Skip to content

[HUDI-1982] Remove unnecessary synchronization #3041

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.

Already on GitHub? Sign in to your account

Closed
chaplinthink wants to merge 1 commit into from
Closed

[HUDI-1982] Remove unnecessary synchronization #3041

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 7 additions & 9 deletions hudi-timeline-service/src/main/java/org/apache/hudi/timeline/service/RequestHandler.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,15 +132,13 @@ private boolean syncIfLocalViewBehind(Context ctx) {
String lastKnownInstantFromClient =
ctx.queryParam(RemoteHoodieTableFileSystemView.LAST_INSTANT_TS, HoodieTimeline.INVALID_INSTANT_TS);
SyncableFileSystemView view = viewManager.getFileSystemView(basePath);
synchronized (view) {
Copy link
Contributor

@leesf leesf Jun 7, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hi, @chaplinthink I think it is necessary to use synchronization to sync view locally since the handler would handle different request from clients concurrently. cc @bvaradar

Copy link
Contributor Author

@chaplinthink chaplinthink Jun 8, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean the implementation of view.sync(); already has WriteLock to handle multiple requests from clients concurrently @leesf

 try {
      writeLock.lock();
      runSync(oldTimeline, newTimeline);
    } finally {
      writeLock.unlock();
    }

Copy link
Contributor

@leesf leesf Jun 8, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@chaplinthink Thanks for the explanation, make sense to me. @vinothchandar @bvaradar do you have any other concern?

Copy link
Member

@vinothchandar vinothchandar Jul 8, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was mulling about the reloading of timeline that happens before the write lock.

@Override
  public void sync() {
    HoodieTimeline oldTimeline = getTimeline();
    HoodieTimeline newTimeline = metaClient.reloadActiveTimeline().filterCompletedAndCompactionInstants();
    try {
      writeLock.lock();
      runSync(oldTimeline, newTimeline);
    } finally {
      writeLock.unlock();
    }
  }

runSync() actually could init/reassign metaClient, so in theory removing synchornized could in theory make it non-serializable.

I would suggest that we either move the timeline reload into the write lock and leave this as-is. Whatever we change, we need to validate with more concurrent testing. So not sure if this is all worth the trouble.

Are you hitting real concurrency bottlenecks around this?

Copy link
Contributor Author

@chaplinthink chaplinthink Jul 9, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks a lot for reply. Do you mean the synchornized is to ensure HoodieTimeline newTimeline = metaClient.reloadActiveTimeline().filterCompletedAndCompactionInstants(); concurrently?

In fact, we are also to do this right ?

@Override
  public void sync() {
    HoodieTimeline oldTimeline = getTimeline();
    try {
      writeLock.lock();
      HoodieTimeline newTimeline = metaClient.reloadActiveTimeline().filterCompletedAndCompactionInstants();
      runSync(oldTimeline, newTimeline);
    } finally {
      writeLock.unlock();
    }
  }

I am confused when I see the code that we use synchornized and writeLock at the same time.

I agree to validate this with more concurrent testing. Currently i have not encountered concurrency bottlenecks.

if (isLocalViewBehind(ctx)) {
HoodieTimeline localTimeline = viewManager.getFileSystemView(basePath).getTimeline();
LOG.info("Syncing view as client passed last known instant " + lastKnownInstantFromClient
+ " as last known instant but server has the following last instant on timeline :"
+ localTimeline.lastInstant());
view.sync();
return true;
}
if (isLocalViewBehind(ctx)) {
HoodieTimeline localTimeline = viewManager.getFileSystemView(basePath).getTimeline();
LOG.info("Syncing view as client passed last known instant " + lastKnownInstantFromClient
+ " as last known instant but server has the following last instant on timeline :"
+ localTimeline.lastInstant());
view.sync();
return true;
}
}
return false;
Expand Down