feat(code-review): Add CodeReviewRun model to track check run lifecycle

[armenzg]

Summary

Adds a CodeReviewRun database model to persist the state of each code review run, enabling visibility into stuck or failed GitHub check runs.

Model (sentry_codereviewrun table):

• Tracks organization_id, repository_id, pull_request_number, commit_sha, github_delivery_id
• Status lifecycle: task_enqueued → seer_request_sent → seer_request_succeeded / seer_request_failed
• Records seer_response_status (HTTP code) and error_message on failure
• Indexed on organization_id, repository_id, github_delivery_id, date_added for efficient querying

Pipeline wiring:

• schedule_task() accepts pull_request_number and github_delivery_id, creates a CodeReviewRun with TASK_ENQUEUED status, and passes code_review_run_id to the Celery task
• pull_request.py and issue_comment.py pass the new params
• Celery task updates status at each stage; retryable errors only mark as failed on the final attempt (not mid-retry)
• Also consolidates sentry_sdk.set_tags + log_seer_request into a single _set_tags_and_log() function using Seer-consistent tag names (scm_provider, scm_owner, pr_id, etc.)

Retention cleanup:

• New periodic task cleanup_old_code_review_runs runs every 6 hours
• Deletes rows with date_added older than 90 days using bulk_delete_objects (10k-row batches) to avoid table locks

Test plan

• CreateCodeReviewRunTest - creates records, handles missing fields, survives DB errors gracefully
• UpdateCodeReviewRunTest - updates status/error_message, no-op on None ID, survives DB errors
• ProcessGitHubWebhookEventRunTrackingTest - lifecycle: success, client error, retryable final, retryable mid-retry, no run_id
• CleanupOldCodeReviewRunsTest - deletes old rows, retains recent rows
• All 159 existing code review tests pass

Made with Cursor

[github-actions]

This PR has a migration; here is the generated SQL for src/sentry/migrations/1031_add_codereviewrun_table.py

for 1031_add_codereviewrun_table in sentry

--
-- Create model CodeReviewRun
--
CREATE TABLE "sentry_codereviewrun" ("id" bigint NOT NULL PRIMARY KEY GENERATED BY DEFAULT AS IDENTITY, "organization_id" bigint NOT NULL, "repository_id" bigint NOT NULL, "pull_request_number" integer NOT NULL, "commit_sha" varchar(64) NOT NULL, "github_delivery_id" varchar(64) NOT NULL, "status" varchar(32) NOT NULL, "seer_response_status" integer NULL, "error_message" text NULL, "date_added" timestamp with time zone DEFAULT (STATEMENT_TIMESTAMP()) NOT NULL, "date_updated" timestamp with time zone DEFAULT (STATEMENT_TIMESTAMP()) NOT NULL);
CREATE INDEX CONCURRENTLY "sentry_codereviewrun_organization_id_3abec94a" ON "sentry_codereviewrun" ("organization_id");
CREATE INDEX CONCURRENTLY "sentry_codereviewrun_repository_id_b785049c" ON "sentry_codereviewrun" ("repository_id");
CREATE INDEX CONCURRENTLY "sentry_codereviewrun_github_delivery_id_e1ef8e8d" ON "sentry_codereviewrun" ("github_delivery_id");
CREATE INDEX CONCURRENTLY "sentry_codereviewrun_github_delivery_id_e1ef8e8d_like" ON "sentry_codereviewrun" ("github_delivery_id" varchar_pattern_ops);
CREATE INDEX CONCURRENTLY "sentry_codereviewrun_date_added_925ef389" ON "sentry_codereviewrun" ("date_added");

[cursor]

Cursor Bugbot has reviewed your changes and found 3 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Cleanup task schedule has wrong namespace and name

High Severity

The schedule config references "seer:sentry.seer.code_review.tasks.cleanup.cleanup_old_code_review_runs" but the task is registered with namespace "seer.code_review" (not "seer") and name "sentry.seer.code_review.tasks.cleanup_old_code_review_runs" (without the extra .cleanup. segment). The scheduler splits on : and calls taskregistry.get_task(namespace, taskname), which will fail to resolve the task since both the namespace and task name are wrong. The cleanup job will never run, causing the sentry_codereviewrun table to grow without bound.

Additional Locations (1)

• src/sentry/seer/code_review/tasks/cleanup.py#L16-L17

 

[cursor]

date_updated never updated by QuerySet update

Medium Severity

_update_code_review_run uses CodeReviewRun.objects.filter(...).update(...) which is Django's QuerySet.update(). This bypasses the auto_now=True behavior on the date_updated field, so it will retain its creation-time value forever. The codebase provides an instance-level update() method (in sentry.db.models.query) that correctly handles auto_now fields, but it isn't used here. Since this model is designed for monitoring stuck runs, an inaccurate date_updated undermines that visibility.

 

[cursor]

Cleanup task module missing from TASKWORKER_IMPORTS

High Severity

The cleanup_old_code_review_runs task module (sentry.seer.code_review.tasks.cleanup) is not listed in TASKWORKER_IMPORTS. Taskworkers discover tasks by importing modules from that list at startup, so the @instrumented_task decorator on cleanup_old_code_review_runs never executes and the task is never registered. This is independent of the previously reported schedule namespace/name mismatch — both issues must be fixed for the cleanup job to run. Without it, the sentry_codereviewrun table grows unboundedly.

Additional Locations (1)

• src/sentry/seer/code_review/tasks/cleanup.py#L15-L21

 

[giovanni-guidini]

There's no mention in the PR about a follow up deleting the stuff we currently have for this (we can track the status creation and completion via run_state.value and we do have a scheduled task that marks runs as "timedout" and has a side effect of also marking the status timed out.

Just confirming that removing that stuff is being tracked too?

[armenzg]

I see @vaind building something similar here: #108531