Skip to content

errors.md

Latest commit

 

History

History
54 lines (42 loc) · 2.31 KB

errors.md

File metadata and controls

54 lines (42 loc) · 2.31 KB
Raw

Error Hierarchy

All exceptions inherit from RunnerError, enabling callers to catch all runner-related failures with a single clause:

from model_arkestra import RunnerError

try:
    result = await runner.ainvoke("my-model", prompt="...")
except RunnerError as e:
    logger.error(f"Runner failed: {e}")

Exception Reference

Exception When raised
ServerReadyTimeout The server did not become ready (health check returned no 200 within ready_timeout). Raised only on total health-check exhaustion.
ModelNotStarted A request was made for a model name that doesn't exist in the config, or the model hasn't been started yet (start() not called). Raised by _dispatch when no tracked context exists, and during BaseModelRunner.start() when model config lookup fails.
ModelShutdown A request was made to a model that has been explicitly stopped via stop() or stop_all() (state is STOPPED or STOPPING). Raised by _dispatch.
MaxRestartsExceeded The runner crashed and exhausted all automatic restart attempts (restart_limit). State transitions to ERROR; raised by _dispatch on any subsequent request.
RunnerError Generic base class for all runner failures, including HTTP errors (502, 503, 504) from the server, connection failures, and general request errors.

Usage Examples

from model_arkestra import ServerReadyTimeout, ModelNotStarted, MaxRestartsExceeded, RunnerError

# Catch everything at once
try:
    result = await runner.ainvoke("my-model", prompt="...")
except RunnerError as e:
    logger.error(f"Runner failed: {e}")

# Handle specific cases
try:
    await runner.start("qwen3-4b")
except ServerReadyTimeout:
    print("Model didn't become ready — check GPU or binary path")
except ModelNotStarted:
    print(f"Model 'qwen3-4b' not found in config")

# Post-error requests raise MaxRestartsExceeded
try:
    await runner.ainvoke("unstable-model", prompt="hi")
except MaxRestartsExceeded:
    print("Model crashed too many times — check logs with admin endpoint")

Related Documentation