[CELEBORN-2330] Fix HA master bootstrap redirect handling#3691
Draft
sunchao wants to merge 5 commits into
Draft
Conversation
sunchao
changed the title
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why are the changes needed?
In HA mode, clients can hit a failover window where the master they contact is no longer the leader and returns a
MasterNotLeaderExceptionwith a suggested leader address. The client-facing symptom is an RPC failure surfaced as:CelebornException: Exception thrown in awaitResultThe redirect signal is still present underneath that wrapper, but the existing bootstrap and retry logic does not consistently preserve and follow it. In particular:
The consequence is that a client can fail to establish a master connection during HA leader transitions even when a reachable leader or another configured master is available. That turns a recoverable redirect/failover event into an avoidable client-visible failure and makes rolling upgrades or leader changes noisier than necessary.
What changes were proposed in this PR?
This port brings the HA redirect handling fix from
openai/celeborn#70onto upstreammain.When a master tells the client which leader to use, the client now keeps that redirect information even if it is wrapped inside another exception, and it actively follows the suggested leader during bootstrap and failover. If that suggested leader points to another leader, the client can continue along that redirect chain instead of giving up too early.
If the redirect path is no longer useful, for example because the suggested leader cannot be reached, no leader is currently presented, or redirects start looping, the client falls back to the remaining configured masters and keeps searching for a usable endpoint. The retry scan also advances correctly after endpoint setup fails, so one bad redirect does not prevent the client from trying the next viable master.
The PR also adds focused HA tests that cover the bootstrap redirect cases, chained redirects, redirect cycles, and fallback to configured masters.
How was this PR tested?
build/mvn test -pl common -am -Dtest=MasterClientSuiteJ -DwildcardSuites=org.apache.celeborn.common.client.__NoSuchSuite__