The http_pool section controls a single shared httpx.AsyncClient that handles all outgoing HTTP traffic — SSE backends, streamable-http backends, registry fetches, and management API requests. Sharing one pool avoids per-backend connection overhead and keeps total file-descriptor usage predictable.
http_pool:
enabled: true
max_connections: 200
max_keepalive: 100
timeout: 30
A conservative example for resource-constrained environments:
http_pool:
enabled: true
max_connections: 50
max_keepalive: 20
timeout: 15
| Field | Type | Default | Description |
|---|---|---|---|
enabled | bool | true | Enable shared HTTP connection pooling. When disabled, each request may open its own connection. |
max_connections | int | 200 | Maximum total simultaneous connections across all backends. Range: 1–2000. |
max_keepalive | int | 100 | Maximum idle keep-alive connections retained in the pool. Range: 0–2000. Set to 0 to disable keep-alive. |
timeout | float | 30.0 | Default request timeout in seconds applied to all HTTP requests unless overridden per-backend. Range: 1–300. |