Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.6

Improved

JetStream

• The stream info and consumer info endpoints now return leader_since (#7189)
• The stream info and consumer info endpoints now return system_account and traffic_account (#7193)
• The jsz monitoring endpoint now returns system_account and traffic_account (#7193)

Fixed

JetStream

• In operator mode, the cluster_traffic state for an account is now restored correctly when enabling JetStream at startup (#7191)

Complete Changes

v2.11.9-RC.1...v2.11.9-RC.2

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.6

Improved

JetStream

• The raftz endpoint now reports the cluster traffic account (#7186)

Fixed

General

• Fix a panic that could happen at startup if building from source using non-Git version control (#7178)
• Fix an issue where issuing an account JWT update with a connection limit could cause older clients to be disconnected instead of newer ones (#7181, #7185)

JetStream

• The Nats-TTL header will now be correct if the subject delete marker TTL overwrites it (#7177)

Tests

• Tests have been improved (#7172)

Complete Changes

v2.11.8...v2.11.9-RC.1

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.6 (#7156)

Dependencies

• golang.org/x/crypto v0.41.0 (#7166)
• golang.org/x/sys v0.35.0 (#7166)

Added

General

• Community-contributed support for building on Solaris and Illumos (#7122, thanks to @jwntree)

Fixed

General

• String-to-integer parsing has been improved in various places to prevent overflows/underflows (#7145)

JetStream

• Fixed an incorrectly formatted log line when failing to load a block when recovering TTLs (#7150)
• Raft will now step down if a higher term is detected during a catchup (#7151)
• Raft will now more reliably ignore entries from previous/cancelled catchups that arrive late (#7151)
• Fix a potential panic that could happen by a division by zero when applying Raft entries (#7151)
• The healthcheck endpoint should no longer report transient errors for newly created or recently deleted consumers (#7154)
• Fix a potential panic when trying to truncate a filestore block that doesn't exist (#7162)
• Clean up stale index.db file when truncating so that it is not inconsistent if the truncate operation is interrupted (#7162)
• Fix an off-by-one problem when Raft truncates to the correct index at startup (#7162)
• Ephemeral consumers will always select an online server when created on a replicated stream (#7165)

Tests

• Tests have been improved (#7140)

Complete Changes

v2.11.7...v2.11.8

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.6 (#7156)

Dependencies

• golang.org/x/crypto v0.41.0 (#7166)
• golang.org/x/sys v0.35.0 (#7166)

Added

General

• Community-contributed support for building on Solaris and Illumos (#7122, thanks to @jwntree)

Fixed

General

• String-to-integer parsing has been improved in various places to prevent overflows/underflows (#7145)

JetStream

• Fixed an incorrectly formatted log line when failing to load a block when recovering TTLs (#7150)
• Raft will now step down if a higher term is detected during a catchup (#7151)
• Raft will now more reliably ignore entries from previous/cancelled catchups that arrive late (#7151)
• Fix a potential panic that could happen by a division by zero when applying Raft entries (#7151)
• The healthcheck endpoint should no longer report transient errors for newly created or recently deleted consumers (#7154)
• Fix a potential panic when trying to truncate a filestore block that doesn't exist (#7162)
• Clean up stale index.db file when truncating so that it is not inconsistent if the truncate operation is interrupted (#7162)
• Fix an off-by-one problem when Raft truncates to the correct index at startup (#7162)
• Ephemeral consumers will always select an online server when created on a replicated stream (#7165)

Tests

• Tests have been improved (#7140)

Complete Changes

v2.11.7...v2.11.8-RC.1

Changelog

This release also contains all changes up to and including v2.11.7.

Go Version

• 1.24.6

Added

General

• Exponential backoff on route/gateway reconnection attempts (#7042, #7048)
• Added allow_insecure_cipher_suites configuration option to tls block, as insecure ciphers are now disabled by default (#7144)

JetStream

• Atomic batch publishes (#6966, #7057, #7060, #7085, #7090, #7091, #7119, #7152)
  • Stage-and-commit style batches for JetStream message publishes with normal consistency checks
  • ADR: https://github.com/nats-io/nats-architecture-and-design/blob/main/adr/ADR-50.md
• Counter streams (#6973, #6988, #7081, #7118)
  • Counters are incremented or decremented by publishing to the stream with Nats-Incr headers, with full aggregate consistency over sources and mirrors
  • ADR: https://github.com/nats-io/nats-architecture-and-design/blob/main/adr/ADR-49.md
• Prioritised mode for consumer priority groups (#7113)
  • Allows for low-latency switching between clients based on the priority set
  • ADR: https://github.com/nats-io/nats-architecture-and-design/blob/main/adr/ADR-42.md
• Added no_hdr flag to message get and direct get API endpoints (#6994)
  • Messages returned with this flag set will not contain headers, allowing fast access to e.g. counter values
  • ADR: https://github.com/nats-io/nats-architecture-and-design/blob/main/adr/ADR-52.md
• Support for trusted proxies (#7153)
  • Allows enforcing that connections arrive via a NATS protocol-aware proxy
  • ADR: https://github.com/nats-io/nats-architecture-and-design/blob/main/adr/ADR-55.md
• Added partition(n) for deterministic hash-based partitioning and random(n) for non-deterministic random partitioning to subject transforms (#6950)
• Add server_metadata block to server config, allowing arbitrary key-pairs to be set on a per-server basis, similar to stream/consumer metadata (#6935)
• Asynchronous flush support for the filestore with the new allow_async_flush stream option (#7018)
• Support for empty votes, which improves how the Raft election logic handles nodes that have lost their stable storage and attempt to rejoin the cluster regardless (#7038)

Leafnodes

• Added disabled option for leafnode remotes, with support for reloading from the config at runtime (#7054)

Changed

JetStream

• The JetStream API level has been increased to 2 (#6969)
• The API surface is now in strict mode by default, erroring if unknown fields are provided in the request body (#7049)

Improved

General

• When logging closed connections, the remote is now logged if known, making it easier to identify where route/gateway/leafnode connections have dropped (#7077)

JetStream

• Raft nodes will no longer respond success to catchup messages and will ignore responses if not leader, fixing some potential stream desync scenarios (#6944)
• The stream leader is now responsible for remapping consumer assignments on scaling/moves, fixing some potential placement races (#7071, #7112, #7083)
• Replicated streams can now be created even if some of the replica nodes are offline (#7075)
• The storage type is now logged when logging resource limits exceeded (#7076)
• JetStream API requests are always handled from the worker pool, improving the semantics of the API request queue and logging when requests take too long (#7125)

Leafnodes

• Leafnode connections without auth no longer unexpectedly connect to the global account (#7116)

Fixed

JetStream

• Improved validation for stream/consumer configs (#7134)

Complete Changes

v2.11.7...v2.12.0-preview.1

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.5 (#7047)

Dependencies

• golang.org/x/crypto v0.40.0 (#7061)
• golang.org/x/sys v0.34.0 (#7061)

Added

General

• The SubjectMatchesFilter function is now available as an exported function for embedded use (#7051)
• The leafz monitoring endpoint now includes the connection ID (#7063)
• The monitoring endpoint index page now includes the endpoint names on hover (#7066, #7087)

Improved

JetStream

• Consumers with inactivity thresholds should no longer age out before processing acks (#7107)
• The Raft layer will no longer request store state on each apply (#7109)
• Tombstones in Raft log compactions will now be written asynchronously, similar to purges (#7109)
• When enabling per-message TTLs on a stream, existing messages with the Nats-TTL header are now scanned and processed (#7117)

Fixed

General

• Message header lookups with common prefixes will now return correctly in all cases, fixing a problem where the headers could be sensitive to ordering (#7065)
• Validate that the default_sentinel JWT is a bearer token for auth callout (7074)
• The $SYS.REQ.USER.INFO endpoint should now only be answered by the local server, fixing cases where the endpoint may sometimes return without full connection details (#7089)

JetStream

• The Raft layer will require recovery and snapshot handling at startup before campaigning for a leadership election, fixing a situation where a node could continue with an outdated stream (#7040)
• The Raft log will no longer be compacted until after a snapshot is written, improving crash resilience (#7043)
• A race condition when shutting down Raft nodes which could result in no snapshot being written has been fixed (#7045)
• Consumer pull requests that use no_wait or expires behaviour has been fixed with replicated consumers (#7046)
• Pull consumers with an inactive threshold will now consider pending acks when determining inactivity, preventing the consumer from being deleted while messages are being processed (#7052)
• Push consumers will now correctly error when trying to configure priority groups (#7053)
• Committed entry objects will now be correctly returned to the pool on error, reducing allocations (#7064)
• The time hash wheel used for per-message TTLs now correctly detects and expires messages with TTLs over an hour, previously it could take double the expected time (#7070)
• A potential panic when selecting message blocks during TTL recovery has been fixed (#7072)
• A KV purge operation with subject delete markers configured will no longer leave behind a redundant extra delete marker (#7026)
• Raft will now correctly attempt to truncate back to a snapshot if applies were not caught up instead of resetting the entire log (#7095)
• Store cipher conversion will now work correctly when combined with store compression (#7099)
• Truncate and erase operations in the filestore should now be consistent after a hard kill (#7100)
• When a filestore message block is deleted and an unclean shutdown results in a stale index.db, the deleted blocks are now correctly marked as lost data and the index is rebuilt (#7123)
• Fixed a potential underflow that could happen when modifying max_bytes reservations (#7131)

Tests

• Tests have been improved (#7031, #7036, #7050, #6728, #7096)

Complete Changes

v2.11.6...v2.11.7

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.5

Improved

JetStream

• When enabling per-message TTLs on a stream, existing messages with the Nats-TTL header are now scanned and processed (#7117)

Fixed

JetStream

• When a filestore message block is deleted and an unclean shutdown results in a stale index.db, the deleted blocks are now correctly marked as lost data and the index is rebuilt (#7123)

Complete Changes

v2.11.7-RC.2...v2.11.7-RC.3

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.5

Improved

JetStream

• Consumers with inactivity thresholds should no longer age out before processing acks (#7107)
• The Raft layer will no longer request store state on each apply (#7109)
• Tombstones in Raft log compactions will now be written asynchronously, similar to purges (#7109)

Fixed

General

• Truncate and erase operations in the filestore should now be consistent after a hard kill (#7100)

Complete Changes

v2.11.7-RC.1...v2.11.7-RC.2

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.5 (#7047)

Dependencies

• golang.org/x/crypto v0.40.0 (#7061)
• golang.org/x/sys v0.34.0 (#7061)

Added

General

• The SubjectMatchesFilter function is now available as an exported function for embedded use (#7051)
• The leafz monitoring endpoint now includes the connection ID (#7063)
• The monitoring endpoint index page now includes the endpoint names on hover (#7066, #7087)

Fixed

General

• Message header lookups with common prefixes will now return correctly in all cases, fixing a problem where the headers could be sensitive to ordering (#7065)
• Validate that the default_sentinel JWT is a bearer token for auth callout (#7074)
• The $SYS.REQ.USER.INFO endpoint should now only be answered by the local server, fixing cases where the endpoint may sometimes return without full connection details (#7089)

JetStream

• The Raft layer will require recovery and snapshot handling at startup before campaigning for a leadership election, fixing a situation where a node could continue with an outdated stream (#7040)
• The Raft log will no longer be compacted until after a snapshot is written, improving crash resilience (#7043)
• A race condition when shutting down Raft nodes which could result in no snapshot being written has been fixed (#7045)
• Consumer pull requests that use no_wait or expires behaviour has been fixed with replicated consumers (#7046)
• Pull consumers with an inactive threshold will now consider pending acks when determining inactivity, preventing the consumer from being deleted while messages are being processed (#7052)
• Push consumers will now correctly error when trying to configure priority groups (#7053)
• Committed entry objects will now be correctly returned to the pool on error, reducing allocations (#7064)
• The time hash wheel used for per-message TTLs now correctly detects and expires messages with TTLs over an hour, previously it could take double the expected time (#7070)
• A potential panic when selecting message blocks during TTL recovery has been fixed (#7072)
• A KV purge operation with subject delete markers configured will no longer leave behind a redundant extra delete marker (#7026)
• Raft will now correctly attempt to truncate back to a snapshot if applies were not caught up instead of resetting the entire log (#7095)
• Store cipher conversion will now work correctly when combined with store compression (#7099)

Tests

• Tests have been improved (#7031, #7036, #7050, #6728, #7096)

Complete Changes

v2.11.6...v2.11.7-RC.1

Changelog

Refer to the 2.11 Upgrade Guide for backwards compatibility notes with 2.10.x.

Go Version

• 1.24.4

Improved

JetStream

• Sources will no longer update their last seen timestamp when stalled, making it clearer how long it has been since the last contact (#7013)
• Overall consumer performance has been improved by reducing allocations and improving the num pending calculation (#7022)

Fixed

General

• The subsz monitoring endpoint now returns the correct total for subscription details, aligning behaviour with other endpoints for pagination (#7009)

JetStream

• Fixed a bug where filestore encryption could corrupt a message block if a write took place before a read after restarting the server (#7008)
• Fixed a performance regression introduced in v2.11.0 which could result in abnormally low throughput from filtered consumers and higher GC pressure (#7015)
• Healthchecks will no longer produce unexpected monitor goroutine warnings when a clustered stream is restored from a snapshot (#7019)
• A race condition that could result in removed streams incorrectly reappearing has been fixed (#7025)
• The reserved_memory and reserved_storage statistics will no longer underflow when no limits are set (#7024)

Complete Changes

v2.11.5...v2.11.6