## Fix OpenTelemetry Logging functionality

[tomhaynes]

Bugfixes

• select otel logging provider according to specified protocol
• normalise otel paths correctly, for different telemetry types

Relevant issues

No ticket, but previously enabling otel logging caused an AttributeError because it was assuming the wrong logging provider.

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🐛 Bug Fix
✅ Test

Changes

[vercel]

@tomhaynes is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[ghost]

hey @tomhaynes do you mind explaining the change you made here?

I don't follow what was causing the previous error - so it'll be helpful to understand the change

[tomhaynes]

Hi sure @krrishdholakia, there were two issues I was bumping into. I'm looking to forward logs to an otel-collector, I initially followed these docs https://docs.litellm.ai/docs/proxy/logging#opentelemetry

Initially there were just no logs generated (metrics / traces come through fine). I found a change from last month that seemed to require LITELLM_OTEL_INTEGRATION_ENABLE_EVENTS to be enabled - when I did this I got the following stacktrace:

Exception while exporting logs.
Traceback (most recent call last):
  File "/usr/lib/python3.13/site-packages/opentelemetry/sdk/_logs/_internal/export/__init__.py", line 311, in _export_batch
    self._exporter.export(self._log_records[:idx])  # type: ignore
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.13/site-packages/opentelemetry/exporter/otlp/proto/grpc/_log_exporter/__init__.py", line 108, in export
    return self._export(batch)
           ~~~~~~~~~~~~^^^^^^^
  File "/usr/lib/python3.13/site-packages/opentelemetry/exporter/otlp/proto/grpc/exporter.py", line 263, in _export
    request=self._translate_data(data),
            ~~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/usr/lib/python3.13/site-packages/opentelemetry/exporter/otlp/proto/grpc/_log_exporter/__init__.py", line 105, in _translate_data
    return encode_logs(data)
  File "/usr/lib/python3.13/site-packages/opentelemetry/exporter/otlp/proto/common/_internal/_log_encoder/__init__.py", line 38, in encode_logs
    return ExportLogsServiceRequest(resource_logs=_encode_resource_logs(batch))
                                                  ~~~~~~~~~~~~~~~~~~~~~^^^^^^^
  File "/usr/lib/python3.13/site-packages/opentelemetry/exporter/otlp/proto/common/_internal/_log_encoder/__init__.py", line 60, in _encode_resource_logs
    sdk_resource = sdk_log.log_record.resource
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'LogRecord' object has no attribute 'resource'

This sounded similar to this issue open-telemetry/opentelemetry-python#3552 (comment) that this LogRecord object should not be instantiated in this way. I also noticed that the code hardcodes the grpc exporter rather than setting it dynamically based on protocol (although the attribute error manifests with either grpc/http). I tweaked the import to import LogRecord as SdkLogRecord to avoid conflict with native python LogRecord.

The second issue I saw was similar to this #13942 - regarding the endpoint configuration, specifically the path appended to it (/v1/traces or /v1/logs etc) - my changes to normalise the endpoint for each telemetry type should resolve this issue also, although they argue that we should be using the dedicated endpoint config for each OTEL_EXPORTER_OTLP_TRACES_ENDPOINT etc. - this arguably would be slightly better, but my normalisation logic works as long as all telemetry is expected to go to the same collector.

Thanks to @mubashir1osmani who was helping me investigate this, and has confirmed that other customers are also failing to get otel logging through.

[ghost]

Thanks for explaining that!

@mubashir1osmani please feel free to submit a PR to update any relevant documentation on our end