[fix][bk] Fix NPE in IsolatedBookieEnsemblePlacementPolicy when policy class does not match

[Shawyeok]

Fixes NPE in IsolatedBookieEnsemblePlacementPolicy

Motivation

A NullPointerException occurs in getExcludedBookiesWithIsolationGroups when getIsolationGroup() returns a Pair whose left value is null. This happens when the EnsemblePlacementPolicyConfig carries a policy class name that does not equal IsolatedBookieEnsemblePlacementPolicy — in that case the if block that calls pair.setLeft() / pair.setRight() is skipped entirely, leaving both sides of the MutablePair as null.

This bug was observed during the upgrade from the 2.8 branch to the 3.0 branch. The same defect may also exist in master.

Full stack trace from production:

2026-05-18T13:28:40.756Z [BookKeeperClientWorker-OrderedExecutor-35-0] ERROR org.apache.bookkeeper.common.util.SingleThreadExecutor - Error while running task: Cannot invoke "java.util.Set.contains(Object)" because the return value of "org.apache.commons.lang3.tuple.Pair.getLeft()" is null
java.lang.NullPointerException: Cannot invoke "java.util.Set.contains(Object)" because the return value of "org.apache.commons.lang3.tuple.Pair.getLeft()" is null
    at org.apache.pulsar.bookie.rackawareness.IsolatedBookieEnsemblePlacementPolicy.getExcludedBookiesWithIsolationGroups(IsolatedBookieEnsemblePlacementPolicy.java:192)
    at org.apache.pulsar.bookie.rackawareness.IsolatedBookieEnsemblePlacementPolicy.getExcludedBookies(IsolatedBookieEnsemblePlacementPolicy.java:141)
    at org.apache.pulsar.bookie.rackawareness.IsolatedBookieEnsemblePlacementPolicy.replaceBookie(IsolatedBookieEnsemblePlacementPolicy.java:127)
    at org.apache.bookkeeper.client.BookieWatcherImpl.replaceBookie(BookieWatcherImpl.java:316)
    at org.apache.bookkeeper.client.EnsembleUtils.replaceBookiesInEnsemble(EnsembleUtils.java:69)
    at org.apache.bookkeeper.client.ReadOnlyLedgerHandle.handleBookieFailure(ReadOnlyLedgerHandle.java:222)
    at org.apache.bookkeeper.client.PendingAddOp.writeComplete(PendingAddOp.java:353)
    at org.apache.bookkeeper.proto.BookieClientImpl.completeAdd(BookieClientImpl.java:287)
    at org.apache.bookkeeper.proto.BookieClientImpl.access$200(BookieClientImpl.java:79)
    at org.apache.bookkeeper.proto.BookieClientImpl$ChannelReadyForAddEntryCallback.lambda$operationComplete$0(BookieClientImpl.java:405)
    at org.apache.bookkeeper.common.util.OrderedExecutor$TimedRunnable.run(OrderedExecutor.java:203)
    at org.apache.bookkeeper.common.util.SingleThreadExecutor.safeRunTask(SingleThreadExecutor.java:137)
    at org.apache.bookkeeper.common.util.SingleThreadExecutor.run(SingleThreadExecutor.java:107)
    at io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
    at java.base/java.lang.Thread.run(Thread.java:840)

This error causes topics to time out during loading. Since there is no automatic recovery path, the only way to restore service is by restarting the broker.

Modifications

Two complementary fixes:

1. getIsolationGroup(): Ensure not returning pair of null: add compatibility with ZkIsolatedBookieEnsemblePlacementPolicy and fallback to defaultIsolationGroups if met a mis-match policy class.

Verifying this change

• Make sure that the change passes the CI checks.

This change added tests and can be verified as follows:

• Added testReplaceBookieWithNonMatchingPolicyClassShouldNotThrowNPE to reproduce the exact production failure: calling replaceBookie with custom metadata whose EnsemblePlacementPolicyConfig policy class does not match IsolatedBookieEnsemblePlacementPolicy, which previously caused the NPE.

Does this pull request potentially affect one of the following parts:

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment

[lhotari]

LGTM

[void-ptr974]

LGTM, thanks for addressing the fallback behavior.

[Shawyeok]

/pulsarbot rerun-failure-checks

[Shawyeok]

@lhotari @dao-jun
All tests have passed, feel free to take a look, thanks.