[fix] Restore interrupt status in InterruptedException catch blocks

[TimurRakhmatullin86]

Motivation

When Java code catches InterruptedException, the JVM automatically clears the interrupt flag on the current thread. If the catch block does not restore the flag by calling Thread.currentThread().interrupt(), all upstream callers lose visibility into the fact that an interrupt was requested. This is a well-known correctness issue documented in Java Concurrency in Practice (section 7.1.2) and the InterruptedException Javadoc.

In a system like Pulsar — where graceful shutdown, topic unloading, and thread pool termination rely on cooperative interrupt signaling — swallowing the interrupt flag can cause:

• Shutdown hangs: threads that should terminate keep running because they never see the interrupt.
• Resource leaks: cleanup code gated on Thread.interrupted() is silently skipped.
• Silent failures: operations that should abort continue with stale or inconsistent state.

Modifications

This PR adds Thread.currentThread().interrupt() to 35 catch blocks across 31 files in the following modules:

Module	Files
pulsar-broker (service)	BrokerService, PersistentTopic, PersistentSubscription, NonPersistentTopic
pulsar-broker (loadbalance)	ServiceUnitStateChannelImpl, TransferShedder, BrokerRegistryImpl
pulsar-broker (namespace)	OwnedBundle, OwnershipCache
pulsar-broker (admin/web)	PulsarWebResource, WebService, PackagesBase, NonPersistentTopics, LoadReportCommand
pulsar-broker (compaction)	StrategicTwoPhaseCompactor
pulsar-broker (other)	BookKeeperClientFactoryImpl, PulsarClusterMetadataTeardown, LocalBookkeeperEnsemble
pulsar-broker-common	AuthorizationService, OneStageAuthenticationState, BookieRackAffinityMapping
pulsar-common	FileUtils, FutureUtil
managed-ledger	ManagedLedgerImpl
pulsar-metadata	PulsarRegistrationManager, PulsarLedgerUnderreplicationManager, PulsarLedgerAuditorManager, ZKSessionWatcher, MetadataStoreTableViewImpl
pulsar-proxy	ProxyStats
pulsar-transaction	MLTransactionLogImpl

For catch blocks that handle only InterruptedException, the fix is:

} catch (InterruptedException e) {
    Thread.currentThread().interrupt();
    // existing handling ...
}

For multi-catch blocks (e.g. catch (InterruptedException | ExecutionException e)), the fix is:

} catch (InterruptedException | ExecutionException e) {
    if (e instanceof InterruptedException) {
        Thread.currentThread().interrupt();
    }
    // existing handling ...
}

Verifying this change

This is a mechanical, behavior-preserving fix. Each change adds exactly one Thread.currentThread().interrupt() call at the start of a catch block; no control flow or logic is altered.

• Existing unit and integration tests continue to pass.
• The fix follows the pattern already used in many other places in the Pulsar codebase (e.g., PulsarService.java, ExtensibleLoadManagerImpl.java, TopicTransactionBuffer.java).

Documentation

• doc-not-needed — No documentation changes needed for a bug fix that preserves existing behavior.

🤖 Generated with Claude Code