NATS
v2.14.6Networking & Messaging2026年8月27日
性能改善と JetStream の正確性修正を中心としたメンテナンスリリースです。単独サーバー構成に関する制約の適用と Go ツールチェーンの更新も含まれます。
要対応 (1)
breakingコンシューマーの
direct、sourcing状態の保護コンシューマーの
directまたはsourcingの状態は、コンシューマー更新によって変更されなくなりました。
影響確認 (1)
breaking非クラスタ環境での複製数設定の制約
クラスタ構成ではないシステムで
replicasを 1 より大きく設定している場合に適用されます。replicasが 1 より大きいストリーム設定の更新は、クラスタ化されていないシステムでは拒否されるようになりました。
その他の記録済み変更 27 件すべてfixes 19 · value changes 7 · constraints 1
fixes (19)
- - WebSocket client buffers are recycled more aggressively when reallocating into larger buffers (#8518)
- - Fixed missing routes after a reconnect when gossiped URLs were rejected (#8527)
- - The filestore block sync no longer incorrectly skips blocks that were compacted during the same pass (#8456)
- - The filestore block sync now performs a single directory sync upon completion rather than once per compacted block (#8461)
- - The filestore I/O semaphore now covers block sync operations (#8462)
- - Fixed some issues that could prevent stream snapshots from taking place on a clean shutdown (#8465)
- - The filestore inline compaction no longer ignores
sync_interval: always(#8475) - - The stream created time is now preserved after recovery on a standalone server (#8471, contributed by @Green7)
- - Fixed a data race that could apply when accessing the batch ID (#8369)
- - Fixed a data race when reading the consumer direct or sourcing status (#8478, contributed by @nberlee)
- - Fixed an issue where the server could fail to distinguish consumer tiers when enforcing limits (#8484)
- - Fixed a flow control problem where replicated consumers could get stuck after a leader change (#8488)
- - Fixed various consumer create issues that could destroy the state of an existing consumer with the same name (#8491)
- - Internal message delete proposals are no longer counted in the JS API statistics (#8502)
- - Dynamic filestore reservations no longer incorrectly shrink based on used storage after restarts (#8503)
- - Fixed a signalling issue that could stall Raft catchups (#8501)
- - Fixed an issue where a stale snapshot from a previous Raft group could be replayed if the group name was unchanged (#8501)
- - Consumer delivery counts will no longer underflow below zero (#8512, contributed by @no-hup)
- -
AckFlowControlconsumers from a work queue stream will no longer incorrectly ack messages outside of the consumer filter (#8431, #8528)
value changes (7)
- - 1.26.7
- - Removal from service import response maps is now constant-time (#8463)
- - The client code no longer resets the deadline more times than necessary when flushing client buffers (#8513)
- - The client write buffers are now freed correctly when the working buffer exceeds a rational size (#8515)
- - Stream reads are now isolated with a separate lock, which improves the performance of direct gets (#8486)
- - Improved the performance of looking up matching sequence numbers for multi-subject gets (#8486)
- - Reduced the number of heap escapes in subject tree matching (#8526)
constraints (1)
- - The pending append entry cache in-memory is now bounded by size as well as entry count (#8501)
NATSをスタックに追加
対応が必要なリリースが出たときに、週次メールでお知らせします。 今回の破壊的変更も、その一例です。