How to Maintain Trace Association in Parallel Python Processes | Arize AI Community

Arize AI Community Icon

How to Maintain Trace Association in Parallel Python Processes | Arize AI Community

RunLLM
APP
·
To ensure that the traces within setup_donation remain associated with the parent play_round span, you need to propagate the trace context to the threads executed by the ThreadPoolExecutor. This can be achieved by explicitly passing the context to each thread. Here's how you can modify your code to achieve this: LBKSfrom opentelemetry import context as context_apiLBKSdef play_round(self): with tracer.start_as_current_span("play_round") as span: span.set_attribute(SpanAttributes.INPUT_VALUE, self.game_state.model_dump_json()) span.set_attribute(SpanAttributes.OPENINFERENCE_SPAN_KIND, OpenInferenceSpanKindValues.CHAIN.value) pairs = self.create_donor_recipient_pairs() span.set_attribute(SpanAttributes.OUTPUT_VALUE, json.dumps([f"{donor.name} -> {recipient.name}" for donor, recipient in pairs]))LBKS # Capture the current context current_context = context_api.get_current()LBKS with ThreadPoolExecutor() as executor: futures = [ executor.submit(context_api.attach, current_context, donor.setup_donation, recipient, self.game_state, self.players) for donor, recipient in pairs ] # Wait for all donations to complete results = [future.result() for future in futures]LBKS for result in results: donor = self.find_player(result.donor_name) recipient = self.find_player(result.recipient_name) donor.execute_donation(recipient, self.game_state, result)LBKS self.game_state.round += 1LBKSThis approach ensures that the trace context is propagated to each thread, maintaining the association with the parent span. Would you like more detail on this process?Sources: - Slack Discussion
💡 Hint: Mention RunLLM in the thread for followups.
Sebastian S.
·
RunLLM yes, currently I am getting this error as submit expects the first variable to be the function and the rest to be variables: attach() takes 1 positional argument but 5 were given
RunLLM
APP
·
To resolve the error, you need to ensure that the context_api.attach function is used correctly. The attach function is used to set the context for the current thread, and it should be called separately from the executor.submit call. Here's how you can modify your code: LBKSfrom opentelemetry import context as context_apiLBKSdef play_round(self): with tracer.start_as_current_span("play_round") as span: span.set_attribute(SpanAttributes.INPUT_VALUE, self.game_state.model_dump_json()) span.set_attribute(SpanAttributes.OPENINFERENCE_SPAN_KIND, OpenInferenceSpanKindValues.CHAIN.value) pairs = self.create_donor_recipient_pairs() span.set_attribute(SpanAttributes.OUTPUT_VALUE, json.dumps([f"{donor.name} -> {recipient.name}" for donor, recipient in pairs]))LBKS # Capture the current context current_context = context_api.get_current()LBKS def setup_donation_with_context(donor, recipient): # Attach the context to the current thread token = context_api.attach(current_context) try: return donor.setup_donation(recipient, self.game_state, self.players) finally: # Detach the context after the function completes context_api.detach(token)LBKS with ThreadPoolExecutor() as executor: futures = [ executor.submit(setup_donation_with_context, donor, recipient) for donor, recipient in pairs ] # Wait for all donations to complete results = [future.result() for future in futures]LBKS for result in results: donor = self.find_player(result.donor_name) recipient = self.find_player(result.recipient_name) donor.execute_donation(recipient, self.game_state, result)LBKS self.game_state.round += 1LBKSThis code ensures that the context is correctly attached and detached for each thread, maintaining the trace association. Would you like more details on this solution?Sources: - https://arize-ai.slack.com/archives/C04R3GXC8HK/p1714665357.760569
💡 Hint: Mention RunLLM in the thread for followups.
Sebastian S.
·
RunLLM this is what setup donation looks like, how should I update it: def setup_donation(self, recipient: "Player", game_state: GameState, players: List["Player"]) -> Decision: with tracer.start_as_current_span(f"execute_donation-{self.name}") as span: span.set_attribute(SpanAttributes.OPENINFERENCE_SPAN_KIND, OpenInferenceSpanKindValues.AGENT.value) span.set_attribute(SpanAttributes.INPUT_VALUE, json.dumps(self.model_dump())) donation_percent = self.generate_donation(game_state, recipient, players) donation_amount = self.wallet * donation_percent # if the donor doesn't have enough funds, donate all of their funds if self.wallet - donation_amount < 0: donation_amount = self.wallet span.set_attribute(SpanAttributes.OUTPUT_VALUE, json.dumps({ "donation_percent": donation_percent, "donation_amount": donation_amount, })) game_state_copy = GameState(**game_state.model_dump()) decision = Decision( donor_name=self.name, recipient_name=recipient.name, dynamic_game_state=game_state_copy, donation_percent=donation_percent, donation_amount=donation_amount, donor_wallet_before=self.wallet, donor_wallet_after=self.wallet - donation_amount ) return decision
RunLLM
APP
·
To ensure that the setup_donation function correctly propagates the trace context when executed in a separate thread, you should modify it to accept the context as a parameter and attach it at the beginning of the function. Here's how you can update the setup_donation function: LBKSfrom opentelemetry import context as context_apiLBKSdef setup_donation(self, recipient: "Player", game_state: GameState, players: List["Player"], context): # Attach the context to the current thread token = context_api.attach(context) try: with tracer.start_as_current_span(f"execute_donation-{self.name}") as span: span.set_attribute(SpanAttributes.OPENINFERENCE_SPAN_KIND, OpenInferenceSpanKindValues.AGENT.value) span.set_attribute(SpanAttributes.INPUT_VALUE, json.dumps(self.model_dump())) donation_percent = self.generate_donation(game_state, recipient, players) donation_amount = self.wallet * donation_percentLBKS # if the donor doesn't have enough funds, donate all of their funds if self.wallet - donation_amount < 0: donation_amount = self.walletLBKS span.set_attribute(SpanAttributes.OUTPUT_VALUE, json.dumps({ "donation_percent": donation_percent, "donation_amount": donation_amount, }))LBKS game_state_copy = GameState(**game_state.model_dump()) decision = Decision( donor_name=self.name, recipient_name=recipient.name, dynamic_game_state=game_state_copy, donation_percent=donation_percent, donation_amount=donation_amount, donor_wallet_before=self.wallet, donor_wallet_after=self.wallet - donation_amount ) return decision finally: # Detach the context after the function completes context_api.detach(token)LBKSThis modification ensures that the trace context is attached at the start of the setup_donation function and detached at the end, maintaining the trace association. Would you like more details on this solution?Sources: - <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1714665357.760569|Slack Discussion>
💡 Hint: Mention RunLLM in the thread for followups.