Volcengine Ark model fails during reconnect / recovery continuation because it does not support the coding plan feature

[wtry]

Title

Volcengine Ark model fails during reconnect / recovery continuation because it does not support the coding plan feature

Environment

• Jcode version: v0.29.0
• OS: macOS
• Architecture: aarch64
• Provider: OpenAI-compatible
• Endpoint: https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions
• Model: volcengine:ark-code-latest
• Auth: ARK_API_KEY

Description

After the session reconnects successfully, Jcode attempts to continue a pending recovery directive. The continuation request fails with HTTP 404 from Volcengine Ark.

The provider error says:

The requested model does not support the coding plan feature.

It looks like Jcode sends a request that uses the coding plan feature during reconnect / recovery continuation, but the configured Ark model volcengine:ark-code-latest does not support that feature. Jcode then retries several times and eventually stops after reaching the auto-retry limit.

Steps to Reproduce

1. Configure an OpenAI-compatible Volcengine Ark provider.
2. Use model volcengine:ark-code-latest.
3. Let a session disconnect or reload while a recovery directive is pending.
4. Jcode reconnects successfully.
5. Jcode tries to continue the pending recovery directive.
6. The request fails with UnsupportedModel.
7. Auto-retry reaches the limit and stops.

Actual Behavior

Jcode reconnects successfully, but recovery continuation fails repeatedly:

✓ Reconnected successfully.

Reload complete - continuing because a recovery directive was pending.

OpenAI-compatible chat request failed
  endpoint: https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions
  model: volcengine:ark-code-latest
  auth: ARK_API_KEY
  status: 404 Not Found
  response: {"error":{"code":"UnsupportedModel","message":"The requested model does not support the coding plan feature. Please refer to the documentation at https://www.volcengine.com/docs/82379/1925114 to select a compatible model. Request id: 02178177399933089a6ab5dfd15f12fdf9b1912203a3c65809198","param":"","type":""}}
Hint: check network connectivity, DNS/TLS, that the base URL includes the API version (usually /v1), and that the model exists on the provider.

╭─────── ⚡ reconnecting ────────╮
│ Retrying · attempt 3/3 · in 6s │
│ Detail ⚠ Remote request failed │
╰────────────────────────────────╯

✓ Retrying continuation... (system message)

OpenAI-compatible chat request failed
  endpoint: https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions
  model: volcengine:ark-code-latest
  auth: ARK_API_KEY
  status: 404 Not Found
  response: {"error":{"code":"UnsupportedModel","message":"The requested model does not support the coding plan feature. Please refer to the documentation at https://www.volcengine.com/docs/82379/1925114 to select a compatible model. Request id: 02178177400190989a6ab5dfd15f12fdf9b1912203a3c653785cd","param":"","type":""}}

✓ Retrying continuation... (system message)

OpenAI-compatible chat request failed
  endpoint: https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions
  model: volcengine:ark-code-latest
  auth: ARK_API_KEY
  status: 404 Not Found
  response: {"error":{"code":"UnsupportedModel","message":"The requested model does not support the coding plan feature. Please refer to the documentation at https://www.volcengine.com/docs/82379/1925114 to select a compatible model. Request id: 02178177400654189a6ab5dfd15f12fdf9b1912203a3c6553bbba","param":"","type":""}}

✓ Retrying continuation... (system message)

OpenAI-compatible chat request failed
  endpoint: https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions
  model: volcengine:ark-code-latest
  auth: ARK_API_KEY
  status: 404 Not Found
  response: {"error":{"code":"UnsupportedModel","message":"The requested model does not support the coding plan feature. Please refer to the documentation at https://www.volcengine.com/docs/82379/1925114 to select a compatible model. Request id: 02178177401319189a6ab5dfd15f12fdf9b1912203a3c658d037b","param":"","type":""}}

⚠ Remote request failed. Auto-retry limit reached after 3 attempts. Use `/poke` again to retry manually.

[1jehuang]

Triage

Thanks for the detailed report. Two distinct things are happening here.

Root cause (provider-side / config)

The endpoint https://ark.cn-beijing.volces.com/api/coding/v3/chat/completions is Volcengine Ark's coding plan endpoint, which only accepts models that support that feature. ark-code-latest is not one of them, so Ark returns:

404 UnsupportedModel: The requested model does not support the coding plan feature.

This is the same error you'd get on the very first request, not something specific to reconnect/recovery. Recovery just happens to be the moment the next request fires after reload. The immediate workaround is to either:

• point the base URL at the standard endpoint (.../api/v3) instead of .../api/coding/v3, or
• use a model that the coding plan endpoint actually supports (see the Ark docs link in the error).

Jcode-side bug (the part we should fix)

A 404 UnsupportedModel is a permanent/deterministic error: retrying the identical request can never succeed. But Jcode's remote auto-retry path treats it as worth a short retry budget and burns all attempts (attempt 3/3) before giving up, producing the noisy repeated failures you saw.

The retry classifier (is_retryable_error in openai_stream_runtime.rs) correctly does not mark 404 as retryable, but the reconnect/recovery continuation layer (schedule_pending_remote_retry_with_limit in tui_lifecycle.rs) still gives deterministic 4xx model/endpoint errors a short auto-poke retry. We should short-circuit non-retryable 4xx model-capability errors (e.g. 404 UnsupportedModel, model-not-found) so they fail fast with an actionable hint instead of retrying.

Status

• Reproducible with a clear fix path.
• Priority: medium (no data loss; affects users on Ark's coding endpoint with an incompatible model).

Labeling accordingly.