-
Notifications
You must be signed in to change notification settings - Fork 424
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.
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
RATIS-2235. Allow only one thread to perform appendLog #1206
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@SzyWilliam , thanks for working this! Please see a comment inlined.
final CompletableFuture<Void> appendLog = appendLogFuture.updateAndGet(f -> f.thenCompose(ignored -> { | ||
final List<CompletableFuture<Long>> futures = entries.isEmpty() ? Collections.emptyList() | ||
: state.getLog().append(requestRef.delegate(entries)); | ||
return JavaUtils.allOf(futures); | ||
})); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should retain the ref first, i.e.
@@ -1641,9 +1645,8 @@ class RaftServerImpl implements RaftServer.Division,
state.updateConfiguration(entries);
}
future.join();
-
- final List<CompletableFuture<Long>> futures = entries.isEmpty() ? Collections.emptyList()
- : state.getLog().append(requestRef.delegate(entries));
+ final CompletableFuture<Void> appendLog = entries.isEmpty()? appendLogFuture.get()
+ : appendLog(requestRef.delegate(entries));
proto.getCommitInfosList().forEach(commitInfoCache::update);
CodeInjectionForTesting.execute(LOG_SYNC, getId(), null);
@@ -1657,7 +1660,7 @@ class RaftServerImpl implements RaftServer.Division,
final long commitIndex = effectiveCommitIndex(proto.getLeaderCommit(), previous, entries.size());
final long matchIndex = isHeartbeat? RaftLog.INVALID_LOG_INDEX: entries.get(entries.size() - 1).getIndex();
- return JavaUtils.allOf(futures).whenCompleteAsync((r, t) -> {
+ return appendLog.whenCompleteAsync((r, t) -> {
followerState.ifPresent(fs -> fs.updateLastRpcTime(FollowerState.UpdateType.APPEND_COMPLETE));
timer.stop();
}, getServerExecutor()).thenApply(v -> {
@@ -1675,6 +1678,13 @@ class RaftServerImpl implements RaftServer.Division,
});
}
+ private CompletableFuture<Void> appendLog(ReferenceCountedObject<List<LogEntryProto>> entriesRef) {
+ entriesRef.retain();
+ return appendLogFuture.updateAndGet(f -> f.thenCompose(
+ ignored -> JavaUtils.allOf(state.getLog().append(entriesRef))))
+ .whenComplete((v, e) -> entriesRef.release());
+ }
+
private long checkInconsistentAppendEntries(TermIndex previous, List<LogEntryProto> entries) {
// Check if a snapshot installation through state machine is in progress.
final long installSnapshot = snapshotInstallationHandler.getInProgressInstallSnapshotIndex();
@szetszwo Thanks for reviewing this pr and help with the code. Now I learned how to use zero-copy in async programming, thank you! |
The test failures seem related. Not yet sure what is the problem. |
@@ -1641,9 +1645,9 @@ leaderId, getMemberId(), currentTerm, followerCommit, inconsistencyReplyNextInde | |||
state.updateConfiguration(entries); | |||
} | |||
future.join(); | |||
final CompletableFuture<Void> appendLog = entries.isEmpty()? appendLogFuture.get() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@SzyWilliam , I see the problem now -- when it is empty, it should not wait for the previous append. Otherwise, it blocks the heartbeats.
final CompletableFuture<Void> appendLog = entries.isEmpty()? CompletableFuture.completedFuture(null)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That is a tricky one. Thanks for spotting and troubleshooting this!
@szetszwo the tests are all passed now, PTAL ~ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
+1 the change looks good.
@szetszwo Thanks for reviewing and merging this PR! |
See https://issues.apache.org/jira/browse/RATIS-2235
See the discussion at https://issues.apache.org/jira/browse/RATIS-2208