[V2] "Unsupported package for ollama/<model>" when Ollama /api/tags takes longer than 1 s (hardcoded discovery timeout) #53341

Activity

1. github-actions

   assigned

   rekram1-node

   on Oct 5, 2026
2. jwefers commented on Oct 8, 2026

   

   jwefers

   on Oct 8, 2026

   More actions

   same now for me in v2.0.22 for llama.cpp. Still workis in v2.0.18. I have not tested the versions in between
3. ynohat commented on Oct 24, 2026

   

   ynohat

   on Oct 24, 2026

   More actions

   Adding a case from vLLM, which uses the same helper (local.ts):

   • The same hard-coded 1s timeout affects vllm.ts and lmstudio.ts, not only Ollama. vLLM discovery calls /health first, so a slow /health blocks the model request too.
   • Our endpoint is reached through a security-related router that gives very poor network conditions: TCP connect about 9s, measured with curl, versus about 20 ms without it. A higher default or a configurable timeout is needed, not only a marginal fix.
   • The failure is silent. With a bare provider config (no models), discovery returns zero models, and the default INFO log has no WARN or ERROR entry. Logging discovery failures at WARN would make this diagnosable, as you suggested.

   Reproduction and measurements are in #58558.
4. opencodejr commented on Oct 27, 2026

   

   opencodejrBot

   on Oct 27, 2026

   Contributor

   More actions

   Thanks for the detailed report and the timings. I could reproduce this on 2.0.18, and it still happens on the latest dev (6e978118c).

   Steps:

   1. Run a mock Ollama server whose GET /api/tags takes 1.5 s.
   2. Configure providers.ollama with settings.baseURL set to http://127.0.0.1:<port>/v1 and qwen3:test listed under models.
   3. Run opencode run --standalone -m ollama/qwen3:test "hi".

   The result is Error: Unsupported package for ollama/qwen3:test: with an empty package name. The server only ever sees the /api/tags call. With a 100 ms delay, the same setup works on dev. The script below reproduces both cases.

   On current dev the discovery logic is shared by Ollama, LM Studio and vLLM in packages/core/src/plugin/provider/local.ts, and it still has the hard-coded Effect.timeout("1 second").

   Related:

   • fix(core): keep configured local models available without discovery #54011: an open PR that keeps models you list in the config usable when discovery fails. It doesn't change the 1 s timeout.
   • provider discovery: hard-coded 1s timeout fails remote endpoints #58558: the same timeout with vLLM behind a slow tunnel, closed as a duplicate of this issue.

   Reproduction script (repro.sh)

   #!/usr/bin/env bash
   # Reproduces #53341: Ollama discovery has a hard-coded 1 s timeout (packages/core/src/plugin/provider/local.ts);
   # when GET /api/tags takes longer, a configured ollama model fails with "Unsupported package for ollama/<model>:".
   # Usage: ./repro.sh                      (uses `opencode` on PATH)
   #        OC_DEV=/path/to/repo ./repro.sh (runs the CLI from source)
   set -u
   here=$(mktemp -d /tmp/opencode/mock-XXXX)
   # Mock Ollama server: /api/tags delayed by $DELAY ms, OpenAI-compatible chat streams a fixed reply.
   cat > "$here/mock.ts" <<'MOCK'
   // Mock Ollama: /api/tags delayed by DELAY ms (default 1500), chat completions streams "hello".
   const DELAY = Number(process.env.DELAY ?? 1500)
   const MODEL = "qwen3:test"
   const DETAILS = { format: "gguf", family: "qwen3", parameter_size: "8B", quantization_level: "Q4_K_M" }
   Bun.serve({ port: Number(process.env.PORT ?? 11999), async fetch(req) {
     const u = new URL(req.url); console.log(new Date().toISOString(), req.method, u.pathname)
     if (u.pathname === "/api/tags") { await Bun.sleep(DELAY); return Response.json({ models: [{ name: MODEL, model: MODEL, modified_at: "2026-01-01T00:00:00Z", digest: "abc", size: 1, details: DETAILS }] }) }
     if (u.pathname === "/api/show") return Response.json({ capabilities: ["completion","tools"], model_info: { "general.architecture": "qwen3", "qwen3.context_length": 32768 }, details: DETAILS })
     if (u.pathname === "/api/version") return Response.json({ version: "0.34.2" })
     if (u.pathname.endsWith("/models")) return Response.json({ object: "list", data: [{ id: MODEL, object: "model" }] })
     if (u.pathname.endsWith("/chat/completions")) {
       const chunk = (d: any) => `data: ${JSON.stringify(d)}\n\n`
       const base = { id: "x", object: "chat.completion.chunk", created: 0, model: MODEL }
       const body = chunk({ ...base, choices: [{ index: 0, delta: { role: "assistant", content: "hello from mock" }, finish_reason: null }] })
         + chunk({ ...base, choices: [{ index: 0, delta: {}, finish_reason: "stop" }], usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 } }) + "data: [DONE]\n\n"
       return new Response(body, { headers: { "content-type": "text/event-stream" } })
     }
     return new Response("nf", { status: 404 })
   }})
   MOCK
   if [ -n "${OC_DEV:-}" ]; then OC=(bun --conditions=browser src/index.ts); CWD=$OC_DEV/packages/cli
   else OC=(opencode); CWD=$(mktemp -d /tmp/opencode/proj-XXXX); fi
   run_case() { # $1=delay ms  $2=port
     (cd "$here" && DELAY=$1 PORT=$2 exec bun mock.ts > "$here/mock-$1.log" 2>&1) & local pid=$!; sleep 1
     local d; d=$(mktemp -d /tmp/opencode/case-XXXX); mkdir -p "$d/.config/opencode"
     echo "{ \"providers\": { \"ollama\": { \"settings\": { \"baseURL\": \"http://127.0.0.1:$2/v1\" }, \"models\": { \"qwen3:test\": {} } } } }" > "$d/.config/opencode/opencode.json"
     echo "=== /api/tags delay ${1} ms ==="
     (cd "$CWD" && XDG_CONFIG_HOME=$d/.config XDG_DATA_HOME=$d/.data XDG_STATE_HOME=$d/.state \
        timeout 90 "${OC[@]}" run --standalone -m ollama/qwen3:test "hi" 2>&1 | sed 's/\x1b\[[0-9;]*m//g' | tail -3)
     echo "--- mock log:"; cat "$here/mock-$1.log"; kill $pid 2>/dev/null
   }
   run_case 100 12100
   run_case 1500 12101

   Run it: bash repro.sh
5. opencodejr

   added

   reproduced

   on Oct 27, 2026
6. svenseeberg commented on Nov 11, 2026

   

   svenseeberg

   on Nov 11, 2026

   Last edited by svenseeberg

   More actions

   Did anyone already come up with a workaround?

   /edit: my problem was actually something else. I still had an outdated line in my opencode config JSON file. I had to replace the "npm": key/value with "package": "@opencode/ai/providers/openai-compatible",

   👍React with 👍1Reacted by Gary Pinkham
7. rekram1-node

   closed this as completedin #63700on Nov 13, 2026