MuseHub — Database Query Runbook
Run these EXPLAIN ANALYZE checks before each production deployment and after
any schema or index change. No sequential scans on tables expected to hold
more than ~10 k rows.
How to run
# Connect to the production DB:
psql $DATABASE_URL
# For each query below, prepend EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT):
EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT)
SELECT ...
Look for:
Seq Scanon large tables — add an index if presentRows Removed by Filter>>Rows— index selectivity is pooractual time=values > 100 ms — investigate or add a covering index
Top-10 highest-traffic queries
Q1 — Repo lookup by URL slug (every page load)
SELECT * FROM musehub_repos
WHERE owner = 'gabriel' AND slug = 'neo-soul'
AND deleted_at IS NULL;
Expected plan: Index Scan using uq_musehub_repos_owner_slug
Q2 — List repos for a user (dashboard)
SELECT * FROM musehub_repos
WHERE owner = 'gabriel' AND visibility = 'public'
AND deleted_at IS NULL
ORDER BY pushed_at DESC
LIMIT 20;
Expected plan: Index Scan using ix_musehub_repos_owner_visibility
Q3 — HEAD commit for a branch (every push/fetch)
SELECT * FROM musehub_commits
WHERE repo_id = '<repo_uuid>' AND branch = 'main'
ORDER BY timestamp DESC
LIMIT 1;
Expected plan: Index Scan using ix_musehub_commits_repo_branch
Q4 — Open issue list (issue tracker)
SELECT * FROM musehub_issues
WHERE repo_id = '<repo_uuid>' AND state = 'open'
ORDER BY created_at DESC
LIMIT 25;
Expected plan: Index Scan using ix_musehub_issues_repo_state
Q5 — Issue by number (direct URL)
SELECT * FROM musehub_issues
WHERE repo_id = '<repo_uuid>' AND number = 42;
Expected plan: Index Scan using ix_musehub_issues_repo_number
Q6 — Open proposals (Proposal list)
SELECT * FROM musehub_proposals
WHERE repo_id = '<repo_uuid>' AND state = 'open'
ORDER BY created_at DESC
LIMIT 25;
Expected plan: Index Scan using ix_musehub_proposals_repo_state
Q7 — Intel result lookup (intel endpoints)
SELECT * FROM musehub_intel_results
WHERE repo_id = '<repo_uuid>' AND intel_type = 'code.symbol_history';
Expected plan: Index Scan using ix_musehub_intel_results_repo_type
SLOW QUERY threshold: 100 ms. Alert if actual time exceeds threshold.
Q8 — Per-repo storage quota sum (every push)
SELECT COALESCE(SUM(o.size_bytes), 0)
FROM musehub_objects o
JOIN musehub_object_refs r ON r.object_id = o.object_id
WHERE r.repo_id = '<repo_uuid>' AND o.deleted_at IS NULL;
Expected plan: Index Scan using pk_musehub_object_refs (repo_id prefix) +
Index Scan using musehub_objects pkey
Q9 — Recent commit feed (repo home page)
SELECT * FROM musehub_commits
WHERE repo_id = '<repo_uuid>'
ORDER BY timestamp DESC
LIMIT 10;
Expected plan: Index Scan using ix_musehub_commits_repo_timestamp
Q10 — Per-user storage sum (MCP push quota)
SELECT COALESCE(SUM(o.size_bytes), 0)
FROM musehub_objects o
JOIN musehub_object_refs ref ON ref.object_id = o.object_id
JOIN musehub_repos r ON r.repo_id = ref.repo_id
WHERE r.owner_user_id = '<user_uuid>'
AND r.deleted_at IS NULL;
Expected plan: Index Scan using ix_musehub_repos_owner_user_id +
Index Scan using pk_musehub_object_refs + Index Scan using musehub_objects pkey
Slow query log
Queries exceeding SLOW_QUERY_THRESHOLD_MS (default 100 ms) are logged at
WARNING level by the SQLAlchemy after_cursor_execute listener in
musehub/db/database.py.
To monitor in production:
# Tail the app log for SLOW QUERY entries:
journalctl -u musehub -f | grep "SLOW QUERY"
# Or set log_min_duration_statement in Postgres for server-side capture:
ALTER SYSTEM SET log_min_duration_statement = 100;
SELECT pg_reload_conf();
Alerts should be wired to any line matching SLOW QUERY in the application
log, or to pg_stat_statements entries with mean_exec_time > 100.