HDFS-17514: RBF: Routers should unset cached stateID when namenode does not set stateID in RPC response header. #6804
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
…havior of old filesystem object.
|
💔 -1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
mccormickt12
reviewed
May 8, 2024
| @Override | ||
| public void receiveResponseState(RpcHeaderProtos.RpcResponseHeaderProto header) { | ||
| sharedGlobalStateId.accumulate(header.getStateId()); | ||
| if (header.getStateId() == 0 && sharedGlobalStateId.get() > 0) { |
Contributor
There was a problem hiding this comment.
stateId 0 means no state id right?
This is different than msync being 0?
Contributor
There was a problem hiding this comment.
I assume stateId is an integer and protobuf will return 0 for an integer if it is not set?
Member
Author
There was a problem hiding this comment.
Yes, stateID 0 means no stateId is set. Msync doesn't have a return value.
There is another bug in the router in that it accepts 0 as a value to advance it's cachedStateID to.
Ideally sharedGlobalStateId.get() > 0 should not be necessary here. For now it captures namenodes that actually had STATE_ID_CONTEXT enabled to begin with. But stale reads could happen with a namenode that has never had STATE_ID_CONTEXT enabled.
Fixing this will touch other tests so I'm debating whether to try fix that in this PR or separately. I'm leaning towards a separate PR.
Contributor
There was a problem hiding this comment.
We can remove the sharedGlobalStateId.get() > 0, right?
If sharedGlobalStateId.get() < 0, routers already fallback to active and no need to reset. If it is > 0 and then we see a request without StateID, we will reset this counter and routers will fallback to active.
Adding sharedGlobalStateId.get() > 0 doesn't seem to make a difference.
Member
Author
There was a problem hiding this comment.
The tests in TestNoNamenodesAvailableLongTime rely on the router allowing a stateId of 0. So having sharedGlobalStateId.get() > 0 allows this behavior while guarding against when the sharedGlobalStateId has advances beyond zero.
The tests in TestNoNamenodesAvailableLongTime do need to be fixed but I would like to limit the scope of this PR.
Contributor
There was a problem hiding this comment.
+1 to fixing the tests and associated check in follow on PR.
xinglin
approved these changes
May 8, 2024
Contributor
xinglin
left a comment
xinglin
left a comment
There was a problem hiding this comment.
thanks for the quick fix!
|
🎊 +1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
xinglin
approved these changes
May 10, 2024
ctrezzo
reviewed
May 10, 2024
| if (header.getStateId() == 0 && sharedGlobalStateId.get() > 0) { | ||
| sharedGlobalStateId.reset(); | ||
| } else { | ||
| sharedGlobalStateId.accumulate(header.getStateId()); |
Contributor
There was a problem hiding this comment.
@simbadzina I have a naive question: What protects us here from the state where header.getStateId() > 0 && header.getStateId() < sharedGlobalStateId?
It seems like if this case were to occur then sharedGlobalStateId would go backwards.
Member
Author
There was a problem hiding this comment.
The sharedGlobalStateID is created as new LongAccumulator(Math::max, Long.MIN_VALUE)
So accumulate either keeps the current value or moves it forward.
ctrezzo
reviewed
May 10, 2024
Contributor
There was a problem hiding this comment.
@simbadzina Should we also add a test case to TestPoolAlignmentContext.java to ensure the sharedGlobalStateId moves under the conditions we expect it to?
|
🎊 +1 overall
This message was automatically generated. |
Member
Author
Good call. Let me add that. |
|
💔 -1 overall
This message was automatically generated. |
ctrezzo
approved these changes
May 14, 2024
Contributor
ctrezzo
left a comment
ctrezzo
left a comment
There was a problem hiding this comment.
+1 LGTM. Thanks @simbadzina !
|
🎊 +1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
Member
Author
|
Failing tests in continuous-integration unrelated to my changes |
K0K0V0K
pushed a commit
to K0K0V0K/hadoop
that referenced
this pull request
May 17, 2024
…es not set stateID in RPC response header. (apache#6804)
K0K0V0K
pushed a commit
to K0K0V0K/hadoop
that referenced
this pull request
May 17, 2024
…es not set stateID in RPC response header. (apache#6804)
NyteKnight
pushed a commit
to NyteKnight/hadoop
that referenced
this pull request
Jun 25, 2024
…es not set stateID in RPC response header. (apache#6804) (Cherry-picked from 6a4f0be)
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.
HDFS-17514
Description of PR
When a namenode that had "dfs.namenode.state.context.enabled" set to true is restarted with the configuration set to false, routers will keep using a previously cached state ID.
Without RBF
With RBF
New clients that are created after the restart should not fetch the stale state ID.
How was this patch tested?
New unit test
Unit test fails with the patch is PoolAlignmentContext in the second commit.
Unit test pass with the patch.
For code changes:
LICENSE,LICENSE-binary,NOTICE-binaryfiles?