Private
Public Access
0
0
Files
manual_slop/src/warmup.py
T
ed 8d58d7fc46 fix(warmup): defer _done_event.set() until after callbacks fire
WarmupManager._record_success and _record_failure used to set
self._done_event.set() inside the with self._lock: block, BEFORE
calling the user-registered on_complete callbacks. This created
a race: a test thread calling mgr.wait() could observe
mgr.is_done() == True and proceed before the worker thread had
finished firing the callbacks. The mgr.on_complete caller would
then assert on state that the callback was supposed to mutate
(e.g. test_warmup_on_complete_callback_fires' `received` list).

Fix: move self._done_event.set() to AFTER the for cb in callbacks:
loop in both _record_success and _record_failure. The done event
is now set last, so wait() cannot return until all callbacks
have completed (or raised, which is swallowed by the try/except).

ALSO fix the previously-corrupted state of warmup.py (the result
of a misused set_file_slice edit that left orphaned code with no
def line for _record_failure). _record_failure is now a proper
class method with the def line restored.

ALSO fix tests/test_warmup.py:
  - test_warmup_on_complete_callback_fires: the test body was
    missing the pool/mgr setup. Added the missing lines.
  - test_warmup_done_event_set_after_all_complete: removed the
    racy `assert not mgr.is_done()` assertion that fires
    immediately after submit. On a fast machine, os/sys warmup
    completes in microseconds, so is_done() is already True
    by the time the assertion runs. The remaining assertion
    (`assert mgr.is_done()` after wait) still tests the
    semantic that the done event is set after completion.
  - Removed both `@pytest.mark.skip` markers; the underlying
    issues are now fixed in production code AND the tests.

Verified: 10/10 tests in tests/test_warmup.py pass (previously
2 skipped, 2 failed).
2026-06-07 16:02:30 -04:00

308 lines
11 KiB
Python

"""WarmupManager: import heavy modules on a background thread pool.
Per spec.md:2.2 Layer 3, the AppController's __init__ submits a warmup
job to the shared _io_pool for each heavy module (provider SDKs,
feature-gated GUI modules, etc.). After warmup completes, those modules
are in sys.modules and any function that calls _require_warmed(name)
gets an instant lookup instead of a multi-hundred-ms import.
Public API on the manager (and exposed on AppController via delegation):
mgr.submit(modules) - start warmup jobs (call once at init)
mgr.status() - {pending, completed, failed}
mgr.is_done() - bool
mgr.wait(timeout) - block until done
mgr.on_complete(callback) - register completion callback
mgr.canaries() - list[dict] of per-module canary records (observability)
mgr.reset() - clear state (for re-warmup, e.g. in tests)
Canary records (one per submitted module) carry:
canary_id: monotonic numeric ID (continues across resets)
module: module name
thread_name: name of the worker thread that did the import (e.g. "controller-io-0")
thread_id: threading.get_ident() of that worker
submit_ts: wall-clock when submit() was called
start_ts: wall-clock when the worker started the import
end_ts: wall-clock when the import finished
elapsed_ms: (end_ts - start_ts) * 1000
status: "running" | "completed" | "failed" | "cancelled"
error: error message string if status == "failed", else None
"""
import importlib
import sys
import threading
import time
from concurrent.futures import Future, ThreadPoolExecutor
from typing import Callable, Optional
CompletionCallback = Callable[[dict], None]
class WarmupManager:
def __init__(self, pool: ThreadPoolExecutor, log_to_stderr: bool = True) -> None:
self._pool = pool
self._lock = threading.Lock()
self._log_lock = threading.Lock()
self._done_event = threading.Event()
self._pending: list[str] = []
self._completed: list[str] = []
self._failed: list[str] = []
self._callbacks: list[CompletionCallback] = []
self._started = False
# Canary observability state (per-module import tracking).
self._canaries: list[dict] = []
self._next_canary_id: int = 1
# Stderr logging: when True, the manager prints a one-line summary
# of each completed/failed canary to stderr, plus a final aggregate
# line when the entire warmup finishes. Default True so production
# runs get observability for free. Tests can opt out.
self._log_to_stderr: bool = log_to_stderr
# Capture the main thread ident at construction time so we can flag
# any canary that ran on the main thread (a main-thread-purity violation).
self._main_thread_ident: int = threading.get_ident()
self._pending: list[str] = []
self._completed: list[str] = []
self._failed: list[str] = []
self._callbacks: list[CompletionCallback] = []
self._started = False
# Canary observability state (per-module import tracking).
self._canaries: list[dict] = []
self._next_canary_id: int = 1
# Stderr logging: when True, the manager prints a one-line summary
# of each completed/failed canary to stderr, plus a final aggregate
# line when the entire warmup finishes. Default True so production
# runs get observability for free. Tests can opt out.
self._log_to_stderr: bool = log_to_stderr
# Capture the main thread ident at construction time so we can flag
# any canary that ran on the main thread (a main-thread-purity violation).
self._main_thread_ident: int = threading.get_ident()
def submit(self, modules: list[str]) -> None:
submit_ts = time.time()
with self._lock:
if self._started:
raise RuntimeError("WarmupManager.submit() called twice; call reset() first")
self._pending = list(modules)
self._completed = []
self._failed = []
self._done_event.clear()
self._started = True
for name in modules:
canary = {
"canary_id": self._next_canary_id,
"module": name,
"thread_name": None,
"thread_id": None,
"submit_ts": submit_ts,
"start_ts": None,
"end_ts": None,
"elapsed_ms": None,
"status": "running",
"error": None,
}
self._next_canary_id += 1
self._canaries.append(canary)
for name in modules:
self._pool.submit(self._warmup_one, name)
def status(self) -> dict[str, list[str]]:
with self._lock:
return {
"pending": list(self._pending),
"completed": list(self._completed),
"failed": list(self._failed),
}
def canaries(self) -> list[dict]:
"""Return a defensive copy of the canary records (per-module import tracking)."""
with self._lock:
return [dict(c) for c in self._canaries]
def is_done(self) -> bool:
return self._done_event.is_set()
def wait(self, timeout: Optional[float] = None) -> bool:
return self._done_event.wait(timeout=timeout)
def on_complete(self, callback: CompletionCallback) -> None:
fire_now = False
with self._lock:
if self._done_event.is_set():
fire_now = True
snap = self._snapshot()
else:
self._callbacks.append(callback)
if fire_now:
try:
callback(snap)
except Exception:
pass
def reset(self) -> None:
with self._lock:
self._pending = []
self._completed = []
self._failed = []
self._done_event.clear()
self._callbacks = []
self._started = False
# Canary records are preserved across resets (full history).
# Any still-running canaries from the prior submit are marked
# "cancelled" so callers can distinguish.
for c in self._canaries:
if c["status"] == "running":
c["status"] = "cancelled"
c["end_ts"] = c.get("end_ts") or time.time()
if c.get("start_ts") and c["elapsed_ms"] is None:
c["elapsed_ms"] = (c["end_ts"] - c["start_ts"]) * 1000
def _warmup_one(self, name: str) -> None:
start_ts = time.time()
thread = threading.current_thread()
thread_name = thread.name
thread_id = thread.ident
# Mark start in the canary record (find by module name; running record exists).
with self._lock:
for c in self._canaries:
if c["module"] == name and c["status"] == "running" and c["start_ts"] is None:
c["thread_name"] = thread_name
c["thread_id"] = thread_id
c["start_ts"] = start_ts
break
try:
importlib.import_module(name)
except BaseException as e:
end_ts = time.time()
self._record_failure(name, e, end_ts)
else:
end_ts = time.time()
self._record_success(name, end_ts)
def _record_success(self, name: str, end_ts: Optional[float] = None) -> None:
if end_ts is None: end_ts = time.time()
callbacks: list[CompletionCallback] = []
canary_snapshot: Optional[dict] = None
all_done = False
with self._lock:
if name in self._pending:
self._pending.remove(name)
self._completed.append(name)
for c in self._canaries:
if c["module"] == name and c["status"] == "running":
c["status"] = "completed"
c["end_ts"] = end_ts
if c["start_ts"] is not None:
c["elapsed_ms"] = (end_ts - c["start_ts"]) * 1000
canary_snapshot = dict(c)
break
done = self._started and not self._pending
if done:
callbacks = list(self._callbacks)
all_done = True
# NOTE: do NOT set _done_event here. We set it AFTER callbacks
# fire (below) so that `wait()` does not return before user
# on_complete callbacks have run. This closes the race where
# a test thread calling `mgr.wait()` could observe `is_done()`
# and proceed before the on_complete side effects were visible.
if canary_snapshot is not None:
self._log_canary(canary_snapshot)
if all_done:
self._log_summary()
for cb in callbacks:
try:
cb(self._snapshot())
except Exception:
pass
if all_done:
self._done_event.set()
def _record_failure(self, name: str, _err: BaseException, end_ts: Optional[float] = None) -> None:
if end_ts is None: end_ts = time.time()
callbacks: list[CompletionCallback] = []
canary_snapshot: Optional[dict] = None
all_done = False
with self._lock:
if name in self._pending:
self._pending.remove(name)
self._failed.append(name)
for c in self._canaries:
if c["module"] == name and c["status"] == "running":
c["status"] = "failed"
c["end_ts"] = end_ts
c["error"] = f"{type(_err).__name__}: {_err}"
if c["start_ts"] is not None:
c["elapsed_ms"] = (end_ts - c["start_ts"]) * 1000
canary_snapshot = dict(c)
break
done = self._started and not self._pending
if done:
callbacks = list(self._callbacks)
all_done = True
if canary_snapshot is not None:
self._log_canary(canary_snapshot)
if all_done:
self._log_summary()
for cb in callbacks:
try:
cb(self._snapshot())
except Exception:
pass
if all_done:
self._done_event.set()
def _log_canary(self, canary: dict) -> None:
if not self._log_to_stderr: return
cid = canary["canary_id"]
module = canary["module"]
thread_name = canary.get("thread_name") or "?"
thread_id = canary.get("thread_id")
elapsed = canary.get("elapsed_ms")
status = canary["status"]
is_main = thread_id is not None and thread_id == self._main_thread_ident
main_tag = " [MAIN-THREAD]" if is_main else ""
elapsed_str = f"{elapsed:.1f}ms" if elapsed is not None else "?ms"
with self._log_lock:
if status == "completed":
line = f"[warmup {cid}] {module} on {thread_name} (id={thread_id}): {elapsed_str}{main_tag}\n"
elif status == "failed":
err = canary.get("error") or "?"
line = f"[warmup {cid}] FAILED {module} on {thread_name} (id={thread_id}): {err}{main_tag}\n"
else:
line = f"[warmup {cid}] {status.upper()} {module} on {thread_name} (id={thread_id}){main_tag}\n"
try:
sys.stderr.write(line)
sys.stderr.flush()
except Exception: pass
def _log_summary(self) -> None:
if not self._log_to_stderr: return
with self._lock:
canaries = list(self._canaries)
if not canaries: return
total = len(canaries)
completed = sum(1 for c in canaries if c["status"] == "completed")
failed = sum(1 for c in canaries if c["status"] == "failed")
cancelled = sum(1 for c in canaries if c["status"] == "cancelled")
main_thread_violations = [c["module"] for c in canaries if c.get("thread_id") == self._main_thread_ident]
total_ms = 0.0
for c in canaries:
if c.get("elapsed_ms"): total_ms += c["elapsed_ms"]
parts = [f"{completed} completed"]
if failed: parts.append(f"{failed} failed")
if cancelled: parts.append(f"{cancelled} cancelled")
with self._log_lock:
try:
sys.stderr.write(f"[warmup done] {total} modules: {', '.join(parts)} (sum of per-module elapsed: {total_ms:.1f}ms)\n")
if main_thread_violations:
sys.stderr.write(f"[warmup WARNING] {len(main_thread_violations)} module(s) loaded on the MAIN THREAD (violates main thread purity invariant): {', '.join(main_thread_violations)}\n")
sys.stderr.flush()
except Exception: pass
def _snapshot(self) -> dict[str, list[str]]:
return {
"pending": list(self._pending),
"completed": list(self._completed),
"failed": list(self._failed),
}