Bug2 #61477

Activity

1. opencodejr commented on Nov 3, 2026

   

   opencodejrBot

   on Nov 3, 2026

   Contributor

   More actions

   Thanks for the report, and sorry this has disrupted your work again.

   I think this is a duplicate of #61118, and it repeats your earlier report #61476. Possible duplicates and closely related issues, ordered by likelihood:

   • fix(core): avoid holding models.dev refresh lock across network timeout #61118: the same problem. The models.dev refresh holds a lock through its 10-second timeout, and /api/model, /api/agent and /api/provider return 503 on 2.2.5. fix(core): avoid holding models.dev refresh lock across network timeout #61120 is an open PR for it.
   • Bug #61476: your earlier issue about the same 503s, already marked as a duplicate of fix(core): avoid holding models.dev refresh lock across network timeout #61118.
   • Desktop: fatal "Location plugins did not finish activating" when the OpenAI login refresh at startup takes over 5 seconds #61443: Desktop 2.2.5 returns a 503 (Location plugins did not finish activating) on /api/provider and /api/model when plugins take longer than 5 seconds to start. If your 503 response body says this, that issue is the closer match.

   The issue mentions attached logs, but none came through. If you have the full server log or the body of one of the 503 responses, please add them to #61118. That helps more than opening new issues.

   About compensation: GitHub issues don't handle account requests like extended quota or premium access. Please email [email protected] with the details, and keep account information out of this public issue.

   The bot will close this issue as a duplicate in 1 day unless you reply explaining how your problem is different.
2. opencodejr

   added

   pending close

   triaging

   on Nov 3, 2026
3. vana30421-cpu commented on Nov 3, 2026

   

   vana30421-cpu

   on Nov 3, 2026

   Author

   More actions

   How my issue is differe

   message.txt

   m the standard #61118 scenario:

   1. Cache fallback does not work. The bot states it cannot reproduce the bug because "the server serves the cached or bundled model catalog without waiting on that lock." However, in my case (Desktop 2.2.5, Windows 11), the server does not serve the cache. It hits the 10-second lock and blocks foreground requests, returning persistent 503 for /api/model, /api/provider, and /api/agent. I have attached my full server log proving this continuous 503 behavior.
   2. Workaround failed. I applied OPENCODE_DISABLE_MODELS_FETCH=1 and restarted the system. It did not fix the issue. The lock is still being held, and models remain unavailable.
   3. Not the plugin issue (Desktop: fatal "Location plugins did not finish activating" when the OpenAI login refresh at startup takes over 5 seconds #61443). My 503 response is a generic Service Unavailable. It does not contain the text "Location plugins did not finish activating".
   4. Extensive troubleshooting. To try and fix this, I have completely reinstalled the application about 7 times from scratch, including full cache and config wipes. The fact that a clean install still results in this lock proves the fallback logic is fundamentally broken in my environment.

   Since this is the second time this has happened and it has completely halted my company's workflow, I need a concrete solution (a patch or a confirmed downgrade version) and a response regarding compensation.

   (Log file message.txt attached below)

   message.txt
4. opencodejr

   removed

   triaging

   pending close

   on Nov 3, 2026
5. opencodejr commented on Nov 3, 2026

   

   opencodejrBot

   on Nov 3, 2026

   Contributor

   More actions

   Thanks for the log. It showed this isn't the models.dev lock from #61118. Your log has no models.dev fetch at all, and every 503 takes exactly 5 seconds. That is the server's limit for waiting on plugins to start. The response is Location plugins did not finish activating.

   I reproduced it on 2.2.5 with opencode serve and a config containing only "plugin": ["opencode-token-tracker@latest"]. I also slowed down access to the npm registry to imitate a slow network. While the plugin was downloading, /api/provider, /api/model and /api/agent all returned 503 after 5 seconds, every time, and the built-in providers weren't available either. With a normal connection the download takes a few seconds and things recover. A slow or filtered connection to npm keeps everything at 503 for as long as the download takes. In your log the plugin only loaded about 140 seconds after startup, from a newly downloaded copy. That would explain why reinstalling and wiping the cache didn't help: each clean install downloads the plugin again. The same thing still happens on the latest dev (156bb61). The script below reproduces it.

   Until this is fixed, removing opencode-token-tracker@latest from the plugin list in your config should let models load. Pinning a version, or making sure npm can be reached quickly, may help too. One thing I couldn't reproduce: your log shows 503s continuing after the plugin had loaded. If they keep happening with the plugin removed, please say so here.

   Related: #61443 (the same 5-second plugin wait, caused by an OpenAI login refresh) and #57011 (a proposed change to keep serving models while plugins reload).

   Reproduction script (repro.sh)

   #!/usr/bin/env bash
   # Repro for #61477: start `opencode serve` with the npm plugin opencode-token-tracker@latest
   # and poll the plugin-gated endpoints.
   set -u
   OPENCODE=${OPENCODE:-opencode}
   T=${T:-$(mktemp -d /tmp/opencode/r61477.XXXX)}
   export HOME=$T/home XDG_CONFIG_HOME=$T/home/.config XDG_DATA_HOME=$T/home/.local/share XDG_CACHE_HOME=$T/home/.cache XDG_STATE_HOME=$T/home/.local/state
   mkdir -p $XDG_CONFIG_HOME/opencode $T/proj
   echo '{"plugin":["opencode-token-tracker@latest"]}' > $XDG_CONFIG_HOME/opencode/opencode.json
   # Simulate a slow npm registry (slow/filtered network): every request is delayed by $DELAY ms
   if [ -n "${DELAY:-}" ]; then
     echo "registry=http://127.0.0.1:4873/" > $HOME/.npmrc
     export NPM_CONFIG_REGISTRY=http://127.0.0.1:4873/ BUN_CONFIG_REGISTRY=http://127.0.0.1:4873/
     DELAY=$DELAY node "$(dirname "$0")/slow-registry.mjs" > $T/registry.log 2>&1 &
     RPID=$!
   fi
   cd $T/proj
   PORT=${PORT:-4977}
   OPENCODE_PASSWORD=pw $OPENCODE serve --port $PORT --print-logs --log-level info > $T/server.log 2>&1 &
   PID=$!
   sleep 3
   for i in $(seq 1 ${ROUNDS:-12}); do
     for ep in provider model agent; do
       code=$(curl -s -u opencode:pw -o $T/body -w '%{http_code} %{time_total}s' "http://127.0.0.1:$PORT/api/$ep?location%5Bdirectory%5D=$T/proj")
       echo "$(date -u +%T) /api/$ep $code $(head -c 160 $T/body)"
     done
   done
   kill $PID; [ -n "${RPID:-}" ] && kill $RPID
   echo '--- registry log'; cat $T/registry.log 2>/dev/null
   echo "--- server log (plugin / models.dev lines)"
   grep -Ei 'plugin|models.dev|error' $T/server.log | head -40

   Run it: bash repro.sh
6. opencodejr

   added

   reproduced

   on Nov 3, 2026
7. neriousy

   marked Bug #61476 as a duplicate of this issue on Nov 4, 2026
8. neriousy

   marked bug2 #61479 as a duplicate of this issue on Nov 4, 2026
9. neriousy commented on Nov 12, 2026

   

   neriousy

   on Nov 12, 2026

   Member

   More actions

   Sorry about the disruption. We don't offer compensation or quota credits for bugs. The models.dev fetch hang itself is tracked separately, and OPENCODE_DISABLE_MODELS_FETCH=1 is the workaround until that lands. Closing this one.
10. neriousy

    closed this as not plannedon Nov 12, 2026