Troubleshooting Metadata Updates in Phoenix Dashboard Traces

Roméo I. · 2024-09-11T17:19:50.015Z

Hi all, sorry to bother you again. I'm trying to add some metadatas to my traces. To do so i download all my traces with query_spans, add the metadatas, and then upload them back to my server with log_traces. Unfortunately my changes doesn't appear in the phoenix dashboard and log_traces execute without any errors. here's a snippet of my code. spans = phoenix_client.query_spans(start_time=last_eval_run_time) metadata_column_name = 'attributes.'+SpanAttributes.METADATA spans[metadata_column_name] = None for index, row in spans.iterrows(): spans.at[index, metadata_column_name] = "{'foo':'bar'}" dataset = TraceDataset(spans) phoenix_client.log_traces(dataset)

Troubleshooting Metadata Updates in Phoenix Dashboard Traces | Arize AI Community

11 comments

· Sorted by Oldest

RunLLM
APP
·
To add metadata to your traces and ensure they appear in the Phoenix dashboard, you need to follow these steps: 1. Retrieve Spans DataFrame: Use get_spans_dataframe to get the spans DataFrame. 2. Add Metadata: Modify the DataFrame to include the desired metadata. 3. Log Updated Traces: Use log_traces to upload the modified DataFrame back to the server.Here's a concise example: LBKSimport phoenix as pxLBKS# Initialize client client = px.Client()LBKS# Retrieve spans DataFrame spans_df = client.get_spans_dataframe()LBKS# Add metadata (example: adding a 'user' column) spans_df['user'] = 'example_user'LBKS# Log updated traces trace_dataset = px.TraceDataset(dataframe=spans_df) client.log_traces(trace_dataset)LBKSEnsure that the metadata is correctly formatted and added to the DataFrame. If the changes still don't appear, verify that the Phoenix server is correctly processing the updated traces.Would you like more detail on any specific part?Sources: - Customize Traces - Send Annotations to Phoenix
💡 Hint: Mention RunLLM in the thread for followups.
Roger Y.
·
The Phoenix server doesn’t allow updates to the same span_id. You would need to clear them from the server before re-uploading them
Xander S.
·
Hey Roméo I., you can also add metadata to traces using these instructions. https://docs.arize.com/phoenix/tracing/how-to-tracing/customize-spans#using_metadata
👍1
Roger Y.
·
you can also change the span_id manually before re-uploading
Roméo I.
·
Hi Xander S., thanks for the share, but unfortunately i can't add my metadatas at inference time, I have to do it later on.
Roméo I.
·
Roger Y. ok that makes sens, thanks for the tips !
Roméo I.
·
If I may, it could be nice to have to possibility to edit existing spans.
Roger Y.
·
np. What we can do is to add a parameter for the server that would tell it to overwrite of existing spans. If you like that, would you mind opening an enhancement request? For the time being, what you could do is to just change the trace and span_ids and re-uploading them perhaps to a different project
Roger Y.
·
if you need an example of how to reset the span_ids, here’s a function that we have been using to a similar task
Roger Y.
·
but don’t directly import that function though, it’s meant to be internal
Roméo I.
·
awesome, thanks Roger Y.!
👍1

RunLLM
APP
·
To add metadata to your traces and ensure they appear in the Phoenix dashboard, you need to follow these steps: 1. Retrieve Spans DataFrame: Use get_spans_dataframe to get the spans DataFrame. 2. Add Metadata: Modify the DataFrame to include the desired metadata. 3. Log Updated Traces: Use log_traces to upload the modified DataFrame back to the server.Here's a concise example: LBKSimport phoenix as pxLBKS# Initialize client client = px.Client()LBKS# Retrieve spans DataFrame spans_df = client.get_spans_dataframe()LBKS# Add metadata (example: adding a 'user' column) spans_df['user'] = 'example_user'LBKS# Log updated traces trace_dataset = px.TraceDataset(dataframe=spans_df) client.log_traces(trace_dataset)LBKSEnsure that the metadata is correctly formatted and added to the DataFrame. If the changes still don't appear, verify that the Phoenix server is correctly processing the updated traces.Would you like more detail on any specific part?Sources: - Customize Traces - Send Annotations to Phoenix
💡 Hint: Mention RunLLM in the thread for followups.
Roger Y.
·
The Phoenix server doesn’t allow updates to the same span_id. You would need to clear them from the server before re-uploading them
Xander S.
·
Hey Roméo I., you can also add metadata to traces using these instructions. https://docs.arize.com/phoenix/tracing/how-to-tracing/customize-spans#using_metadata
👍1
Roger Y.
·
you can also change the span_id manually before re-uploading
Roméo I.
·
Hi Xander S., thanks for the share, but unfortunately i can't add my metadatas at inference time, I have to do it later on.
Roméo I.
·
Roger Y. ok that makes sens, thanks for the tips !
Roméo I.
·
If I may, it could be nice to have to possibility to edit existing spans.
Roger Y.
·
np. What we can do is to add a parameter for the server that would tell it to overwrite of existing spans. If you like that, would you mind opening an enhancement request? For the time being, what you could do is to just change the trace and span_ids and re-uploading them perhaps to a different project
Roger Y.
·
if you need an example of how to reset the span_ids, here’s a function that we have been using to a similar task
Roger Y.
·
but don’t directly import that function though, it’s meant to be internal
Roméo I.
·
awesome, thanks Roger Y.!
👍1