Build applications that make decisions (chatbots, agents, simulations, etc...). Monitor, trace, persist, and execute on your own infrastructure.
Python aiburrchatbot-frameworkdagsgenerative-aigraphshacktoberfestllmopsllmsmlopspersistent-data-structurestate-machinestate-managementvisibility
Brings S3 tracker to parity with standard tracker
elijahbenizzy317ea6ae433e
diff --git a/burr/tracking/s3client.py b/burr/tracking/s3client.pyindex bb5de6ae..e2ce9c62 100644--- a/burr/tracking/s3client.py+++ b/burr/tracking/s3client.py@@ -17,6 +17,7 @@ from burr.common import types as burr_typesfrom burr.core import Action, ApplicationGraph, State, serdefrom burr.integrations.base import require_pluginfrom burr.tracking.base import SyncTrackingClient+from burr.tracking.client import StateKey, StreamStatefrom burr.tracking.common.models import (ApplicationMetadataModel,ApplicationModel,@@ -25,6 +26,9 @@ from burr.tracking.common.models import (BeginSpanModel,EndEntryModel,EndSpanModel,+ EndStreamModel,+ FirstItemStreamModel,+ InitializeStreamModel,PointerModel,)from burr.visibility import ActionSpan@@ -86,7 +90,16 @@ def _allowed_project_name(project_name: str, on_windows: bool) -> bool:return bool(re.match(pattern, project_name))-EventType = Union[BeginEntryModel, EndEntryModel, BeginSpanModel, EndSpanModel, AttributeModel]+EventType = Union[+ BeginEntryModel,+ EndEntryModel,+ BeginSpanModel,+ EndSpanModel,+ AttributeModel,+ InitializeStreamModel,+ FirstItemStreamModel,+ EndStreamModel,+]def unique_ordered_prefix() -> str:@@ -171,6 +184,7 @@ class S3TrackingClient(SyncTrackingClient):self.initialized = Falseself.running = Trueself.init()+ self.stream_state: Dict[StateKey, StreamState] = dict()def _get_time_partition(self):time = datetime.datetime.utcnow().isoformat()@@ -398,3 +412,85 @@ class S3TrackingClient(SyncTrackingClient):span_id=span.uid,)self.submit_log_event(end_span_model, app_id, partition_key)++ def copy(self):+ return S3TrackingClient(+ project=self.project,+ bucket=self.bucket,+ region=self.region,+ endpoint_url=self.endpoint_url,+ non_blocking=self.non_blocking,+ serde_kwargs=self.serde_kwargs,+ unique_tracker_id=self.unique_tracker_id,+ flush_interval=self.flush_interval,+ )++ def pre_start_stream(+ self,+ *,+ action: str,+ sequence_id: int,+ app_id: str,+ partition_key: Optional[str],+ **future_kwargs: Any,+ ):+ initialize_stream_model = InitializeStreamModel(+ action_sequence_id=sequence_id,+ span_id=None,+ stream_init_time=system.now(),+ )+ self.submit_log_event(initialize_stream_model, app_id, partition_key)+ self.stream_state[app_id, action, partition_key] = StreamState(+ stream_init_time=system.now(),+ count=0,+ )++ def post_stream_item(+ self,+ *,+ item: Any,+ item_index: int,+ stream_initialize_time: datetime.datetime,+ first_stream_item_start_time: datetime.datetime,+ action: str,+ sequence_id: int,+ app_id: str,+ partition_key: Optional[str],+ **future_kwargs: Any,+ ):+ stream_state = self.stream_state[app_id, action, partition_key]+ if stream_state.count == 0:+ stream_state.count += 1+ self.submit_log_event(+ FirstItemStreamModel(+ action_sequence_id=sequence_id,+ span_id=None,+ first_item_time=system.now(),+ ),+ app_id,+ partition_key,+ )+ else:+ stream_state.count += 1++ def post_end_stream(+ self,+ *,+ action: str,+ sequence_id: int,+ app_id: str,+ partition_key: Optional[str],+ **future_kwargs: Any,+ ):+ stream_state = self.stream_state[app_id, action, partition_key]+ self.submit_log_event(+ EndStreamModel(+ action_sequence_id=sequence_id,+ span_id=None,+ end_time=system.now(),+ items_streamed=stream_state.count,+ ),+ app_id,+ partition_key,+ )+ del self.stream_state[app_id, action, partition_key]