fix(observability): support langfuse_otel in litellm handler and add opentelemetry dependencies

[sduccao]

Summary

Enables seamless Langfuse v3 tracing via OpenTelemetry (langfuse_otel) across PR-Agent by adding required OpenTelemetry dependencies and supporting langfuse_otel metadata in LiteLLMAIHandler.

Motivation & Problem

1. Langfuse v3 SDK Compatibility: PR-Agent uses langfuse==3.14.5 (required by MOSAICO's direct tracing APIs such as get_client, propagate_attributes, and start_as_current_observation).
2. sdk_integration Crash: In Langfuse v3, the legacy callback ["langfuse"] in LiteLLM raises TypeError: Langfuse.__init__() got an unexpected keyword argument 'sdk_integration'.
3. Missing OTel Dependencies & Metadata: When using langfuse_otel, LiteLLM requires OpenTelemetry packages (opentelemetry-api, opentelemetry-sdk, opentelemetry-exporter-otlp), and LiteLLMAIHandler previously only injected PR metadata (trace_name, tags, trace_metadata) when "langfuse" was in callbacks, omitting "langfuse_otel".

Solution

• Dependencies: Added opentelemetry-api==1.28.0, opentelemetry-sdk==1.28.0, and opentelemetry-exporter-otlp==1.28.0 to requirements.txt.
• LiteLLMAIHandler Callback Normalization: Automatically map legacy "langfuse" callback entries to "langfuse_otel" during handler initialization, ensuring backwards compatibility for users configuring LITELLM.SUCCESS_CALLBACK: ["langfuse"].
• Metadata Injection: Updated add_litellm_callbacks to inject trace metadata (command, pr_url, tags) for "langfuse_otel" callbacks.

[qodo-code-review]

PR Summary by Qodo

Pin Langfuse 2.59.7 for LiteLLM callback compatibility

🐞 Bug fix ⚙️ Configuration changes 🕐 Less than 5 minutes

AI Description

• Pins Langfuse 2.59.7 to satisfy LiteLLM’s supported v2 constraint.
• Prevents native Langfuse callbacks from failing on the removed sdk_integration argument.
• Documents alignment with LiteLLM’s official dependency configuration.

High-Level Assessment

The exact v2 pin is appropriate because the repository already uses exact dependency versions, LiteLLM explicitly requires Langfuse below v3, and 2.59.7 matches its official configuration. A broader >=2.59.7,<3.0 range would allow compatible updates but reduce build reproducibility.

Files changed (1) +3 / -1

Other (1) +3 / -1

requirements.txtDowngrade Langfuse to LiteLLM-compatible v2 release +3/-1

Downgrade Langfuse to LiteLLM-compatible v2 release

• Replaces Langfuse 3.14.5 with 2.59.7 to restore compatibility with LiteLLM’s native callback integration. Comments document the upstream constraint and the 'sdk_integration' incompatibility.

requirements.txt

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (1) 📘 Rule violations (1) 📜 Skill insights (0)

1. Duplicate Langfuse callbacks registered 🐞 Bug ≡ Correctness ⭐ New

Description

Mapping each entry independently turns ["langfuse", "langfuse_otel"] into two langfuse_otel
entries, causing duplicate tracing callbacks and duplicate Langfuse events. The same defect affects
success, failure, and service callback lists.

Code

pr_agent/algo/ai_handlers/litellm_ai_handler.py[153]

+                success_cb = ["langfuse_otel" if c == "langfuse" else c for c in success_cb]

Relevance

●●● Strong

Direct precedent accepts removing legacy Langfuse callbacks to prevent duplicate instrumentation;
same callback-normalization context.

PR-#2459

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

The new list comprehensions replace names but never deduplicate them. The repository's existing
MOSAICO registration deliberately removes legacy entries and appends langfuse_otel only when
absent, while its tests require the resulting callback count to equal one.

pr_agent/algo/ai_handlers/litellm_ai_handler.py[151-164]
pr_agent/mosaico/env_bridge.py[64-71]
tests/unittest/test_mosaico_env_bridge.py[89-108]
PR-#2459

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
Callback normalization replaces every legacy `langfuse` entry independently, so configurations containing both legacy and OTel callback names produce duplicate `langfuse_otel` registrations.

## Issue Context
Normalize each callback collection so non-Langfuse callbacks are preserved while legacy entries are removed and exactly one `langfuse_otel` entry remains. Apply this consistently to success, failure, and service callbacks, and add coverage for mixed and repeated entries.

## Fix Focus Areas
- pr_agent/algo/ai_handlers/litellm_ai_handler.py[151-164]
- tests/unittest/test_mosaico_env_bridge.py[89-108]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Downgrade disables MOSAICO tracing ✓ Resolved 🐞 Bug ◔ Observability

Description

Pinning langfuse==2.59.7 removes the Langfuse client APIs used by MOSAICO (get_client,
propagate_attributes, and start_as_current_observation). langfuse_span() and client
initialization catch the resulting failures and continue, so MOSAICO requests run without the
intended Langfuse spans while only emitting a warning.

Code

requirements.txt[R37-39]

+# litellm 1.96.2 requires langfuse v2 (<3.0) for the native 'langfuse' callback to avoid TypeError with sdk_integration.
+# Version matches litellm's own constraint in pyproject.toml (>=2.59.7,<3.0).
+langfuse==2.59.7

Relevance

●●● Strong

Accepted MOSAICO Langfuse compatibility findings show the team prioritizes preserving tracing when
dependency/API mismatches silently disable instrumentation.

PR-#2560
PR-#2459

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

The changed dependency forces the v2 SDK, while the repository directly imports and invokes the
newer client interface. Both affected paths intentionally swallow exceptions and proceed untraced,
proving the dependency change causes an observability regression in MOSAICO.

pr_agent/mosaico/observability.py[63-79]
pr_agent/mosaico/server.py[65-75]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The new global `langfuse==2.59.7` pin is compatible with LiteLLM's native callback but is incompatible with MOSAICO's direct instrumentation, which imports and uses the newer Langfuse client API. Because MOSAICO catches these exceptions, tracing is silently disabled rather than failing loudly.

## Issue Context
Preserve LiteLLM's requirement for a v2 client while ensuring MOSAICO retains functional tracing. Either update MOSAICO to the Langfuse 2.59.7 API and verify its trace-context behavior, or isolate the LiteLLM callback's dependency/API compatibility so both integrations can coexist. Add/adjust tests that exercise `_configure_langfuse()` and `langfuse_span()` with the pinned dependency.

## Fix Focus Areas
- requirements.txt[37-39]
- pr_agent/mosaico/observability.py[63-79]
- pr_agent/mosaico/server.py[65-75]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

3. Callback normalization lacks tests 📘 Rule violation ▣ Testability ⭐ New

Description

The new callback normalization changes configured langfuse callbacks to langfuse_otel, but the
PR does not add or update tests for this behavior. Regressions in success, failure, or service
callback handling could therefore go undetected.

Code

pr_agent/algo/ai_handlers/litellm_ai_handler.py[R151-154]

+            success_cb = get_settings().litellm.success_callback
+            if isinstance(success_cb, list):
+                success_cb = ["langfuse_otel" if c == "langfuse" else c for c in success_cb]
+            litellm.success_callback = success_cb

Relevance

●●● Strong

Recent handler feedback accepts regression-test requests for changed behavior, and callback changes
need branch coverage.

PR-#2584
PR-#2404

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

PR Compliance ID 2694678 requires tests to change when production behavior changes. Lines 151-164
introduce new conditional normalization for three callback settings, while the PR diff contains no
test-file changes exercising these branches.

Rule 2694678: Require tests to change when production code behavior changes
pr_agent/algo/ai_handlers/litellm_ai_handler.py[151-164]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The new `langfuse` to `langfuse_otel` callback normalization changes production behavior without corresponding test coverage.

## Issue Context
Add unit tests covering success, failure, and service callback lists, including preservation of unrelated callback entries and non-list values.

## Fix Focus Areas
- pr_agent/algo/ai_handlers/litellm_ai_handler.py[151-164]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

Context

✅ Compliance rules (platform): 34 rules

Review mode: ⚖️ Balanced: This behavioral callback normalization and metadata change affects three runtime callback paths plus dependency integration, so it carries real compatibility risk but is not dense enough to warrant redundant review passes.

_{ⓘ  1 issues published inline · 2 in summary}

• Author self-review: I have reviewed the code review findings, and addressed the relevant ones.

Tip of the day

💡 Did you know, you can keep summaries lean with Finding overflow, which tucks the rest behind 'View more'

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

Previous review results

Review updated until commit c2dcad2 ⚖️ Balanced

Results up to commit 2a12c5f 🚀 Fast

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0) 🎨 UX issues (0) 📜 Skill insights (0)

1. Downgrade disables MOSAICO tracing ✓ Resolved 🐞 Bug ◔ Observability

Description

Pinning langfuse==2.59.7 removes the Langfuse client APIs used by MOSAICO (get_client,
propagate_attributes, and start_as_current_observation). langfuse_span() and client
initialization catch the resulting failures and continue, so MOSAICO requests run without the
intended Langfuse spans while only emitting a warning.

Code

requirements.txt[R37-39]

+# litellm 1.96.2 requires langfuse v2 (<3.0) for the native 'langfuse' callback to avoid TypeError with sdk_integration.
+# Version matches litellm's own constraint in pyproject.toml (>=2.59.7,<3.0).
+langfuse==2.59.7

Relevance

●●● Strong

Accepted MOSAICO Langfuse compatibility findings show the team prioritizes preserving tracing when
dependency/API mismatches silently disable instrumentation.

PR-#2560
PR-#2459

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

The changed dependency forces the v2 SDK, while the repository directly imports and invokes the
newer client interface. Both affected paths intentionally swallow exceptions and proceed untraced,
proving the dependency change causes an observability regression in MOSAICO.

pr_agent/mosaico/observability.py[63-79]
pr_agent/mosaico/server.py[65-75]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
The new global `langfuse==2.59.7` pin is compatible with LiteLLM's native callback but is incompatible with MOSAICO's direct instrumentation, which imports and uses the newer Langfuse client API. Because MOSAICO catches these exceptions, tracing is silently disabled rather than failing loudly.

## Issue Context
Preserve LiteLLM's requirement for a v2 client while ensuring MOSAICO retains functional tracing. Either update MOSAICO to the Langfuse 2.59.7 API and verify its trace-context behavior, or isolate the LiteLLM callback's dependency/API compatibility so both integrations can coexist. Add/adjust tests that exercise `_configure_langfuse()` and `langfuse_span()` with the pinned dependency.

## Fix Focus Areas
- requirements.txt[37-39]
- pr_agent/mosaico/observability.py[63-79]
- pr_agent/mosaico/server.py[65-75]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

[qodo-code-review]

2. Duplicate langfuse callbacks registered 🐞 Bug ≡ Correctness

Mapping each entry independently turns ["langfuse", "langfuse_otel"] into two langfuse_otel
entries, causing duplicate tracing callbacks and duplicate Langfuse events. The same defect affects
success, failure, and service callback lists.

Agent Prompt

## Issue description
Callback normalization replaces every legacy `langfuse` entry independently, so configurations containing both legacy and OTel callback names produce duplicate `langfuse_otel` registrations.

## Issue Context
Normalize each callback collection so non-Langfuse callbacks are preserved while legacy entries are removed and exactly one `langfuse_otel` entry remains. Apply this consistently to success, failure, and service callbacks, and add coverage for mixed and repeated entries.

## Fix Focus Areas
- pr_agent/algo/ai_handlers/litellm_ai_handler.py[151-164]
- tests/unittest/test_mosaico_env_bridge.py[89-108]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

[qodo-code-review]

Code review by qodo was updated up to the latest commit c2dcad2