Priority don't seems to work

[kelvkhiu]

I added 2 provider

1 set to 0 another set to 1 for the priority.

It use 1 first. By right it should be using 0 first

[github-actions]

Thank you for reporting this issue.

Analysis:

The provider selection logic in this codebase uses lower number = higher priority. A provider with priority 0 should be selected before a provider with priority 1. This is confirmed in the core selection code:

• src/app/v1/_lib/proxy/provider-selector.ts (lines 1029-1039) - The selectTopPriority() method uses Math.min() to find the lowest priority value and only returns providers at that level.
• src/drizzle/schema.ts (line 164) - Priority defaults to 0 (highest).
• The UI form description also states: "Lower number = higher priority (0 is highest)" (messages/en/settings/providers/form/strings.json, line 96).

So with your setup (one provider at priority 0, another at priority 1), the system should use the priority-0 provider first. If you are seeing the priority-1 provider being used instead, there are a few possible explanations:

1. Session Reuse: The system has a session reuse mechanism (provider-selector.ts, lines 500-599) that binds a provider to a session via Redis. If a previous request in the same session already selected the priority-1 provider (e.g., before you changed priorities), subsequent requests within that session will continue reusing it regardless of priority. Sessions expire after a configurable TTL (default 300 seconds).

2. Health Filtering: Before priority selection occurs, the system filters out providers that are circuit-broken or over rate/spending limits (provider-selector.ts, lines 857-897). If your priority-0 provider has tripped a circuit breaker (e.g., due to consecutive failures) or exceeded a spending limit, it would be filtered out, leaving only the priority-1 provider.

3. Provider Type / Model Mismatch: The system also filters providers by type compatibility (claude, openai, codex, gemini) and model support. If your priority-0 provider does not match the request format or does not support the requested model, it will be excluded.

To help diagnose this, please provide:

• Are both providers the same type (e.g., both claude) and support the same models?
• Did you change the priority settings after already making requests? (Session reuse could explain the behavior)
• Do both providers have the same group tag, or is one untagged?
• Check the Provider Chain section in the dashboard - it shows the full decision context including which providers were filtered and why. This will reveal exactly why the priority-1 provider was selected.

Potential workaround:
If session reuse is the cause, waiting for the session TTL to expire (default 5 minutes) or restarting with a fresh session/API key should cause the system to respect the updated priorities.

---

Automated response from Claude AI

[kelvkhiu]

Are both providers the same type (e.g., both claude) and support the same models?
yes same claude, same model

Did you change the priority settings after already making requests? (Session reuse could explain the behavior)
Yes

Do both providers have the same group tag, or is one untagged?
Both group same

[ding113]

hi, you can click STATUS badge in Usage Logs Menu, then check Decision chain in Logic Trace Section.