[fix][broker] Fix ownership-generation races in OwnershipCache remove… by SongOf · Pull Request #1 · SongOf/pulsar

SongOf · 2026-07-16T06:18:14Z

…Ownership and lock-expiry cleanup

Fixes #xyz

Main Issue: #xyz

PIP: #xyz

Motivation

Modifications

Verifying this change

Make sure that the change passes the CI checks.

(Please pick either of the following options)

This change is a trivial rework / code cleanup without any test coverage.

(or)

This change is already covered by existing tests, such as (please describe tests).

(or)

This change added tests and can be verified as follows:

(example:)

Added integration tests for end-to-end deployment with large payloads (10MB)
Extended integration test for recovery after broker failure

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

If the box was checked, please highlight the changes

…Ownership and lock-expiry cleanup

# Conflicts: # pulsar-broker/src/main/java/org/apache/pulsar/broker/service/BrokerService.java # pulsar-broker/src/test/java/org/apache/pulsar/broker/service/BrokerServiceTest.java

…d TLS material provider plugin interface (apache#25890)

… orderly consumer scaling (apache#26208)

…p-client-api SPI modules (apache#26222)

…er (apache#26269)

…cy (apache#26251)

…to-creation is disabled (apache#26279)

…pache#26271)

… are no context attributes (apache#26285)

…lsar-txn-consumable for peek messages (apache#26073)

…futures (apache#26287)

…pache#26300)

…che#26293)

…aned delayed-delivery buckets (apache#26309)

…acementPolicyLedgerFragmentsOnRack (apache#26308)

…r instead of cast the class (apache#26298)

…lowed the cluster to access (apache#26276)

…ort (apache#26314)

… factory integration (core migration) (apache#26282)

…ary glibc baseline (apache#26315)

apache#26316)

…#26255)

… in the client (apache#26317)

…y for testing (apache#26323)

…he is disabled (apache#26296)

…t provider is registered (apache#26154) Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

…ad manager (apache#26192)

…servlet additional servlets (apache#26444)

…ar-function-go (apache#26446) Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

…26449)

…lean close (apache#26433)

…cers (apache#25896)

…ets auto-scale policy (apache#26434)

…apache#26460)

… paths and restore configured ordering (apache#26462)

…ng broker entry metadata (apache#26464)

…ead of instances (apache#26467)

…26465)

…ange units and static stats state (apache#26466)

…ust memory parameters (apache#26468)

…ient (apache#26480) ### Motivation `pulsar-perf` was migrated to the V5 client API in apache#25887 and `pulsar-client` in apache#25917, so every subcommand of both tools now drives the V5 SDK. Pulsar 5.0 does not deprecate ordinary (non-scalable) topics and the v4 client remains fully supported, so it is still useful to drive the CLI tools with the v4 client: for benchmark results comparable across broker versions (the v4 commands can also target pre-5.0 brokers, which the V5 SDK cannot talk to at all), and for testing the v4 client and ordinary topics without the V5 SDK in the path. The migration also left a set of options that are still accepted but no longer do what they say, with no way to get the behaviour back — round-robin partition routing and the outstanding-message limits on `produce`, the real subscription types and the receiver-queue knobs on `consume`, the v4 `Reader` and its `<ledgerId>:<entryId>` start position on `read`, the v4 transaction coordinator on `transaction`, KeyValue schemas and non-durable subscriptions on `pulsar-client`, and `loadConf` for `client.conf` keys without a dedicated CLI flag. ### Modifications Each command is split into an abstract base holding everything that is not client-specific, plus two thin sibling subclasses that bind the client types — the existing V5 command and a new v4 one: ``` PerformanceProducerBase -> PerformanceProducer / PerformanceProducerV4 (produce-v4) PerformanceConsumerBase -> PerformanceConsumer / PerformanceConsumerV4 (consume-v4) PerformanceReaderBase -> PerformanceReader / PerformanceReaderV4 (read-v4) PerformanceTransactionBase -> PerformanceTransaction / PerformanceTransactionV4 (transaction-v4) AbstractCmdProduce -> CmdProduce / CmdProduceV4 (produce-v4) AbstractCmdConsumeCommand -> CmdConsume / CmdConsumeV4 (consume-v4) AbstractCmdReadCommand -> CmdRead / CmdReadV4 (read-v4) ``` `PulsarClientTool` builds a second, lazily-supplied v4 `ClientBuilder` for the `*-v4` commands; it keeps `loadConf` and accepts `http://` service URLs. `pulsar-shell` picks the new `pulsar-client` commands up for free, and both tools' `gen-doc` / `generate_documentation` render them. New subcommands only. The V5 commands keep their behaviour apart from a few deliberate fixes that the split surfaced (the `ProducerSocket` send/ack ordering and its `close()` null guard, the transaction receive-failure path), plus some added "use the `-v4` command" pointers on existing warnings — all listed in the PR description. ### Verifying this change Covered by `PulsarPerfTestToolTest`, `PerformanceV4CommandsTest`, `PerformanceTransactionV4Test`, `CmdV4CommandsTest`, `PulsarClientToolTest` (including the three KeyValue cases disabled by apache#25917, now re-enabled against `produce-v4`), and new `produce-v4` / `consume-v4` / `read-v4` cases in the `PerfToolTest` integration test.

…ad of under the wrapper write lock (apache#26463)

…ck allocation hotspot (apache#26500)

…xMetadataInterceptor isn't configured (apache#26497)

…the ownership barrier without recursion

…sting setter instead of reflection

SongOf force-pushed the fix/broker-ownership-cache-generation-race branch from 1f7e054 to 8f4d0b6 Compare July 22, 2026 14:38

[fix][broker] Fix ownership-generation races in OwnershipCache remove…

724a40a

…Ownership and lock-expiry cleanup

SongOf force-pushed the fix/broker-ownership-cache-generation-race branch from 8f4d0b6 to 724a40a Compare July 23, 2026 19:05

maxlisongsong and others added 27 commits July 27, 2026 23:03

Merge branch 'master' into fix/broker-ownership-cache-generation-race

c1211bc

# Conflicts: # pulsar-broker/src/main/java/org/apache/pulsar/broker/service/BrokerService.java # pulsar-broker/src/test/java/org/apache/pulsar/broker/service/BrokerServiceTest.java

[feat][pip] PIP-478: Asynchronous v5 client auth plugin interfaces an…

bb4768d

…d TLS material provider plugin interface (apache#25890)

[improve][broker] PIP-486 (PR5): broker-side entry-bucket handoff for…

f61d66c

… orderly consumer scaling (apache#26208)

[feat][misc] PIP-478: dedicated pulsar-tls-factory-api and pulsar-htt…

c61fc19

…p-client-api SPI modules (apache#26222)

[improve][io] Replace Spring Framework with cron-utils in CronTrigger…

534fd89

…er (apache#26269)

[fix][broker] Fix bucket delayed delivery counter and dedup consisten…

34c7c94

…cy (apache#26251)

[fix][admin] Avoid creating subscriptions when peeking messages if au…

2fbf416

…to-creation is disabled (apache#26279)

[feat][misc] PIP-478: default file-based TLS factory implementation (a…

bab5c77

…pache#26271)

[fix][misc] Remove unnecessary " {}" suffix from log lines when there…

af54f64

… are no context attributes (apache#26285)

[fix][txn] Fix X-Pulsar-txn-uncommitted header semantics and add X-Pu…

17de556

…lsar-txn-consumable for peek messages (apache#26073)

[fix][client] Make V5 async producer flush() await the caller's send …

e2d706b

…futures (apache#26287)

[fix][sec] Upgrade Netty to 4.2.17 to address several CVEs and bugs (a…

cb90c12

…pache#26300)

[improve][misc] Upgrade Jetty to 12.1.12 (apache#26302)

403aae6

[fix][ml] Skip contiguous deleted ranges during reads (apache#26299)

a87e36d

[cleanup][ml] Remove unused metadata quorum configuration fields (apa…

afa93b5

…che#26293)

[fix][broker] Wait for in-flight snapshot creation when trimming orph…

9a5bed3

…aned delayed-delivery buckets (apache#26309)

[fix][test] Fix flaky TestReplicationWorker.testRepairedNotAdheringPl…

a5e7937

…acementPolicyLedgerFragmentsOnRack (apache#26308)

[improve][broker] Expose interface for the replicator in ManagedLedge…

3412427

…r instead of cast the class (apache#26298)

[Fix][broker]Get a fenced error when loading a topic that does not al…

2babfbc

…lowed the cluster to access (apache#26276)

[improve][misc] Upgrade Conscrypt to 2.6.1 to add aarch64 native supp…

86ff200

…ort (apache#26314)

[feat][client] PIP-478: asynchronous v5 client authentication and TLS…

fd7eebb

… factory integration (core migration) (apache#26282)

[improve][misc] Upgrade Conscrypt to 2.6.2 to restore the native libr…

416d9d7

…ary glibc baseline (apache#26315)

[fix][test] Fix flaky AuditorPeriodicCheckTest.testFailedWriteRecovery (

96dd35a

apache#26316)

[improve][broker][proxy] Add httpMaxResponseHeaderSize config (apache…

ddd043b

…#26255)

[improve][client] PIP-478: drive the v5 authentication model natively…

75cc51e

… in the client (apache#26317)

[improve][ci] Gate CI in apache/pulsar on the pull request being read…

e712ffe

…y for testing (apache#26323)

[fix][ml] Make inflight reads limiter work even if managed ledger cac…

3061275

…he is disabled (apache#26296)

david-streamlio and others added 25 commits August 28, 2026 08:54

[improve][client] Seed MessageCryptoBc from the BC-FIPS DRBG when tha…

367caff

…t provider is registered (apache#26154) Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

[improve][build] Upgrade Gradle to 9.7.1 (apache#26440)

10c6cf1

[fix][broker] Add missing bundle Prometheus metrics for extensible lo…

41255fb

…ad manager (apache#26192)

[improve][build] Upgrade lightproto to 0.8.1 (apache#26445)

19c69ee

[improve][misc] PIP-472: Drop Jetty ee8 usage and adapt legacy javax.…

f3c7707

…servlet additional servlets (apache#26444)

[fix][sec] Bump google.golang.org/grpc from 1.82.1 to 1.83.1 in /puls…

8ae58a1

…ar-function-go (apache#26446) Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

[improve][test] Cover reverse equality for topic message IDs (apache#…

5951b6d

…26449)

[fix][client] PIP-486: explicit unsubscribe for scalable consumers' c…

c70039c

…lean close (apache#26433)

[fix][client] Fix incorrect topic in OTel spans for partitioned produ…

d4e8677

…cers (apache#25896)

[improve][broker] PIP-486: rebucket rollover and the segments-vs-buck…

94af75a

…ets auto-scale policy (apache#26434)

[improve][build] Profile Pulsar tests with async-profiler from Gradle (…

23fa956

…apache#26460)

[improve][broker] Avoid iterator allocation in BrokerInterceptors hot…

46182ff

… paths and restore configured ordering (apache#26462)

[improve][broker] Copy small entries into a flat buffer when prependi…

7010231

…ng broker entry metadata (apache#26464)

[fix][cli] pulsar-perf: register subcommand classes with picocli inst…

f41e620

…ead of instances (apache#26467)

[fix][cli] Apply PULSAR_MEM and PULSAR_GC in bin/pulsar-perf (apache#…

ffd9b6a

…26465)

[improve][cli] pulsar-perf: cut latency histogram memory ~100x, fix r…

bc1785d

…ange units and static stats state (apache#26466)

[improve][test] Make PulsarProfilingTest test Scalable Topics and adj…

8207618

…ust memory parameters (apache#26468)

[improve][build] Allow building with Java 26 (apache#26461)

67205b9

[fix][ml] Track all pending read callbacks for timeouts (apache#26081)

9ba61bd

[fix][misc] Resolve void by name in Reflections.loadClass (apache#26478)

dfe30a6

[improve][ml] Parse message metadata at entry cache insert time inste…

ce60798

…ad of under the wrapper write lock (apache#26463)

[improve][ml] Back LongBitmap with the on-heap RoaringBitmap to fix a…

00d475f

…ck allocation hotspot (apache#26500)

[fix][broker] Fix NPE in ManagedLedgerInterceptorImpl when AppendInde…

bef7b8f

…xMetadataInterceptor isn't configured (apache#26497)

Merge branch 'master' into fix/broker-ownership-cache-generation-race

b08c6f9

github-actions Bot added the PIP label Sep 8, 2026

SongOf force-pushed the fix/broker-ownership-cache-generation-race branch from b9ae6cc to 0adab8c Compare September 9, 2026 16:50

[fix][broker] Make the lock-expiry unload generation-aware and drain …

2166893

…the ownership barrier without recursion

SongOf force-pushed the fix/broker-ownership-cache-generation-race branch from 0adab8c to 2166893 Compare September 9, 2026 16:51

[improve][test] Inject the ownership cache spy through a VisibleForTe…

f39ab2b

…sting setter instead of reflection

Navigation Menu

[fix][broker] Fix ownership-generation races in OwnershipCache remove… - #1

[fix][broker] Fix ownership-generation races in OwnershipCache remove…#1
SongOf wants to merge 109 commits into
masterfrom
fix/broker-ownership-cache-generation-race

SongOf commented Jul 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Conversation

SongOf commented Jul 16, 2026

Motivation

Modifications

Verifying this change

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

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants