v2.9.0-rc.1

This release adds OAuth2/OIDC SSO login to the management UI, sends publish confirms only after messages are flushed to disk (with a --no-sync opt-out), verifies PROXY protocol trusted sources, switches to Crystal's native kTLS support and supports per-queue-type policy apply-to targets. It also brings a number of stream, clustering and HTTP API fixes.

Added

• OAuth2/OIDC SSO login to the management UI #1768
• proxy_protocol_trusted_sources config option supporting individual IPs and CIDR notation #1601
• Startup warning when PROXY protocol is enabled without proxy_protocol_trusted_sources configured #1601
• load_definitions config option #1828
• Per-queue-type policy apply-to targets (classic_queues, quorum_queues, streams) for RabbitMQ definitions compatibility #2015
• Bindings count in overview, CLI status and Prometheus metrics #2018
• Per-vhost message stats and labeled channel metrics in /metrics/detailed #1926
• --no-sync option to disable syncfs on publish confirm #1987
• control_unix_path config option to make the lavinmqctl control socket path configurable #2029
• User tag classes to control management UI element visibility #1892

Changed

• Send publish confirms only after messages are flushed to disk with syncfs #1891
• Use Crystal's native kTLS support #1937
• Fiber-free synchronous replication a011ec99
• On-demand deliver_loop fibers for AMQP consumers #1722
• Idle fiber management for queue message expiration #1614
• Replace periodic GC.collect with on-demand GC #2016
• Batch persistence during definitions import #2014
• tcp_proxy_protocol now accepts boolean values (true/false/yes/no); legacy 1/2 are treated as enabled, 0 disables. Protocol version is auto-detected #1601
• Print clean error messages on boot failures instead of stacktraces aba7361b
• Group lavinmqctl help output by command category #1830
• Use amq.default in UI related operations #1913
• New action bar for tables with row counter and column selector #1900
• Bake static assets as string literals to reduce compile memory #1944
• Bump amq-protocol to 1.2.0 #1996
• Update amqp-client to 1.3.3, lz4 and systemd 3.0.1 dependencies cd2f4c58

Removed

• unix_proxy_protocol config option; Unix sockets always auto-detect PROXY protocol headers #1601

Fixed

• GET /api/queues/:vhost/:name no longer emits the message_stats field twice #2031
• GET /api/channels and GET /api/connections no longer include the per-metric rate-history log arrays in every list row, matching GET /api/queues. The logs are only needed by the per-object detail pages, so they are now returned solely by GET /api/channels/:name and GET /api/connections/:name, greatly reducing list response size and latency on large deployments #2025
• Skip the message-expire fiber on streams to avoid a Closed mfile crash when the last consumer disconnects after retention dropped segments ded4b69b
• Discard settlement frames (ack/nack/reject) for an already-closed channel instead of closing the connection with CHANNEL_ERROR f421d6e3
• Treat a publish to a concurrently-closed queue as dropped instead of raising, which previously surfaced as an HTTP publish 500 93f88429
• Stop federation and shovel links before tearing down vhosts on shutdown, and keep the embedded amqp-client's routine connection-teardown logging (already reported via the lmq.* federation/shovel layers) out of the broker log 2b213d3c
• Reply with Basic.GetEmpty instead of erroring when a basic_get races a concurrent queue delete 9fbf4dc4
• Serialize store saves (vhosts, parameters, users) so concurrent create/delete (under churn) don't race on the shared .tmp file and fail the rename f2316d0d
• Don't crash the MQTT brokers on a vhost Closed event for a vhost that was never registered (e.g. one whose create didn't finish) fcf592c4
• Serialize per-resource policy application and run stream drop_overflow under the message-store lock, fixing a segfault when policies were applied to a queue concurrently (e.g. under policy churn) with publishing/consuming 7f4b8399
• Race between stream queue delete and drop_overflow #1939
• Stream queue crash when delivery-limit policy is applied d3667f99
• Handle empty trailing stream segment on load 4efe79c4
• Serialize stream consumer-offset access under the message-store lock for MT safety #1973
• Skip delete_message after queue close #1961
• MessageStore size leaks in purge_all and requeued shift? #1957
• Handle reject-publish overflow in HTTP publish and multi-queue routes #1958
• Keep closing AMQP channels until close-ok #1980
• MQTT max inflight message flow control #1935
• Exit if etcd lease expires while waiting to be in sync #1969
• Wait lease TTL + 2 seconds before retrying etcd lease grant 57d898cd
• Delete empty queue directories on followers during streaming #2005
• Broker start failure for passwordless users with null hashing_algorithm #1898
• Anchor server uptime and CPU counters to OS process lifetime #2012
• Use :vhost named parameter for DELETE vhost-limits route #1994
• OpenAPI docs fixes #2004
• Log kTLS=off instead of kTLS= when offload is inactive #2019
• MQTT packet ID overflow in lavinmqperf #1997
• -q flag collision between QoS and quiet in mqttperf #1951
• Smooth pacing in lavinmqperf AMQP throughput 85711b32