Use the bugscache trigram index in search and cut N+1 queries in the job details panel

[camd]

Summary

Optimizes the database queries behind the job details / failure summary panel, where the bug_suggestions endpoint has been reported as slow. The main fix makes an already-shipped index actually get used; two smaller commits remove N+1 queries on the same path.

1. Bugscache.search() never used the trigram index (main fix)

search() filtered with summary__icontains, which Django compiles to UPPER(summary) LIKE UPPER(%s) on PostgreSQL. Because the predicate is on UPPER(summary) rather than the bare column, the planner cannot use the gin_trgm_ops index on bugscache.summary (added in #9555) and runs a full sequential scan on every call — and search() runs once per distinct error term in the bug suggestions panel.

A small TrigramILike lookup applies ILIKE directly to the column (same wildcard-escaping as __icontains), so the existing trigram GIN index serves the query. No migration required.

Measured on the prod replica (bugscache ≈ 52,800 rows / 106 MB)

EXPLAIN (ANALYZE, BUFFERS) for three representative terms:

Term	Before	After
test-verify (generic)	Seq Scan, 52,806 rows removed, 128 ms	Bitmap Index Scan, 5.8 ms
shutdownleaks (leak)	Seq Scan, 52,793 rows removed, 131 ms	Bitmap Index Scan, 5.3 ms
long test-path string	Seq Scan, whole table, 123 ms	Bitmap Index Scan, 64 ms

Shared buffers per call drop from ~5,724 to 76–702. The cost goes from O(table size) — which grows unbounded as bugs accumulate — to O(matches).

2. Batch the per-row occurrences count

Bugscache.serialize() ran jobmap.count() for every internal-issue row, i.e. one query per internal issue in a search result. Added internal_occurrences() to compute all counts in a single grouped query and pass them into serialize().

3. Remove N+1 in job note serialization

• JobNoteSerializer.failure_classification_id used SlugRelatedField(slug_field="id"), loading the whole related FailureClassification per note just to read its id (already present as the FK column). Switched to PrimaryKeyRelatedField (pk-only, no query).
• push_notes reads failure_classification.name, so its queryset now select_relateds the classification instead of fetching it per note.

Testing

• All existing Bugscache.search() parametrized tests pass, confirming behaviour parity (literal _/% handling and case-insensitivity are unchanged).
• Added a test asserting the compiled search SQL uses ILIKE on the bare column (no UPPER(...) wrapping) and escapes wildcards.
• Added a test for internal_occurrences verifying per-bug counts, window exclusion, and that it runs in a single query.
• error_summary, note, and job-note suites pass. ruff clean.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.68%. Comparing base (6a820e6) to head (85c36e4).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9689      +/-   ##
==========================================
+ Coverage   82.66%   82.68%   +0.02%     
==========================================
  Files         626      626              
  Lines       36753    36793      +40     
  Branches     3348     3333      -15     
==========================================
+ Hits        30383    30424      +41     
+ Misses       5988     5987       -1     
  Partials      382      382              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.