fix(providers): add threading lock to get_provider_registry() singleton
CI / push-validation (pull_request) Successful in 40s
CI / helm (pull_request) Successful in 46s
CI / build (pull_request) Successful in 53s
CI / lint (pull_request) Successful in 1m16s
CI / quality (pull_request) Successful in 1m22s
CI / typecheck (pull_request) Successful in 1m29s
CI / security (pull_request) Successful in 1m28s
CI / unit_tests (pull_request) Successful in 6m30s
CI / docker (pull_request) Successful in 1m47s
CI / integration_tests (pull_request) Successful in 10m48s
CI / coverage (pull_request) Successful in 11m33s
CI / status-check (pull_request) Successful in 3s

Added import threading to src/cleveragents/providers/registry.py to enable a module-level lock.
Defined _registry_lock: threading.Lock = threading.Lock() at module scope.
Wrapped get_provider_registry() body with with _registry_lock: to ensure thread-safety.
Wrapped reset_provider_registry() body with with _registry_lock: to ensure thread-safety.
Added TDD support: features/tdd_registry_thread_safety.feature with @tdd_issue @tdd_issue_10409 @mock_only.
Added step definitions: features/steps/tdd_registry_thread_safety_steps.py.

ISSUES CLOSED: #10478
This commit is contained in:
2026-04-19 09:59:38 +00:00
committed by drew
parent 14fa9f2124
commit e1cd306f6f
3 changed files with 116 additions and 5 deletions
@@ -0,0 +1,80 @@
"""Step definitions for TDD Issue #10409 — get_provider_registry() thread safety.
Verifies that get_provider_registry() is safe for concurrent access from
multiple threads, preventing the race condition where two threads can each
observe _registry is None and independently construct a ProviderRegistry,
violating the singleton contract.
"""
from __future__ import annotations
import threading
from typing import Any
from behave import given, then, when
from cleveragents.providers.registry import (
get_provider_registry,
reset_provider_registry,
)
__all__: list[str] = []
@given("the global provider registry has been reset")
def step_given_registry_reset(context: Any) -> None:
"""Reset the global provider registry to None before the test."""
reset_provider_registry()
context.registry_instances: list[object] = []
context.construction_count = 0
@when("two threads call get_provider_registry() simultaneously")
def step_when_two_threads_call_simultaneously(context: Any) -> None:
"""Spawn two threads that call get_provider_registry() at the same time."""
results: list[object] = []
results_lock = threading.Lock()
barrier = threading.Barrier(2)
def worker() -> None:
barrier.wait() # Synchronise both threads to maximise race window
result = get_provider_registry()
with results_lock:
results.append(result)
threads = [threading.Thread(target=worker) for _ in range(2)]
for t in threads:
t.start()
for t in threads:
t.join()
context.registry_instances = results
# Count is always 1 with the lock fix; without the fix it could be 2
# We infer construction count from whether instances are identical
context.construction_count = (
1 if (len(results) == 2 and results[0] is results[1]) else 2
)
@then("both threads should receive the identical registry instance")
def step_then_same_instance(context: Any) -> None:
"""Assert both threads received the exact same ProviderRegistry object."""
instances = context.registry_instances
assert len(instances) == 2, (
f"Expected 2 registry instances from threads, got {len(instances)}"
)
assert instances[0] is instances[1], (
f"Threads received different ProviderRegistry instances: "
f"id(instances[0])={id(instances[0])}, id(instances[1])={id(instances[1])}. "
f"This indicates a thread-safety race condition in get_provider_registry()."
)
@then("only one ProviderRegistry should have been constructed")
def step_then_one_construction(context: Any) -> None:
"""Assert that only one ProviderRegistry was constructed."""
assert context.construction_count == 1, (
f"Expected exactly 1 ProviderRegistry construction, "
f"got {context.construction_count}. "
f"This indicates a thread-safety race condition in get_provider_registry()."
)
@@ -0,0 +1,20 @@
@tdd_issue @tdd_issue_10409 @mock_only
Feature: TDD Issue #10409 — get_provider_registry() singleton is not thread-safe
As a developer running concurrent agent workers
I want get_provider_registry() to be thread-safe
So that only one ProviderRegistry instance is ever created under concurrent access
This test captures bug #10478. The get_provider_registry() function in
cleveragents.providers.registry implements a singleton pattern without
thread-safety guards. Under concurrent access, two threads can each observe
_registry is None before either has finished constructing the ProviderRegistry,
causing both to independently instantiate a new registry. This violates the
singleton contract and can lead to inconsistent provider state.
See CONTRIBUTING.md > Bug Fix Workflow > TDD Issue Test Tags.
Scenario: Concurrent calls to get_provider_registry return the same instance
Given the global provider registry has been reset
When two threads call get_provider_registry() simultaneously
Then both threads should receive the identical registry instance
And only one ProviderRegistry should have been constructed
+16 -5
View File
@@ -14,6 +14,7 @@ Following ADR-008 (Provider Plugin Architecture), this registry provides:
from __future__ import annotations
import os
import threading
from dataclasses import dataclass
from enum import StrEnum
from typing import TYPE_CHECKING, Any, ClassVar
@@ -794,13 +795,18 @@ def resolve_provider_by_name(
return registry.create_ai_provider(provider_type=provider_type)
# Global registry instance
# Global registry instance and its thread-safety lock
_registry: ProviderRegistry | None = None
_registry_lock: threading.Lock = threading.Lock()
def get_provider_registry(settings: Settings | None = None) -> ProviderRegistry:
"""Get the global provider registry instance.
This function is thread-safe: all reads and writes to ``_registry`` are
protected by ``_registry_lock``, ensuring only one ``ProviderRegistry``
instance is ever created under concurrent access.
Args:
settings: Optional settings instance. Creates new registry if provided.
@@ -808,15 +814,20 @@ def get_provider_registry(settings: Settings | None = None) -> ProviderRegistry:
The global ProviderRegistry instance.
"""
global _registry
if _registry is None or settings is not None:
_registry = ProviderRegistry(settings)
return _registry
with _registry_lock:
if _registry is None or settings is not None:
_registry = ProviderRegistry(settings)
return _registry
def reset_provider_registry() -> None:
"""Reset the global provider registry.
This function is thread-safe: the write to ``_registry`` is protected by
``_registry_lock``.
Useful for testing to ensure clean state between tests.
"""
global _registry
_registry = None
with _registry_lock:
_registry = None