Repository navigation
feat: add MLflow tracing to all agent templates - #34
Merged
mpk-droid merged 9 commits intoMar 25, 2026
Merged
Conversation
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdds MLflow tracing: new tracing modules, startup calls to enable_tracing(), conditional wrapping of tool/agent callables for MLflow spans, updates to environment templates, and expanded README tracing documentation for Local and OpenShift deployments. Changes
Sequence Diagram(s)mermaid Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes 🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 7
🧹 Nitpick comments (3)
agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.py (1)
69-69: Unused attribute_mlflow_enabled.This attribute is set but never referenced in this file. Consider removing it if it's not needed elsewhere.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.py` at line 69, The attribute self._mlflow_enabled is assigned using getenv("MLFLOW_TRACKING_URI") in the constructor but never used elsewhere; either remove this unused attribute assignment or wire it into logic that needs it (e.g., replace other MLflow checks with self._mlflow_enabled or guard MLflow-related behavior using this flag). Locate the assignment to self._mlflow_enabled in the class constructor in agent.py and delete that line if not needed, or update any MLflow-related methods to reference self._mlflow_enabled consistently.agents/crewai/websearch_agent/pyproject.toml (1)
11-11: Consider adding a version constraint for stability.Removing all version constraints from
crewai[litellm]may lead to unexpected breakages when upstream releases breaking changes. Consider specifying at least a minimum version (e.g.,crewai[litellm]>=1.10.0) to document the tested baseline.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/crewai/websearch_agent/pyproject.toml` at line 11, The pyproject dependency "crewai[litellm]" lacks a version constraint which can cause breakages; update the dependency string in agents/crewai/websearch_agent/pyproject.toml (the "crewai[litellm]" entry) to include a minimum tested version (for example "crewai[litellm]>=1.10.0") or another appropriate version range to pin a stable baseline.agents/llamaindex/websearch_agent/README.md (1)
48-53: Add language specifiers to fenced code blocks.The environment variable configuration blocks are missing language specifiers. Adding
bashorinihelps with syntax highlighting and satisfies markdown linting.📝 Proposed fix
-``` +```bash MLFLOW_TRACKING_URI="http://localhost:5000"-``` +```bash MLFLOW_TRACKING_URI="https://<openshift-dashboard-url>/mlflow"Also applies to: 85-91
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/llamaindex/websearch_agent/README.md` around lines 48 - 53, Add language specifiers (e.g., bash or ini) to the fenced code blocks that show environment variables in agents/llamaindex/websearch_agent/README.md so markdown linting and syntax highlighting work; update the three backtick fences surrounding the MLFLOW_* blocks (the blocks around the sample env lines at the shown sections, including the second block at lines ~85-91) to start with ```bash (or ```ini) instead of just ```, ensuring each opening fence includes the language token and the closing fence remains ```; no other content changes are required.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py`:
- Around line 38-44: The warning log in tracing.py uses response.reason_phrase
which doesn't exist on requests.Response; update the log to use response.reason
instead (keep response.status_code, response.text[:500], mlflow_url and logger
as-is) so the logger.warning call references response.reason instead of
response.reason_phrase.
- Around line 99-107: The except block in tracing.py is catching HTTPException
but check_mlflow_health raises RuntimeError, so the handler is unreachable;
update the exception handler to catch RuntimeError (e.g., change "except
HTTPException as e" to "except RuntimeError as e") and ensure you use the caught
exception (e) when logging and when re-raising the RuntimeError with the
existing message; locate this in the block that calls check_mlflow_health and
updates logger.info/logger.warning.
In `@agents/langgraph/react_agent/src/react_agent/tracing.py`:
- Around line 20-54: check_mlflow_health currently uses raw requests and misses
MLflow client settings (MLFLOW_TRACKING_TOKEN, MLFLOW_TRACKING_INSECURE_TLS,
MLFLOW_TRACKING_AUTH, MLFLOW_HTTP_REQUEST_TIMEOUT,
MLFLOW_HTTP_REQUEST_MAX_RETRIES) and references response.reason_phrase
(invalid); update it to use the MLflow client or replicate its settings: create
an mlflow.tracking.MlflowClient (or configure requests.Session with token/auth,
verify flag from MLFLOW_TRACKING_INSECURE_TLS, timeout from
MLFLOW_HTTP_REQUEST_TIMEOUT and retries from MLFLOW_HTTP_REQUEST_MAX_RETRIES) to
call the server health endpoint, replace response.reason_phrase with
response.reason, and keep the same retry loop and logging in check_mlflow_health
so the probe behaves identically to the real MLflow client.
In `@agents/llamaindex/websearch_agent/src/websearch_agent/tracing.py`:
- Around line 38-44: The logger.warning call is using the non-existent attribute
response.reason_phrase which raises AttributeError; update the log to use
response.reason instead (leave response.status_code, response.text[:500], and
mlflow_url as-is) so the warning message accesses the correct requests.Response
attribute (locate the logger.warning block that references
response.reason_phrase and replace that symbol with response.reason).
- Around line 77-85: The except block is catching HTTPException although
check_mlflow_health() raises RuntimeError, so replace the except HTTPException
as e handler with except RuntimeError as e to ensure the handler runs; update
the logger.warning to include the original error info and re-raise the
RuntimeError with the wrapped message using exception chaining (raise
RuntimeError(...) from e) so the original traceback is preserved; refer to the
check_mlflow_health call and the surrounding logger.info/logger.warning and the
existing raise RuntimeError block when making the change.
In
`@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.py`:
- Around line 111-116: The _messages_to_responses_input function currently never
sets instructions and treats system messages as input_text; update it to detect
messages with role "system", set the instructions variable to that message's
content (use m.get("content","") or "" as a fallback), skip adding that message
to input_items, and continue processing other messages (mapping "assistant" to
content_type "output_text" and all other roles to "input_text"); return the
instructions string and the constructed input_items list.
In
`@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py`:
- Around line 20-54: The check_mlflow_health function currently uses raw
requests and response.reason_phrase which breaks configured MLflow
tracing/auth/tls settings and raises AttributeError; update check_mlflow_health
to use the MLflow client (mlflow.tracking.MlflowClient) by setting
mlflow.set_tracking_uri(mlflow_tracking_uri) and invoking the client to probe
the /health endpoint (or use the client's internal endpoint caller) so all
MLFLOW_* settings (MLFLOW_TRACKING_TOKEN, MLFLOW_TRACKING_INSECURE_TLS,
MLFLOW_TRACKING_AUTH, MLFLOW_HTTP_REQUEST_TIMEOUT,
MLFLOW_HTTP_REQUEST_MAX_RETRIES) are honored, and replace any use of
response.reason_phrase with response.reason if you keep any raw requests usage.
---
Nitpick comments:
In `@agents/crewai/websearch_agent/pyproject.toml`:
- Line 11: The pyproject dependency "crewai[litellm]" lacks a version constraint
which can cause breakages; update the dependency string in
agents/crewai/websearch_agent/pyproject.toml (the "crewai[litellm]" entry) to
include a minimum tested version (for example "crewai[litellm]>=1.10.0") or
another appropriate version range to pin a stable baseline.
In `@agents/llamaindex/websearch_agent/README.md`:
- Around line 48-53: Add language specifiers (e.g., bash or ini) to the fenced
code blocks that show environment variables in
agents/llamaindex/websearch_agent/README.md so markdown linting and syntax
highlighting work; update the three backtick fences surrounding the MLFLOW_*
blocks (the blocks around the sample env lines at the shown sections, including
the second block at lines ~85-91) to start with ```bash (or ```ini) instead of
just ```, ensuring each opening fence includes the language token and the
closing fence remains ```; no other content changes are required.
In
`@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.py`:
- Line 69: The attribute self._mlflow_enabled is assigned using
getenv("MLFLOW_TRACKING_URI") in the constructor but never used elsewhere;
either remove this unused attribute assignment or wire it into logic that needs
it (e.g., replace other MLflow checks with self._mlflow_enabled or guard
MLflow-related behavior using this flag). Locate the assignment to
self._mlflow_enabled in the class constructor in agent.py and delete that line
if not needed, or update any MLflow-related methods to reference
self._mlflow_enabled consistently.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: bd40c6ff-5599-4e59-9d2b-5e4a1aae9857
📒 Files selected for processing (20)
.gitignoreagents/crewai/websearch_agent/README.mdagents/crewai/websearch_agent/main.pyagents/crewai/websearch_agent/pyproject.tomlagents/crewai/websearch_agent/src/crewai_web_search/crew.pyagents/crewai/websearch_agent/src/crewai_web_search/tracing.pyagents/crewai/websearch_agent/template.envagents/langgraph/react_agent/README.mdagents/langgraph/react_agent/main.pyagents/langgraph/react_agent/src/react_agent/tracing.pyagents/langgraph/react_agent/template.envagents/llamaindex/websearch_agent/README.mdagents/llamaindex/websearch_agent/main.pyagents/llamaindex/websearch_agent/src/websearch_agent/tracing.pyagents/llamaindex/websearch_agent/template.envagents/vanilla_python/openai_responses_agent/README.mdagents/vanilla_python/openai_responses_agent/main.pyagents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.pyagents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.pyagents/vanilla_python/openai_responses_agent/template.env
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (3)
agents/langgraph/react_agent/src/react_agent/tracing.py (1)
79-84: Preserve the original exception cause on startup failure.At Line 81, the new
RuntimeErroris raised without chaining, which hides the original stack context fromcheck_mlflow_health().Proposed fix
except RuntimeError as e: logger.warning(f"[Tracing] MLflow server is unreachable at {tracking_uri}") raise RuntimeError( f"MLFLOW_TRACKING_URI is set but server is unreachable: {tracking_uri}. " f"Start the server or check the URI. Error: {e}" - ) + ) from e🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/langgraph/react_agent/src/react_agent/tracing.py` around lines 79 - 84, In the except RuntimeError block inside tracing.py (handling check_mlflow_health/tracking URI), preserve the original exception by raising the new RuntimeError using exception chaining (raise RuntimeError(... ) from e) instead of raising it plain; update the raise in the except block that currently logs logger.warning(f"[Tracing] MLflow server is unreachable at {tracking_uri}") to include "from e" so the original stack context is retained for debugging.agents/crewai/websearch_agent/src/crewai_web_search/tracing.py (1)
19-146: Extract shared MLflow bootstrap logic to a common tracing utility.
check_mlflow_health+ startup wiring is duplicated across multiple templates, which increases drift risk and patch overhead. Consider centralizing shared logic and keeping only framework-specificautolog()/span wiring per template.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py` around lines 19 - 146, The check_mlflow_health, enable_tracing, and wrap_func_with_mlflow_trace logic is duplicated across templates — extract the shared MLflow bootstrap (health check, tracking URI handling, mlflow.set_tracking_uri, set_experiment, enable_async_logging, and provider autolog selection logic) into a common utility function/class (e.g., tracing.bootstrap or tracing.enable_mlflow) and replace the duplicated blocks in this file with calls to that utility; keep only template-specific wiring here (e.g., mlflow.crewai.autolog() or any framework-specific autolog calls) and ensure you reference the original symbols when refactoring: check_mlflow_health, enable_tracing, wrap_func_with_mlflow_trace, provider_autolog_map, and mlflow.crewai.autolog to guide where to move code and what to retain locally.agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py (1)
100-105: Chain the original exception when re-raising.At Line 102, use
from eso the root startup failure cause is preserved.Proposed fix
except RuntimeError as e: logger.warning(f"[Tracing] MLflow server is unreachable at {tracking_uri}") raise RuntimeError( f"MLFLOW_TRACKING_URI is set but server is unreachable: {tracking_uri}. " f"Start the server or check the URI. Error: {e}" - ) + ) from e🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py` around lines 100 - 105, In the except RuntimeError block in tracing.py (the handler that logs MLflow server unreachability using logger and tracking_uri), re-raise the new RuntimeError using exception chaining so the original exception is preserved—replace the current "raise RuntimeError(...)" with "raise RuntimeError(... ) from e" so the original RuntimeError (variable e) is attached as the __cause__.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py`:
- Around line 55-73: The function wrap_func_with_mlflow_trace shadows the
builtin type() and misdeclares name's type; rename the parameter type to
span_type (or similar) and update any internal references from type to
span_type, change the signature to use name: Optional[str] = None (and import
Optional if not already), and update all call sites of
wrap_func_with_mlflow_trace to pass the renamed argument
(span_type="tool"/"agent") so the comparisons and mlflow.trace calls use
span_type instead of the builtin name.
In
`@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py`:
- Around line 56-74: Rename the parameter in wrap_func_with_mlflow_trace from
type to span_type to avoid shadowing the builtin and change its annotation to
Literal["tool","agent"] so only valid span values are accepted; update all
internal references (the comparisons and error message) to use span_type and add
a static type hint import from typing for Literal, ensuring the function
signature and control flow only allow "tool" or "agent" (raise ValueError for
anything else).
---
Nitpick comments:
In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py`:
- Around line 19-146: The check_mlflow_health, enable_tracing, and
wrap_func_with_mlflow_trace logic is duplicated across templates — extract the
shared MLflow bootstrap (health check, tracking URI handling,
mlflow.set_tracking_uri, set_experiment, enable_async_logging, and provider
autolog selection logic) into a common utility function/class (e.g.,
tracing.bootstrap or tracing.enable_mlflow) and replace the duplicated blocks in
this file with calls to that utility; keep only template-specific wiring here
(e.g., mlflow.crewai.autolog() or any framework-specific autolog calls) and
ensure you reference the original symbols when refactoring: check_mlflow_health,
enable_tracing, wrap_func_with_mlflow_trace, provider_autolog_map, and
mlflow.crewai.autolog to guide where to move code and what to retain locally.
In `@agents/langgraph/react_agent/src/react_agent/tracing.py`:
- Around line 79-84: In the except RuntimeError block inside tracing.py
(handling check_mlflow_health/tracking URI), preserve the original exception by
raising the new RuntimeError using exception chaining (raise RuntimeError(... )
from e) instead of raising it plain; update the raise in the except block that
currently logs logger.warning(f"[Tracing] MLflow server is unreachable at
{tracking_uri}") to include "from e" so the original stack context is retained
for debugging.
In
`@agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py`:
- Around line 100-105: In the except RuntimeError block in tracing.py (the
handler that logs MLflow server unreachability using logger and tracking_uri),
re-raise the new RuntimeError using exception chaining so the original exception
is preserved—replace the current "raise RuntimeError(...)" with "raise
RuntimeError(... ) from e" so the original RuntimeError (variable e) is attached
as the __cause__.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: a20d39b4-fdfa-4566-b8aa-22ab5e68df6f
📒 Files selected for processing (9)
agents/crewai/websearch_agent/README.mdagents/crewai/websearch_agent/src/crewai_web_search/tracing.pyagents/langgraph/react_agent/README.mdagents/langgraph/react_agent/src/react_agent/tracing.pyagents/llamaindex/websearch_agent/README.mdagents/llamaindex/websearch_agent/src/websearch_agent/tracing.pyagents/vanilla_python/openai_responses_agent/README.mdagents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.pyagents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py
✅ Files skipped from review due to trivial changes (4)
- agents/vanilla_python/openai_responses_agent/README.md
- agents/langgraph/react_agent/README.md
- agents/crewai/websearch_agent/README.md
- agents/llamaindex/websearch_agent/README.md
🚧 Files skipped from review as they are similar to previous changes (1)
- agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/agent.py
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
agents/crewai/websearch_agent/src/crewai_web_search/tracing.py (1)
101-106: Chain the exception for better tracebacks.When re-raising an exception within an
exceptblock, useraise ... from eto preserve the exception chain for debugging.♻️ Proposed fix
except RuntimeError as e: logger.warning(f"[Tracing] MLflow server is unreachable at {tracking_uri}") raise RuntimeError( f"MLFLOW_TRACKING_URI is set but server is unreachable: {tracking_uri}. " f"Start the server or check the URI. Error: {e}" - ) + ) from e🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py` around lines 101 - 106, The except block in tracing.py that catches "RuntimeError as e" and re-raises a new RuntimeError should chain the original exception to preserve tracebacks; update the raise in that except handler to use "raise RuntimeError(... ) from e" (referencing the existing tracking_uri and e variables) so the original exception is linked to the new one.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@agents/crewai/websearch_agent/src/crewai_web_search/tracing.py`:
- Around line 101-106: The except block in tracing.py that catches "RuntimeError
as e" and re-raises a new RuntimeError should chain the original exception to
preserve tracebacks; update the raise in that except handler to use "raise
RuntimeError(... ) from e" (referencing the existing tracking_uri and e
variables) so the original exception is linked to the new one.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 588457c6-c08e-44f3-b297-923f8f410ab4
📒 Files selected for processing (2)
agents/crewai/websearch_agent/src/crewai_web_search/crew.pyagents/crewai/websearch_agent/src/crewai_web_search/tracing.py
🚧 Files skipped from review as they are similar to previous changes (1)
- agents/crewai/websearch_agent/src/crewai_web_search/crew.py
Wojciech-Rebisz
approved these changes
Mar 24, 2026
mpk-droid
reviewed
Mar 24, 2026
mpk-droid
left a comment
mpk-droid
left a comment
Contributor
There was a problem hiding this comment.
Overall looks good. Added few comments. Could you also fix "major" coderabbit comments?
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@agents/langgraph/react_agent/src/react_agent/tracing.py`:
- Around line 29-33: Replace use of time.time() and hardcoded timeout=5 in the
health-check loop with a monotonic timer and a per-request remaining budget:
initialize start_time using time.monotonic(), compute remaining = max_wait_time
- (time.monotonic() - start_time) before each iteration, pass timeout=min(5,
remaining) to requests.get(mlflow_url, timeout=...), and when sleeping use
sleep_duration = min(backoff_seconds, remaining) (or skip sleeping if remaining
<= 0) so neither the request nor the sleep can exceed the configured
max_wait_time; apply these changes in the loop that contains start_time, the
while True block, and the requests.get call.
- Around line 72-85: The mlflow imports and timeout parsing in tracing.py can
raise ModuleNotFoundError or ValueError and must be guarded; wrap the "import
mlflow" and "import mlflow.langchain" in a try/except ModuleNotFoundError and
set a flag (e.g., mlflow_available) to False when missing so the tracing block
skips MLflow logic, and when reading MLFLOW_HEALTH_CHECK_TIMEOUT use a safe
parse with a fallback (e.g., default 5) catching ValueError before calling
check_mlflow_health(tracking_uri=tracking_uri,
max_wait_time=health_check_timeout); keep existing RuntimeError handling for
check_mlflow_health and log appropriate messages via logger so startup
gracefully degrades when mlflow is absent or the env var is malformed.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 33ce0614-28a1-4f57-b4ff-466c5926282d
📒 Files selected for processing (4)
agents/crewai/websearch_agent/src/crewai_web_search/tracing.pyagents/langgraph/react_agent/src/react_agent/tracing.pyagents/llamaindex/websearch_agent/src/websearch_agent/tracing.pyagents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py
🚧 Files skipped from review as they are similar to previous changes (2)
- agents/crewai/websearch_agent/src/crewai_web_search/tracing.py
- agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py
mpk-droid
requested changes
Mar 24, 2026
mpk-droid
left a comment
mpk-droid
left a comment
Contributor
There was a problem hiding this comment.
thanks for fixes. couple more comments.
… Responses agent templates Add optional MLflow tracing support to 4 agent templates. Tracing is enabled when MLFLOW_TRACKING_URI is set, with health checks and graceful fallback. - Add tracing.py with health check, enable_tracing(), and framework-specific autolog for each agent - Add enable_tracing() call in each main.py lifespan function - Add MLflow env vars (local + OpenShift) to all template.env files - Add tracing setup docs to all README files - CrewAI: provider-aware LLM autolog via LLM_PROVIDER env var, pin crewai[litellm]==1.9.2 - OpenAI Responses: wrap tools/agent with MLflow spans, fix next(reader) StopIteration crash, fix _messages_to_responses_input content type for assistant messages - Add *.db, CLAUDE.md, agents/demo/ to .gitignore Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add wrap_func_with_mlflow_trace to CrewAI tracing.py for TOOL spans - Wrap CrewAI tool _run methods with MLflow tracing in crew.py - Unpin crewai[litellm] version (was ==1.9.2) to fix streaming+tools bug - Add wrap_func_with_mlflow_trace to OpenAI Responses streaming path - Remove stale `from anyio import current_time` import in OpenAI agent Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix response.reason_phrase → response.reason (requests API) - Fix except HTTPException → except RuntimeError in health check error handling - Fix e.detail → str(e) for RuntimeError (no .detail attribute) - Remove unused HTTPException import from all tracing files - Extract system messages to instructions param in OpenAI Responses agent - Remove unused self._mlflow_enabled attribute - Add crewai[litellm]>=1.10.1 version constraint - Add ini language specifiers to README code blocks Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…rsions mlflow.crewai.autolog() does not capture tool spans in CrewAI >=1.10. Manual wrapping via wrap_func_with_mlflow_trace is needed until this is fixed. Comments warn to remove manual wrapping if autolog is fixed in the future to avoid duplicate spans. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…OW_HEALTH_CHECK_TIMEOUT env var - Rename `type` to `span_type` in wrap_func_with_mlflow_trace to avoid shadowing Python builtin, use Literal["tool", "agent"] for type safety - Fix `name: str = None` to `Optional[str] = None` in CrewAI tracing - Add MLFLOW_HEALTH_CHECK_TIMEOUT to template.env files (default 5s) - Update all call sites to use span_type= keyword Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of crashing the app, log a warning with connection timeout duration and continue without tracing when MLflow is unavailable. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…v var - Use min(5, remaining) for per-request timeout so it never exceeds the configured MLFLOW_HEALTH_CHECK_TIMEOUT budget - Add ValueError guard for MLFLOW_HEALTH_CHECK_TIMEOUT parsing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…w version Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sanafayyaz315
force-pushed
the
mlflow-integration-v2
branch
from
March 24, 2026 13:54
8c55e3c to
f98b24e
Compare
mpk-droid
approved these changes
Mar 24, 2026
mpk-droid
left a comment
mpk-droid
left a comment
Contributor
There was a problem hiding this comment.
LGTM. Thanks for pushing all the fixes!
tarun-etikala
approved these changes
Mar 24, 2026
tarun-etikala
referenced
this pull request
in tarun-etikala/agentic-starter-kits
Apr 16, 2026
- Return HTTP 503 from /health when agent is uninitialized so K8s readiness probes correctly gate traffic (8 agents, audit #5) - Bind local dev servers to 127.0.0.1 to prevent DNS rebinding attacks; container Dockerfiles remain at 0.0.0.0 (audit #36) - Standardize FastAPI to >=0.135.1 across all agents (audit #33) - Bump uvicorn in mcp_automl_template to >=0.41.0 (audit #34) - Remove env:init dependency in react_with_database_memory (audit red-hat-data-services#46) - Update post-helm-migration-audit.md (14 resolved, 4 accepted) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3 tasks
tarun-etikala
referenced
this pull request
in tarun-etikala/agentic-starter-kits
Apr 16, 2026
- Return HTTP 503 from /health when agent is uninitialized so K8s readiness probes correctly gate traffic (8 agents, audit #5) - Bind local dev servers to 127.0.0.1 to prevent DNS rebinding attacks; container Dockerfiles remain at 0.0.0.0 (audit #36) - Standardize FastAPI to >=0.135.1 across all agents (audit #33) - Bump uvicorn in mcp_automl_template to >=0.41.0 (audit #34) - Remove env:init dependency in react_with_database_memory (audit red-hat-data-services#46) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
tarun-etikala
referenced
this pull request
in tarun-etikala/agentic-starter-kits
Apr 20, 2026
- Return HTTP 503 from /health when agent is uninitialized so K8s readiness probes correctly gate traffic (8 agents, audit #5) - Bind local dev servers to 127.0.0.1 to prevent DNS rebinding attacks; container Dockerfiles remain at 0.0.0.0 (audit #36) - Standardize FastAPI to >=0.135.1 across all agents (audit #33) - Bump uvicorn in mcp_automl_template to >=0.41.0 (audit #34) - Remove env:init dependency in react_with_database_memory (audit red-hat-data-services#46) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
tarun-etikala
referenced
this pull request
in tarun-etikala/agentic-starter-kits
Apr 20, 2026
- Return HTTP 503 from /health when agent is uninitialized so K8s readiness probes correctly gate traffic (8 agents, audit #5) - Bind local dev servers to 127.0.0.1 to prevent DNS rebinding attacks; container Dockerfiles remain at 0.0.0.0 (audit #36) - Standardize FastAPI to >=0.135.1 across all agents (audit #33) - Bump uvicorn in mcp_automl_template to >=0.41.0 (audit #34) - Remove env:init dependency in react_with_database_memory (audit red-hat-data-services#46) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add optional MLflow tracing integration to all 4 agent templates
(CrewAI, LangGraph, LlamaIndex, OpenAI Responses).
What's included
1. MLflow tracing integration for all 4 frameworks
mlflow.crewai.autolog()+ provider-specific autolog(openai)
mlflow.langchain.autolog()mlflow.llama_index.autolog()mlflow.openai.autolog()+ manualwrap_func_with_mlflow_tracefor agent/tool spans2. Per-template changes
tracing.py(new) —enable_tracing(),check_mlflow_health(),conditional autolog
main.py— callenable_tracing()in lifespantemplate.env— MLflow config vars (local + OpenShift cluster)README.md— MLflow setup instructions3. Fixes
wrap_func_with_mlflow_traceon tool_runfor explicitTOOL spans
4. Design decisions
pyproject.toml— userinstalls manually
MLFLOW_TRACKING_URIstartup (fail-fast)