Router.predict_batch dispatched on_predict_start, on_error and on_predict_end without a timeout, so Router(hooks_timeout=...) was silently ignored on the batched path while it applied to predict. Thread the instance value through and add a per-call hooks_timeout, validated like the other entry points. Documents the lock interaction: a timed-out hook keeps running outside the hooks_concurrent=False lock, so an overrunning hook no longer blocks the ones behind it.
7.5 KiB
Error handling
Hooks run inside the request they observe, so how their failures are handled matters. This page is the exact policy.
- The two policies
- What runs when something fails
- Exception chaining
- BaseException
- Configuration errors
- Warnings
- Choosing a policy
The two policies
hooks_raise controls what happens when a hook raises.
hooks_raise |
behaviour |
|---|---|
True (default) |
the hook exception propagates out of the call. |
False |
the hook is skipped with a RuntimeWarning and the call continues. |
It is set per instance and can be overridden per call (hooks_raise= on predict_batch,
system_one, Router.route, Router.predict). Per-call None means "use the instance value".
# strict: a broken audit hook fails the request
laya.load("convaiinnovations/laya", on_predict_end=audit, hooks_raise=True)
# lenient: telemetry must never take down a served request
laya.load("convaiinnovations/laya", on_predict_end=metrics, hooks_raise=False)
dispatch catches Exception. Anything that is not an Exception (see
BaseException) is never swallowed, even with hooks_raise=False.
What runs when something fails
The predict lifecycle is wrapped in try / except / finally, so cleanup hooks run on failure.
try:
on_predict_start
inference
except BaseException as exc:
ctx.error = exc
on_error (best effort; cannot mask exc)
raise
finally:
elapsed_ms, usage
on_predict_end (best effort; cannot mask exc on the failure path)
Failure matrix, per event and runtime:
| event | runtime | if it raises |
|---|---|---|
on_predict_start |
Agent / Router | hooks_raise=True: on_error and on_predict_end still run, then the exception propagates. False: warn and continue (mutations made before the raise remain). |
| inference | Agent / Router | ctx.error set, on_error runs, on_predict_end runs, exception propagates. |
on_error |
Agent / Router | never masks the original exception; chained as __context__. |
on_predict_end (success path) |
Agent / Router | hooks_raise=True: propagates (the result is computed but the call fails). False: warn. |
on_predict_end (failure path) |
Agent / Router | never masks the original exception; chained as __context__. |
on_route |
Router | propagates directly; there is no predict context yet. |
on_load |
Router | propagates directly; the checkpoint stays built and resident. |
on_evict |
Router | propagates directly; the checkpoint is already freed. |
Consequences worth knowing:
- A failing
on_loadleaves the model cached, so the nextloadreturns it without firingon_loadagain. - A failing
on_predict_endon the success path means the caller gets an exception instead of a result, even though inference succeeded. Usehooks_raise=Falsefor end hooks that are pure side effects.
Exception chaining
When a hook fails while another exception is already propagating, the original exception is
re-raised and the hook's exception is attached as __context__. The root cause is never lost.
class BadTelemetry:
def on_error(self, ctx):
raise RuntimeError("telemetry down")
try:
agent.system_one(state, questions, hooks=[BadTelemetry()])
except RuntimeError as exc:
assert exc.__context__ is not None # the telemetry failure
The same rule applies to a failing on_predict_end on the failure path.
BaseException
dispatch catches Exception, not BaseException, so KeyboardInterrupt and SystemExit
always propagate. They still trigger the except BaseException branch of the predict lifecycle,
which means on_error and on_predict_end run before the process unwinds. Keep those hooks
fast and non-blocking if you care about interrupt latency.
Configuration errors
Bad configuration fails fast with TypeError, before any inference:
| case | raised at | example |
|---|---|---|
| class instead of instance | construction | hooks=[MyHook] |
| no lifecycle method | construction | hooks=[object()] |
| non-callable event | construction | on_predict_start = 5 |
| non-callable convenience hook | construction | on_predict_start=123 |
plain callable in hooks= |
construction | hooks=[lambda ctx: None] |
Per-call hooks are validated when the call is made, so a bad per-call hook raises TypeError
from predict/system_one rather than at construction.
Warnings
With hooks_raise=False, each failing hook emits one RuntimeWarning naming the hook type and
event:
laya: hook Metrics.on_predict_end failed: connection reset
The warning is emitted once per failure, not once per hook definition, so a flaky hook under load can be noisy. Aggregate or rate-limit inside the hook if that matters.
Timeouts
hooks_timeout bounds each hook call in seconds. A hook still running after the limit is treated
as a hook failure: TimeoutError when hooks_raise=True, a RuntimeWarning when False. None
(the default) means no limit.
laya.load("convaiinnovations/laya", on_predict_end=metrics, hooks_timeout=2.0)
It can be set per instance or overridden per call on predict_batch, system_one,
Router.route, Router.predict and ONNXAgent.system_one. The value must be positive; 0 or a
negative number raises ValueError at the point it is set, rather than racing on a zero-length
join.
A timed hook runs on a worker thread in a copy of the caller's contextvars context, so a
request id or tracing span set by the caller is visible to the hook.
One honest caveat: Python cannot interrupt a thread, so a timed-out hook keeps running in the
background. The timeout bounds how long the request waits, not how long the hook lives. Use it to
keep a served request responsive, not to reclaim the work. For a hook that can hang, also give the
underlying call its own timeout (a socket or HTTP timeout). Because the thread cannot be
reclaimed, a hook that hangs on every call grows threads one per call; give a hook that can hang
its own bound rather than relying on hooks_timeout to stop it.
For an async hook, the coroutine runs on the event loop; a timeout on the calling side still returns after the limit, and the coroutine keeps running on the loop.
The timeout also releases the hooks_concurrent=False lock: dispatch waits for the hook only up
to the limit, then moves on, while the timed-out hook keeps running outside the lock. So the lock
serialises the hooks that finish in time, not every hook that was ever started; a hook that
overruns no longer blocks the ones behind it.
Choosing a policy
| hook kind | recommended | why |
|---|---|---|
| policy / guardrail / redaction | hooks_raise=True |
a policy that silently fails is a security hole. |
| audit / logging | hooks_raise=True in tests, often False in production |
losing an audit record should be loud, but not necessarily fatal. |
| metrics / tracing | hooks_raise=False |
observability must not fail the request. |
| cache read/write | hooks_raise=True |
a broken cache should surface, not silently miss. |
You can mix: install a guardrail with the instance default and give a telemetry hook its own
try/except, or use a separate hooks_raise per call.
See also
- Lifecycle: the
try/except/finallyshape in context. - Patterns and anti-patterns: common mistakes with error handling.