadoop.hdfs.TestLeaseRecovery2
Subsystem Report/Notes
Docker ClientAPI=1.41 ServerAPI=1.41 base: https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/1/artifact/out/Dockerfile
GITHUB PR #5170
Optional Tests dupname asflicense compile javac javadoc mvninstall mvnsite unit shadedclient spotbugs checkstyle codespell detsecrets
uname Linux 0f9749a8f192 4.15.0-191-generic #202-Ubuntu SMP Thu Aug 4 01:49:29 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
Build tool maven
Personality dev-support/bin/hadoop.sh
git revision trunk / 54a0786
Default Java Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
Multi-JDK versions /usr/lib/jvm/java-11-openjdk-amd64:Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04 /usr/lib/jvm/java-8-openjdk-amd64:Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
Test Results https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/1/testReport/
Max. process+thread count 2123 (vs. ulimit of 5500)
modules C: hadoop-hdfs-project/hadoop-hdfs U: hadoop-hdfs-project/hadoop-hdfs
Console output https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/1/console
versions git=2.25.1 maven=3.6.3 spotbugs=4.2.2
Powered by Apache Yetus 0.14.0 https://yetus.apache.org

This message was automatically generated.

@MingXiangLi
Copy link
Contributor
MingXiangLi commented Dec 1, 2022 •
edited
Loading

@dingshun3016 This seems to only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, so why not handle it for example like forbid refresh() when ReplicaCachingGetSpaceUsed#init() at first time ?

@dingshun3016 dingshun3016 reopened this Dec 1, 2022
@dingshun3016
Copy link
Contributor Author

@dingshun3016 This seems to only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, so why not handle it for example like forbid refresh() when ReplicaCachingGetSpaceUsed#init() at first time ?

@MingXiangLi thanks reply

forbid refresh() when ReplicaCachingGetSpaceUsed #init() at first time，it will cause the value of dfsUsage to be 0 until the next time refresh().

if remove the BLOCK_POOl level write lock in the org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl#addBlockPool(String bpid, Configuration conf) method, what will be the impact ?

do you have any other suggestions?

@MingXiangLi
Copy link
Contributor
MingXiangLi commented Dec 1, 2022 •
edited
Loading

The BLOCK_POOl level lock is to protect replica consistency for FsDataSetImpl when read or write operating happend at same time.

forbid refresh() when ReplicaCachingGetSpaceUsed #init() at first time，it will cause the value of dfsUsage to be 0 until the next time refresh().

For example we can use df command instead at first time or other way.

On my side It's less risky to change ReplicaCachingGetSpaceUsed logic than remove the write lock.
Or we can further discussion to make sure no case will lead to consistency problem if we remove write lock.

@dingshun3016
Copy link
Contributor Author

according to the situation discussed so far, it seems that there are several ways to solve this problem

  • remove the BLOCK_POOl level write lock in #addBlockPool

    but worry about having replica consistency problem

  • forbid refresh() when ReplicaCachingGetSpaceUsed#init() at first time

    it will cause the value of dfsUsage to be 0 until the next time refresh()

  • use du or df command instead at first time

    du is very expensive and slow
    df is inaccurate when the disk sharing by other servers
    reference HDFS-14313

Now that, this case only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, I have an idea, is it possible to add a switch, not add lock when ReplicaCachingGetSpaceUsed#init() at first time , and add it at other times

do you think it's possible？@MingXiangLi
8000

@hadoop-yetus
Copy link

💔 -1 overall

Vote Subsystem Runtime Logfile Comment
+0 🆗 reexec 1m 11s Docker mode activated.
_ Prechecks _
+1 💚 dupname 0m 0s No case conflicting files found.
+0 🆗 codespell 0m 0s codespell was not available.
+0 🆗 detsecrets 0m 0s detect-secrets was not available.
+1 💚 @author 0m 0s The patch does not contain any @author tags.
-1 ❌ test4tests 0m 0s The patch doesn't appear to include any new or modified tests. Please justify why no new tests are needed for this patch. Also please list what manual steps were performed to verify this patch.
_ trunk Compile Tests _
+1 💚 mvninstall 42m 16s trunk passed
+1 💚 compile 1m 36s trunk passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 compile 1m 30s trunk passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 checkstyle 1m 17s trunk passed
+1 💚 mvnsite 1m 39s trunk passed
+1 💚 javadoc 1m 18s trunk passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javadoc 1m 41s trunk passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 spotbugs 3m 51s trunk passed
+1 💚 shadedclient 26m 27s branch has no errors when building and testing our client artifacts.
_ Patch Compile Tests _
+1 💚 mvninstall 1m 23s the patch passed
+1 💚 compile 1m 32s the patch passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javac 1m 32s the patch passed
+1 💚 compile 1m 21s the patch passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 javac 1m 21s the patch passed
+1 💚 blanks 0m 0s The patch has no blanks issues.
+1 💚 checkstyle 0m 59s the patch passed
+1 💚 mvnsite 1m 25s the patch passed
+1 💚 javadoc 0m 57s the patch passed with JDK Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04
+1 💚 javadoc 1m 31s the patch passed with JDK Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
+1 💚 spotbugs 3m 30s the patch passed
+1 💚 shadedclient 26m 7s patch has no errors when building and testing our client artifacts.
_ Other Tests _
-1 ❌ unit 415m 49s /patch-unit-hadoop-hdfs-project_hadoop-hdfs.txt hadoop-hdfs in the patch passed.
+1 💚 asflicense 1m 5s The patch does not generate ASF License warnings.
535m 40s
Reason Tests
Failed junit tests hadoop.hdfs.qjournal.server.TestJournalNode
hadoop.hdfs.TestLeaseRecovery2
Subsystem Report/Notes
Docker ClientAPI=1.41 ServerAPI=1.41 base: https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/2/artifact/out/Dockerfile
GITHUB PR #5170
Optional Tests dupname asflicense compile javac javadoc mvninstall mvnsite unit shadedclient spotbugs checkstyle codespell detsecrets
uname Linux c8aeace311b7 4.15.0-191-generic #202-Ubuntu SMP Thu Aug 4 01:49:29 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
Build tool maven
Personality dev-support/bin/hadoop.sh
git revision trunk / 54a0786
Default Java Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
Multi-JDK versions /usr/lib/jvm/java-11-openjdk-amd64:Ubuntu-11.0.16+8-post-Ubuntu-0ubuntu120.04 /usr/lib/jvm/java-8-openjdk-amd64:Private Build-1.8.0_342-8u342-b07-0ubuntu1~20.04-b07
Test Results https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/2/testReport/
Max. process+thread count 1906 (vs. ulimit of 5500)
modules C: hadoop-hdfs-project/hadoop-hdfs U: hadoop-hdfs-project/hadoop-hdfs
Console output https://ci-hadoop.apache.org/job/hadoop-multibranch/job/PR-5170/2/console
versions git=2.25.1 maven=3.6.3 spotbugs=4.2.2
Powered by Apache Yetus 0.14.0 https://yetus.apache.org

This message was automatically generated.

@MingXiangLi
Copy link
Contributor
MingXiangLi commented Dec 2, 2022 •
edited
Loading

Now that, this case only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, I have an idea, is it possible to add a switch, not add lock when ReplicaCachingGetSpaceUsed#init() at first time , and add it at other times

This makes sense to me, get replicas usage message no need strong consistency.@Hexiaoqiao any suggestion?

@Hexiaoqiao
Copy link
Contributor

Now that, this case only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, I have an idea, is it possible to add a switch, not add lock when ReplicaCachingGetSpaceUsed#init() at first time , and add it at other times

This makes sense to me, get replicas usage message no need strong consistency.@Hexiaoqiao any suggestion?

Thanks for the detailed discussions. +1. it seems good to me.
BTW, I try to dig PR to fix this bug but no found. It just at out internal branch which not refresh space used at init stage. And refresh-used is one complete async thread (at CachingGetSpaceUsed) , thus it could not dead lock when DataNode instance restart. Thanks.

@dingshun3016
Copy link
Contributor Author

Now that, this case only happen when invoke addBlockPool() and CachingGetSpaceUsed#used < 0, I have an idea, is it possible to add a switch, not add lock when ReplicaCachingGetSpaceUsed#init() at first time , and add it at other times

This makes sense to me, get replicas usage message no need strong consistency.@Hexiaoqiao any suggestion?

Thanks for the detailed discussions. +1. it seems good to me. BTW, I try to dig PR to fix this bug but no found. It just at out internal branch which not refresh space used at init stage. And refresh-used is one complete async thread (at CachingGetSpaceUsed) , thus it could not dead lock when DataNode instance restart. Thanks.

Thanks replay. looks like this PR HDFS-14986, forbid refresh() when ReplicaCachingGetSpaceUsed#init() at first time.
I prefer to not add lock when ReplicaCachingGetSpaceUsed#init() at first time

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@dingshun3016 @hadoop-yetus @MingXiangLi @Hexiaoqiao
< 3171 /div>
0