Exhausted provider budgets are retried: 402 and structured quota codes are classified from message text only #47685

Activity

1. github-actions

   assigned

   neriousy

   on Sep 6, 2026
2. opencodejr commented on Oct 28, 2026

   

   opencodejrBot

   on Oct 28, 2026

   More actions

   Thanks for the detailed report. I tested it against a mock OpenAI-compatible provider on @opencode/cli@2.2.0 and on the latest dev (b6c43a524), counting how many requests OpenCode sends for each error:

   • HTTP 402 and 429 with insufficient_quota are already handled on both. They're classified as an exhausted quota and not retried: the mock gets one attempt per call and the provider's message is shown. The other OpenAI billing codes you list usually come with a 400/401 status, which is already not retried.
   • Google per-day QuotaFailure still reproduces. A 429 with status: "RESOURCE_EXHAUSTED", the message "Resource has been exhausted", and a QuotaFailure detail whose quotaId is GenerateRequestsPerDayPerProjectPerModel is treated as an ordinary rate limit. The mock got 12 requests (each call tried six times) before the error appeared. The classifier only reads reason from error.details[] and never looks at violations[].quotaId.

   The script below starts the mock provider and prints the request count for each case.

   Reproduction script (repro.sh)

   #!/usr/bin/env bash
   # Reproduce #47685: counts how many times OpenCode sends a request to a provider
   # that reports an exhausted budget. Usage: ./repro.sh [opencode command...]
   # Default command: `opencode` from PATH. For a source checkout:
   #   PRELOAD=<repo>/packages/tui/node_modules/@opentui/solid/scripts/preload.js ./repro.sh bun --conditions=browser --preload=<repo>/packages/tui/node_modules/@opentui/solid/scripts/preload.js <repo>/packages/cli/src/index.ts
   set -u
   CMD=("$@"); [ ${#CMD[@]} -eq 0 ] && CMD=(opencode)
   DIR=$(mktemp -d); cd "$DIR"; git init -q
   cat > server.ts <<'TS'
   const mode = process.env.MODE!
   let count = 0
   const bodies: Record<string, [number, unknown]> = {
     openai_insufficient_quota: [429, { error: { message: "You exceeded your current quota, please check your plan and billing details.", type: "insufficient_quota", code: "insufficient_quota" } }],
     http_402: [402, { error: { message: "Payment required" } }],
     google_per_day_quota: [429, { error: { code: 429, message: "Resource has been exhausted (e.g. check quota).", status: "RESOURCE_EXHAUSTED",
       details: [{ "@type": "type.googleapis.com/google.rpc.QuotaFailure", violations: [{ quotaMetric: "generativelanguage.googleapis.com/generate_requests_per_model_per_day", quotaId: "GenerateRequestsPerDayPerProjectPerModel" }] }] } }],
   }
   Bun.serve({ port: Number(process.env.PORT), fetch(req) {
     const u = new URL(req.url)
     if (u.pathname.endsWith("/chat/completions")) { count++; console.error(`REQUEST ${count} ${mode}`); const [s, b] = bodies[mode]; return Response.json(b, { status: s }) }
     if (u.pathname.endsWith("/models")) return Response.json({ data: [{ id: "m", object: "model" }] })
     return new Response("nf", { status: 404 })
   } })
   TS
   PORT=18765
   # When running from source, spawned server processes need the OpenTUI preload too.
   [ -n "${PRELOAD:-}" ] && printf "preload = [\"%s\"]\n" "$PRELOAD" > bunfig.toml
   cat > opencode.json <<JSON
   { "\$schema": "https://opencode.ai/config.json",
     "provider": { "mock": { "npm": "@ai-sdk/openai-compatible", "name": "Mock", "options": { "baseURL": "http://127.0.0.1:$PORT/v1", "apiKey": "x" }, "models": { "m": { "name": "m" } } } },
     "model": "mock/m" }
   JSON
   for MODE in openai_insufficient_quota http_402 google_per_day_quota; do
     MODE=$MODE PORT=$PORT bun server.ts 2> server.log & SP=$!; sleep 1
     timeout 90 "${CMD[@]}" run --standalone --model mock/m "hi" > out.log 2>&1; RC=$?
     kill $SP; wait $SP 2>/dev/null
     echo "== $MODE: exit=$RC requests=$(grep -c REQUEST server.log) (timeout=124)"; tail -3 out.log | cut -c1-300
   done

   Run it: bash repro.sh
3. opencodejr

   added

   reproduced

   on Oct 28, 2026