Automate browser based workflows with AI
Python aiapiautomationbrowserbrowser-automationcomputergptllmplaywrightpowerautomatepuppeteerpythonrpaseleniumvisionworkflow
Bound feature-flag flip staleness to snapshot sync (SKY-12962) (#7704)
pedrohsdb79371567980b
diff --git a/skyvern/forge/sdk/experimentation/providers.py b/skyvern/forge/sdk/experimentation/providers.pyindex 2b2e273f7..25a4ad179 100644--- a/skyvern/forge/sdk/experimentation/providers.py+++ b/skyvern/forge/sdk/experimentation/providers.py@@ -178,6 +178,15 @@ class BaseExperimentationProvider(ABC):async def _prepare_feature_flag_resolution(self, feature_name: str, *, cached: bool) -> None:return None+ def invalidate_resolution_caches(self) -> int:+ """Drop all cached resolutions; returns how many entries were dropped."""+ dropped = len(self.result_map) + len(self.tri_state_map) + len(self.variant_map) + len(self.payload_map)+ self.result_map.clear()+ self.tri_state_map.clear()+ self.variant_map.clear()+ self.payload_map.clear()+ return dropped+async def is_feature_enabled(self, feature_name: str, distinct_id: str, properties: dict | None = None) -> bool:await self._prepare_feature_flag_resolution(feature_name, cached=False)feature_flag_value = await self._is_feature_enabled(feature_name, distinct_id, properties)@@ -195,6 +204,9 @@ class BaseExperimentationProvider(ABC):await self._prepare_feature_flag_resolution(feature_name, cached=True)return await self._is_feature_enabled(feature_name, distinct_id, properties)+ # The prepare hook is the only suspension point in the cached resolvers:+ # nothing may await between cache consult and store, or a concurrent data+ # refresh could interleave and hits would lose generation consistency.async def is_feature_enabled_cached(self, feature_name: str, distinct_id: str, properties: dict | None = None) -> bool:@@ -202,12 +214,15 @@ class BaseExperimentationProvider(ABC):if should_bypass_feature_flag_cache(feature_name):await self._prepare_feature_flag_resolution(feature_name, cached=False)feature_flag_value = await self._is_feature_enabled(feature_name, distinct_id, properties)- elif cache_key in self.result_map:- feature_flag_value = self.result_map[cache_key]else:+ # Freshness first: a provider that reloads flag data here may+ # invalidate the maps, so a hit below is generation-consistent.await self._prepare_feature_flag_resolution(feature_name, cached=True)- feature_flag_value = await self._is_feature_enabled(feature_name, distinct_id, properties)- self.result_map[cache_key] = feature_flag_value+ if cache_key in self.result_map:+ feature_flag_value = self.result_map[cache_key]+ else:+ feature_flag_value = await self._is_feature_enabled(feature_name, distinct_id, properties)+ self.result_map[cache_key] = feature_flag_valuerecord_feature_flag_resolution(feature_name=feature_name,resolution_kind="enabled",@@ -232,12 +247,13 @@ class BaseExperimentationProvider(ABC):if should_bypass_feature_flag_cache(feature_name):await self._prepare_feature_flag_resolution(feature_name, cached=False)resolved = await self._resolve_feature_flag(feature_name, distinct_id, properties)- elif cache_key in self.tri_state_map:- resolved = self.tri_state_map[cache_key]else:await self._prepare_feature_flag_resolution(feature_name, cached=True)- resolved = await self._resolve_feature_flag(feature_name, distinct_id, properties)- self.tri_state_map[cache_key] = resolved+ if cache_key in self.tri_state_map:+ resolved = self.tri_state_map[cache_key]+ else:+ resolved = await self._resolve_feature_flag(feature_name, distinct_id, properties)+ self.tri_state_map[cache_key] = resolvedrecord_feature_flag_resolution(feature_name=feature_name,resolution_kind="enabled",@@ -275,10 +291,10 @@ class BaseExperimentationProvider(ABC):async def get_value_cached(self, feature_name: str, distinct_id: str, properties: dict | None = None) -> str | None:cache_key = _make_cache_key(feature_name, distinct_id, properties)+ await self._prepare_feature_flag_resolution(feature_name, cached=True)if cache_key in self.variant_map:variant = self.variant_map[cache_key]else:- await self._prepare_feature_flag_resolution(feature_name, cached=True)variant = await self._get_value(feature_name, distinct_id, properties)self.variant_map[cache_key] = variantrecord_feature_flag_resolution(@@ -290,10 +306,10 @@ class BaseExperimentationProvider(ABC):async def get_payload_cached(self, feature_name: str, distinct_id: str, properties: dict | None = None) -> Any:cache_key = _make_cache_key(feature_name, distinct_id, properties)+ await self._prepare_feature_flag_resolution(feature_name, cached=True)if cache_key in self.payload_map:payload = self.payload_map[cache_key]else:- await self._prepare_feature_flag_resolution(feature_name, cached=True)payload = await self._get_payload(feature_name, distinct_id, properties)self.payload_map[cache_key] = payloadrecord_feature_flag_resolution(diff --git a/tests/unit/test_feature_flag_data_hash_cache.py b/tests/unit/test_feature_flag_data_hash_cache.pyindex 35898c19d..0c87cf745 100644--- a/tests/unit/test_feature_flag_data_hash_cache.py+++ b/tests/unit/test_feature_flag_data_hash_cache.py@@ -88,10 +88,13 @@ async def test_feature_enabled_cached_bypasses_known_kill_switch_flags() -> Noneassert "RATE_LIMITING_ENABLED" in FEATURE_FLAG_CACHE_BYPASS_NAMESassert provider.enabled_calls == 3+ # Freshness runs before every cached consult (TTL-gated inside the+ # provider), so hits also record a cached-mode prepare call.assert provider.prepare_calls == [("RATE_LIMITING_ENABLED", False),("RATE_LIMITING_ENABLED", False),("NOT_A_KILL_SWITCH", True),+ ("NOT_A_KILL_SWITCH", True),]