fix(core): sync model_serve_count gauge on stream request completion

[m199369309]

Summary

• decrease_serve_count (called by restful_api.py for streaming requests) only decremented _serve_count without writing the new value back to the Prometheus registry.
• This caused xinference:model_serve_count to monotonically increase with every streaming request and never decrease, making the gauge diverge from the actual in-flight request count over time. A worker restart was the only way to reset it.
• Fix: make decrease_serve_count async and call record_metrics after decrementing, mirroring the non-stream path in the decorator's finally block (model.py:147-153).

All four call sites in restful_api.py (lines 1222, 1365, 1617, 2468) already use await model.decrease_serve_count(), so no caller changes are needed.

Root cause

Path	_serve_count	Prometheus gauge
Request enters decorator	+1	set ✓
Non-stream ends (decorator finally)	-1	set ✓
Stream ends (restful_api finally → decrease_serve_count)	-1	not set ✗

Test plan

• Start xinference-local, load a streaming-capable LLM
• Send non-stream requests → verify xinference:model_serve_count returns to 0 after each request
• Send stream requests → verify gauge returns to 0 after each stream completes (primary verification)
• Send concurrent stream requests (N=4) → verify gauge returns to 0 after all streams complete
• Trigger client disconnect mid-stream → verify gauge returns to 0
• Exceed request_limits → verify gauge does not inflate abnormally

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the decrease_serve_count method in xinference/core/model.py to be asynchronous and records the updated model_serve_count metric. The reviewer suggested guarding the decrement operation with max(0, self._serve_count - 1) to prevent the serve count from becoming negative, which could otherwise bypass rate limiting checks.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[qinxuye]

LGTM