2.6.3 – Sep 26, 2026

2.6.3 (2026-09-26)

Bug Fixes

• tracking: recheck segment ownership after registering waiters (6e8779d)

  A rebalance can notify a tracker before its waiter is published. Recheck the assignment so available work does not wait for a long-poll deadline.

• websocket: drain terminal releases before closing transport (7f3cb97)

  Give pending tracker releases and admitted response callbacks bounded shutdown grace. Preserve abortive fallback and avoid waiting on the closing callback itself.

• tracking: preserve custom disconnect hooks during cleanup (f26c9a5)

  Retain virtual dispatch for ordinary connection cleanup while keeping frozen handover claims protected.

• tracking: release terminated consumers after pending reads (7d11a21)

  Keep batch processing intact during cancellation and release ownership after delayed reads and cached fallback reads finish. Reuse the existing disconnect protocol so a connected application does not retain abandoned segments.

• tracking: preserve batch ownership across runtime handovers (764cdd8)

  Freeze and restore active segment reservations without rewriting consumer positions. Keep late reads and explicit disconnects retryable while connection cleanup preserves transferred ownership.

Tests

• test-server: bind lifecycle fixtures to explicit loopback (b4cc745)

  Wildcard ephemeral listeners can share a port with an IPv4-specific listener on macOS. Bind clients and their test server to the same interface and bound connection retry time.

• proxy: keep capacity fixtures on distinct segments (c41c01d)

  Random routing keys could serialize two fixtures onto one segment and invalidate the capacity-only expectation.

Changes

• Merge pull request #404 from fluxzero-io/fix/runtime-batch-handoff (3c425b6)

  fix(tracking): preserve active batches during runtime handover