fix: add is_available and extract_params to MongoDB, ClickHouse, Kafka, and CloudWatch tools by devankitjuneja · Pull Request #732 · Tracer-Cloud/opensre

devankitjuneja · 2026-04-21T12:42:03Z

Fixes #730

Describe the changes you have made in this PR -

10 tools defaulted to _always_available, causing them to appear in the planner for every investigation regardless of pipeline context. In pipelines like rds-postgres-synthetic (Grafana-only), the LLM would select MongoDB, ClickHouse, Kafka, and CloudWatch tools, they would fail with TypeError: missing required argument, and the agent wasted investigation loops on dead-end actions.

Root cause: BaseTool.is_available() returns True by default. The requires=[...] field on @tool is documentation-only — not wired to the availability check.

Changes:

Added mongodb_is_available + mongodb_extract_params to app/integrations/mongodb.py
Added clickhouse_is_available + clickhouse_extract_params to app/integrations/clickhouse.py
Added kafka_is_available + kafka_extract_params to app/integrations/kafka.py
Added cloudwatch_is_available to app/tools/utils/availability.py (no extract_params — CloudWatch uses IAM auth, job_queue is alert-specific)
Wired is_available + extract_params into all 10 affected tool decorators

extract_params follows the established PostgreSQL/MongoDB pattern — credentials are auto-filled from resolved integrations so the LLM never needs to supply connection_string, host, or bootstrap_servers. ClickHouse and Kafka detect_sources.py wiring is tracked in a follow-up issue; the helpers are ready to activate once that lands.

Screenshots of the UI changes (If any) -

N/A

Code Understanding and AI Usage

Did you use AI assistance (ChatGPT, Claude, Copilot, etc.) to write any part of this code?

No, I wrote all the code myself
Yes, I used AI assistance (continue below)

If you used AI assistance:

I have reviewed every single line of the AI-generated code
I can explain the purpose and logic of each function/component I added
I have tested edge cases and understand how the code handles them
I have modified the AI output to follow this project's coding standards and conventions

Explain your implementation approach:

The planner filters tools via action.is_available(available_sources) before passing them to the LLM. Tools without a custom is_available always pass this filter. The fix adds named helper functions in each integration module (matching the pattern already used by PostgreSQL and MongoDB tools) and wires them into the @tool decorator. extract_params auto-fills connection credentials from available_sources so the LLM receives only alert-specific parameters it can reason about.

Checklist before requesting a review

I have added proper PR title and linked to the issue
I have performed a self-review of my code
I can explain the purpose of every function, class, and logic block I added
I understand why my changes work and have tested them thoroughly
I have considered potential edge cases and how my code handles them
If it is a core feature, I have added thorough tests
My code follows the project's style guidelines and conventions

Note: Please check Allow edits from maintainers if you would like us to assist in the PR.

…a, and CloudWatch tools (Tracer-Cloud#730) 10 tools defaulted to _always_available, causing them to appear in the planner for every pipeline regardless of context. The LLM would select them, they would fail with TypeError on missing required arguments, and the agent wasted investigation loops on dead-end actions. - Add mongodb_is_available + mongodb_extract_params to integrations/mongodb.py - Add clickhouse_is_available + clickhouse_extract_params to integrations/clickhouse.py - Add kafka_is_available + kafka_extract_params to integrations/kafka.py - Add cloudwatch_is_available to tools/utils/availability.py - Wire is_available + extract_params into all 10 affected tool decorators extract_params ensures credentials are auto-filled from resolved integrations so the LLM never needs to supply connection strings. ClickHouse and Kafka source blocks in detect_sources.py are tracked separately; the helpers are ready to activate once that wiring lands.

greptile-apps · 2026-04-21T12:45:31Z

Greptile Summary

This PR fixes a planner-pollution bug where 10 tools (MongoDB ×5, ClickHouse ×2, Kafka ×2, CloudWatch ×1) defaulted to BaseTool.is_available() → True, causing the LLM to select them even in pipelines that had none of those integrations configured, wasting investigation loops on dead-end TypeError failures.

Key changes:

mongodb_is_available gates on connection_verified (confirmed written by detect_sources.py at line 1099 for MongoDB integrations)
mongodb_database_is_available added for MongoDBProfilerTool and MongoDBCollectionStatsTool, which require a database name — prevents those tools appearing when only a bare connection string is configured
clickhouse_is_available / kafka_is_available gate on connection_verified; per the PR description, these tools will remain hidden until the ClickHouse/Kafka detect_sources.py wiring lands in a follow-up, which is a safe interim state
cloudwatch_is_available correctly gates on the presence of the cloudwatch source key (not connection_verified, which IAM-based auth never writes) — tool activates only when cloudwatch_log_group is present in alert annotations
All three previous review concerns (MongoDB connection_verified inconsistency, profiler/collection-stats missing-database footgun, CloudWatch always-False) are addressed in this revision

Confidence Score: 4/5

Safe to merge; all three previously flagged issues are resolved and the implementation is consistent with established patterns.

The core bug is properly fixed. mongodb_is_available now uses connection_verified (matching ClickHouse/Kafka), the database-gated helpers correctly guard profiler/collection-stats tools, and CloudWatch uses key-presence (correct for IAM auth). The only open item is a minor robustness nit on the int() port cast in clickhouse_extract_params, which is a P2 style suggestion rather than a blocking issue.

app/integrations/clickhouse.py — minor port-cast robustness nit; all other files are clean.

Important Files Changed

Filename	Overview
app/integrations/clickhouse.py	Adds `clickhouse_is_available` (gates on `connection_verified`) and `clickhouse_extract_params`; minor robustness issue with unguarded `int()` port cast.
app/integrations/kafka.py	Adds `kafka_is_available` and `kafka_extract_params` following the established pattern; clean implementation.
app/integrations/mongodb.py	Adds `mongodb_is_available` (gates on `connection_verified`, which `detect_sources.py` does write at line 1099), `mongodb_database_is_available` for profiler/collection-stats tools, and `mongodb_extract_params`; previous thread concerns resolved.
app/tools/utils/availability.py	Adds `cloudwatch_is_available` gating on key presence (correct for IAM-based auth where `connection_verified` is never written); previous thread concern about always-False resolved.
app/tools/MongoDBProfilerTool/init.py	Correctly wires `mongodb_database_is_available` to prevent the tool appearing in the planner without a configured database.
app/tools/MongoDBCollectionStatsTool/init.py	Correctly wires `mongodb_database_is_available`; `collection` param remains LLM-supplied (alert-specific), consistent with other alert-specific params.
app/tools/CloudWatchBatchMetricsTool/init.py	Wires `cloudwatch_is_available`; no `extract_params` (correct, IAM auth, `job_queue` is alert-specific).
app/tools/KafkaConsumerGroupTool/init.py	Wires `kafka_is_available` and `kafka_extract_params`; `group_id` correctly left for LLM to supply.
app/tools/KafkaTopicHealthTool/init.py	Clean wiring of `kafka_is_available` and `kafka_extract_params`.
app/tools/ClickHouseQueryActivityTool/init.py	Clean wiring of `clickhouse_is_available` and `clickhouse_extract_params`.
app/tools/ClickHouseSystemHealthTool/init.py	Clean wiring of `clickhouse_is_available` and `clickhouse_extract_params`.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Planner: filter tools via is_available] --> B{sources dict}

    B --> C[sources.get 'mongodb']
    C --> D{connection_verified?}
    D -- No --> E[Tools hidden from planner]
    D -- Yes --> F{database present?}
    F -- No --> G[ServerStatus / CurrentOps / ReplicaStatus visible]
    F -- Yes --> H[All 5 MongoDB tools visible]

    B --> I[sources.get 'clickhouse']
    I --> J{connection_verified?\ndetect_sources wiring pending}
    J -- No --> K[Tools hidden until follow-up]
    J -- Yes --> L[ClickHouse tools visible]

    B --> M[sources.get 'kafka']
    M --> N{connection_verified?\ndetect_sources wiring pending}
    N -- No --> O[Tools hidden until follow-up]
    N -- Yes --> P[Kafka tools visible]

    B --> Q[sources.get 'cloudwatch']
    Q -- key absent --> R[Tool hidden]
    Q -- key present\nlog_group in annotations --> S[CloudWatchBatchMetricsTool visible\nLLM supplies job_queue]

_{Reviews (3): Last reviewed commit: "fix: cloudwatch_is_available should chec..." | Re-trigger Greptile}

…tion_verified and add database guard - mongodb_is_available now checks connection_verified (consistent with clickhouse_is_available and kafka_is_available; connection_verified is set by detect_sources when MongoDB is configured and validated) - Add mongodb_database_is_available for tools that require a database name (MongoDBProfilerTool, MongoDBCollectionStatsTool); without it those tools would appear in the planner and return an error when no database is configured, repeating the original bug in a narrower scope

devankitjuneja · 2026-04-21T12:56:50Z

@greptile-apps re-review

…nection_verified detect_sources.py populates sources["cloudwatch"] from alert annotations (cloudwatch_log_group) but never writes connection_verified — CloudWatch uses ambient IAM auth, not an API key. Checking connection_verified caused the tool to be permanently hidden even when CloudWatch context was present.

devankitjuneja · 2026-04-21T14:26:31Z

@greptile-apps re-review

VaibhavUpreti

awesome @devankitjuneja! 🚀 Good catch with the bug

greptile-apps Bot reviewed Apr 21, 2026

View reviewed changes

Comment thread app/integrations/mongodb.py Outdated

Comment thread app/integrations/mongodb.py

greptile-apps Bot reviewed Apr 21, 2026

View reviewed changes

Comment thread app/tools/utils/availability.py Outdated

VaibhavUpreti approved these changes Apr 21, 2026

View reviewed changes

VaibhavUpreti merged commit 0cafa99 into Tracer-Cloud:main Apr 21, 2026
7 checks passed

canetizen mentioned this pull request Apr 25, 2026

run_diagnostic_code is registered without is_available/extract_params, crashes on every dispatch #956

Closed

AarushSharmaa mentioned this pull request Apr 27, 2026

fix: stop run_diagnostic_code from crashing every investigation dispatch #988

Merged

13 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: add is_available and extract_params to MongoDB, ClickHouse, Kafka, and CloudWatch tools#732

fix: add is_available and extract_params to MongoDB, ClickHouse, Kafka, and CloudWatch tools#732
VaibhavUpreti merged 3 commits intoTracer-Cloud:mainfrom
devankitjuneja:fix/730-tool-is-available-leak

devankitjuneja commented Apr 21, 2026

Uh oh!

greptile-apps Bot commented Apr 21, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

devankitjuneja commented Apr 21, 2026

Uh oh!

Uh oh!

devankitjuneja commented Apr 21, 2026

Uh oh!

VaibhavUpreti left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

devankitjuneja commented Apr 21, 2026

Describe the changes you have made in this PR -

Screenshots of the UI changes (If any) -

Code Understanding and AI Usage

Checklist before requesting a review

Uh oh!

greptile-apps Bot commented Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Flowchart

Uh oh!

Uh oh!

Uh oh!

devankitjuneja commented Apr 21, 2026

Uh oh!

Uh oh!

devankitjuneja commented Apr 21, 2026

Uh oh!

VaibhavUpreti left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

greptile-apps Bot commented Apr 21, 2026 •

edited

Loading