[BUG] Unavailable providers prevent others from immediately working

[rbray89]

Problem (one or two sentences)

When DNS to a provider is blocked, it prevents other providers from working until the original request(s) to load models gives up, even if those other providers are not configured.

Context (who is affected and when)

Anyone who uses a restricted DNS environment, at first use.

Reproduction steps

Have DNS fail for the following:
glama.ai
openrouter.ai
router.requesty.ai
api.getunbound.ai
ai-gateway.vercel.sh
api.hicap.ai

Expected result

Attempting to use openai/ollama/etc. will result in a normal quick response.

Actual result

Attempting to use openai/ollama/etc. will result in very delayed response.

Variations tried (optional)

Adding 0.0.0.0 glama.ai openrouter.ai router.requesty.ai api.getunbound.ai ai-gateway.vercel.sh api.hicap.ai to /etc/hosts results in expected behavior, where inference behaves correctly.

App Version

v3.50.5

API Provider (optional)

OpenAI Compatible

Model Used (optional)

N/A

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause:
The model fetcher functions in src/api/providers/fetchers/ (openrouter.ts, requesty.ts, unbound.ts, vercel-ai-gateway.ts) use axios.get() without any timeout configuration. Axios defaults to timeout=0 (infinite), so when DNS is blocked for these domains, each request hangs until the OS-level DNS timeout (often 30-120 seconds).

Meanwhile, the requestRouterModels handler in src/core/webview/webviewMessageHandler.ts uses Promise.allSettled() and only sends the aggregated routerModels message to the webview after all promises settle. So even one blocked DNS lookup delays model availability for all providers.

Proposed Fix (two parts):

1. Add timeouts to all model fetcher HTTP requests - Add a 10-second timeout to all axios.get() calls in the fetcher files (openrouter.ts, requesty.ts, unbound.ts, vercel-ai-gateway.ts). The roo.ts fetcher already has a 10-second timeout via AbortController, so this brings the others in line.

2. Send router models incrementally - Modify the requestRouterModels handler to send each provider's models to the webview as soon as they resolve, rather than waiting for all providers to finish. This way, providers with working DNS will have their models available immediately, while blocked providers fail gracefully in the background. Each resolved provider will send an intermediate routerModels message, and the final aggregated message will still be sent at the end for backward compatibility.

Files to modify:

• src/api/providers/fetchers/openrouter.ts - Add timeout to axios calls
• src/api/providers/fetchers/requesty.ts - Add timeout to axios call
• src/api/providers/fetchers/unbound.ts - Add timeout to axios call
• src/api/providers/fetchers/vercel-ai-gateway.ts - Add timeout to axios call
• src/core/webview/webviewMessageHandler.ts - Send models incrementally as each provider resolves

This approach ensures that functional providers are not blocked by unreachable ones, while maintaining backward compatibility.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[rbray89]

@roomote This looks like a good solution.

[roomote]

Thanks for the approval! I'll proceed with implementing the fix now -- adding timeouts to the fetcher HTTP requests and sending router models incrementally. I'll open a PR to share progress shortly.

[roomote]

Opened a draft PR with my attempt at addressing this: #11748

The PR adds 10-second timeouts to the fetcher HTTP requests (openrouter, requesty, unbound, vercel-ai-gateway) and sends router models incrementally to the webview as each provider resolves, rather than waiting for all to finish. Feedback welcome!