Issues

• postgres: a reply buffered on the sync path / direct mode holds about 5x its data in live heap, and maxDirectResultBytes is checked after the append

  goceleris/celeris#1147 · Filter by author FumingPower3925 opened on Oct 27, 2026

• native engines ignore out-of-range SETTINGS values in HTTP2-Settings after a 101 (std ends the conn)

  goceleris/celeris#1146 · Filter by author FumingPower3925 opened on Oct 27, 2026

• adaptive: the oscillation lock and the cooldown suppress the io_uring error-rate safety revert

  goceleris/celeris#1145 · Filter by author FumingPower3925 opened on Oct 27, 2026

• test: TestNICIRQReportAddsNoLoop1036 can read the log before Listen logs 'listening' (start909 returns on the published address)

  bug
  goceleris/celeris#1108 · Filter by author FumingPower3925 opened on Oct 25, 2026 · v1.6.0

• test: TestRunSurvivesARingWaitEBUSY884/submit-and-wait-timeout fails its apparatus guard in Coverage's -race -covermode=atomic build (9 of ~264 runs)

  bug
  goceleris/celeris#1107 · Filter by author FumingPower3925 opened on Oct 25, 2026 · v1.6.0

• iouring: TestPort0HoldIsReleasedWhenListenFails895 failed once on main CI (1 socket still bound after a failed Listen); re-run passed

  goceleris/celeris#1104 · Filter by author FumingPower3925 opened on Oct 21, 2026 · v1.6.0

• test: TestBackloggedPeerIsClosed failed once under -race in CI (epoll/adaptive async-route, 2 of 4 responses)

  bug
  goceleris/celeris#1068 · Filter by author FumingPower3925 opened on Oct 18, 2026 · v1.6.0
  1

• arm64 (msr1) loop engines lose 15-62% rps at 128c-256c and on post-4k since the bench on dbbaaee; #973's pin policy leaves 4 of 12 loops unpinned and is the lead suspect

  goceleris/celeris#1036 · Filter by author FumingPower3925 opened on Oct 16, 2026 · v1.6.0
  3

• arm64 big.LITTLE (Cortex-A520): Go runtime preemption can stall goroutines on little cores for seconds (upstream Go issue; tracking, mitigated by #973)

  bugplatform/linuxupstream
  goceleris/celeris#1028 · Filter by author FumingPower3925 opened on Oct 13, 2026
  1

• h2 (epoll, io_uring, adaptive): the event loop keeps using a stream it looked up while the stream's pool handler can release it (WINDOW_UPDATE, SETTINGS and server-RST paths)

  bugprotocol/h2security
  goceleris/celeris#951 · Filter by author FumingPower3925 opened on Oct 4, 2026 · v1.6.0
  2

• v1.6.0 polish: measurements and evidence records

  performance
  goceleris/celeris#892 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  2

• v1.6.0 polish: documentation

  documentation
  goceleris/celeris#891 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  7

• v1.6.0 polish: CI and tests

  area/ciarea/testtesting
  goceleris/celeris#890 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  20

• v1.6.0 polish: protocols and router

  area/apiarea/protocolprotocol/h1protocol/h2
  goceleris/celeris#889 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  6

• v1.6.0 polish: middleware

  middleware
  goceleris/celeris#888 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  12

• v1.6.0 polish: drivers

  area/driver
  goceleris/celeris#887 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  1

• v1.6.0 polish: adaptive engine

  area/engine
  goceleris/celeris#886 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  4

• v1.6.0 polish: epoll engine

  area/engineengine/epollplatform/linux
  goceleris/celeris#885 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  2

• v1.6.0 polish: io_uring engine

  area/engineengine/iouringplatform/linux
  goceleris/celeris#884 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  4

• epoll (suspected): about 93 KiB left unread at a zero receive window for 1.2-3.9 s while the WebSocket handler waits idle

  bugengine/epollmiddlewareneeds-triage
  goceleris/celeris#871 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  3

• ws-hub-broadcast-128 on amd64: the loop engines' p99 frame gap rose from 1.29-1.59 ms to ~2.0 ms between v1.5.8 and 9f4d89b (same-run +16.8%, 8/8 engines) while CPU per frame fell 43%; competitors hold 1.09-1.44 ms

  engine/epollengine/iouringmiddlewareperformance
  goceleris/celeris#858 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  1

• arm64: celeris trails fasthttp on the memcached driver rows (0.66-0.84x) at 1.30-1.45x its SUT CPU per request, against parity at 0.89-0.94x on amd64; standing since v1.5.8

  area/driverperformanceplatform/linux
  goceleris/celeris#857 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0

• driver/redis: on arm64 the AsyncHandlers columns stall on a 10-command Pipeline: 2.0-6.4k rps against ~15k (sync) and 14-16k (competitors), with 11-31 of 90 s completing no request (since v1.5.8)

  area/driverbugdriver/redisperformanceplatform/linux
  goceleris/celeris#854 · Filter by author FumingPower3925 opened on Oct 2, 2026 · v1.6.0
  1

• Range requests: multi-range responses (multipart/byteranges)

  area/apienhancement
  goceleris/celeris#831 · Filter by author FumingPower3925 opened on Sep 29, 2026 · v1.7.0

• HTTP/2: a 64 MiB response to x/net's http2.Transport takes ~16 s on epoll, io_uring and adaptive, 0.07 s on std (net/http's client: 0.3-0.6 s)

  area/engineengine/epollengine/iouringperformanceprotocol/h2
  goceleris/celeris#809 · Filter by author FumingPower3925 opened on Sep 28, 2026 · v1.6.0
  2