Capturing User Feedback with LangChainInstrumentor in Arize

·Apr 25, 2025 09:59 PM

Hello guys, if I use LangChainInstrumentor, how I can https://docs.arize.com/phoenix/tracing/how-to-tracing/feedback-and-annotations/capture-feedback capture user feedback? The problem is I'm not able to get span id by using LangChainInstrumentor. 🥺

8 comments

· Sorted by Oldest

RunLLM
APP
·
To capture user feedback using `LangChainInstrumentor`, you need to retrieve the current span ID and then send annotations to Phoenix. You can use the `get_current_span` method from the `openinference.instrumentation.langchain` module to get the current span ID. Once you have the span ID, you can send feedback as annotations to Phoenix using its REST API. Here's a brief overview of the process: 1. *Retrieve Current Span ID*: Use `get_current_span` to get the current span ID. 2. *Send Annotations*: Use the Phoenix API to send annotations to the `/v1/span_annotations` endpoint. The annotation should include the span ID, feedback label, score, and any additional metadata. Would you like more detail on how to implement this in your application? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-interact-with-traces/capture-feedback.md|Capture Feedback>
💡 Hint: Mention RunLLM in the thread for followups.
Mikyo
·
Hey Junqi, this should be better documented (cc John G. when you are back) - but runllm is correct in that you can get the current span of a run using a utility in the langchain instrumentor. See https://github.com/Arize-ai/openinference/blob/2c7bd4af321504cdd7261411fe62d889c37[…]ngchain/src/openinference/instrumentation/langchain/__init__.py
Mikyo
·
so if you have the run_id, you can get the span

Roger Y.

the Send Annotations to Phoenix section in the doc page above has a code snippet for LangChain regarding span_id

from opentelemetry.trace import format_span_id
from openinference.instrumentation.langchain import get_current_span

span = get_current_span()
if span is not None:
   span_id = format_span_id(span.get_span_context().span_id)

RunLLM
APP
·
We've found some places in your existing documentation that contradict the support staff knowledge you've provided! See the docs <https://github.com/Arize-ai/phoenix/issues/4158|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1734948344.476019|here>, <https://github.com/Arize-ai/phoenix/issues/6399|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1716248614.485509|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1724005981.501479|here>, <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-interact-with-traces/capture-feedback.md|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1715202681.227929|here> and <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1715994066.749849|here> Please go <https://app.runllm.com/issues/assistant/172|here> to view and resolve all the inconsistencies found for this assistant.

Junqi Z.

from opentelemetry.trace import format_span_id
from openinference.instrumentation.langchain import get_current_span

span = get_current_span()
if span is not None:
   span_id = format_span_id(span.get_span_context().span_id)

I actually had this snippet in code. It neither works inside langgraph nor works outside.

Mikyo
·
hmm, - I'm less familiar with that method. Roger Y. might know better.
Roger Y.
·
i tried applying to this example, and was able to print out the span_id

Capturing User Feedback with LangChainInstrumentor in Arize | Arize AI Community

RunLLM
APP
·
To capture user feedback using `LangChainInstrumentor`, you need to retrieve the current span ID and then send annotations to Phoenix. You can use the `get_current_span` method from the `openinference.instrumentation.langchain` module to get the current span ID. Once you have the span ID, you can send feedback as annotations to Phoenix using its REST API. Here's a brief overview of the process: 1. *Retrieve Current Span ID*: Use `get_current_span` to get the current span ID. 2. *Send Annotations*: Use the Phoenix API to send annotations to the `/v1/span_annotations` endpoint. The annotation should include the span ID, feedback label, score, and any additional metadata. Would you like more detail on how to implement this in your application? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-interact-with-traces/capture-feedback.md|Capture Feedback>
💡 Hint: Mention RunLLM in the thread for followups.
Mikyo
·
Hey Junqi, this should be better documented (cc John G. when you are back) - but runllm is correct in that you can get the current span of a run using a utility in the langchain instrumentor. See https://github.com/Arize-ai/openinference/blob/2c7bd4af321504cdd7261411fe62d889c37[…]ngchain/src/openinference/instrumentation/langchain/__init__.py
Mikyo
·
so if you have the run_id, you can get the span

Roger Y.

the Send Annotations to Phoenix section in the doc page above has a code snippet for LangChain regarding span_id

from opentelemetry.trace import format_span_id
from openinference.instrumentation.langchain import get_current_span

span = get_current_span()
if span is not None:
   span_id = format_span_id(span.get_span_context().span_id)

RunLLM
APP
·
We've found some places in your existing documentation that contradict the support staff knowledge you've provided! See the docs <https://github.com/Arize-ai/phoenix/issues/4158|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1734948344.476019|here>, <https://github.com/Arize-ai/phoenix/issues/6399|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1716248614.485509|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1724005981.501479|here>, <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/how-to-interact-with-traces/capture-feedback.md|here>, <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1715202681.227929|here> and <https://arize-ai.slack.com/archives/C04R3GXC8HK/p1715994066.749849|here> Please go <https://app.runllm.com/issues/assistant/172|here> to view and resolve all the inconsistencies found for this assistant.

Junqi Z.

from opentelemetry.trace import format_span_id
from openinference.instrumentation.langchain import get_current_span

span = get_current_span()
if span is not None:
   span_id = format_span_id(span.get_span_context().span_id)

I actually had this snippet in code. It neither works inside langgraph nor works outside.

Mikyo
·
hmm, - I'm less familiar with that method. Roger Y. might know better.
Roger Y.
·
i tried applying to this example, and was able to print out the span_id