id
stringclasses
6 values
title
stringclasses
6 values
category
stringclasses
6 values
expected_detector
stringclasses
5 values
expected_root_cause
stringclasses
6 values
fix_target
stringclasses
3 values
naive_answer
stringclasses
6 values
naive_answer_correct
bool
2 classes
telemetry
dict
scenario_description
stringclasses
5 values
healthy_baseline
Healthy database
none
null
Nothing is wrong. The correct answer is no finding.
null
find something anyway
false
{ "statements": [ { "queryid": 455725342067307140, "query": "SELECT id, status, total_cents, created_at\n FROM orders\n WHERE user_id = $1\n ORDER BY created_at DESC\n LIMIT $2", "calls": 30, "total_exec_time": 20.982371999999998, "mean_exec_time": 0.6994123999999999, ...
null
missing_index
Missing index on a hot foreign key
indexing
missing_index
order_items.product_id has no index; the product-lookup query falls back to a sequential scan over the whole table.
database
add an index
true
{ "statements": [ { "queryid": -160921144486592860, "query": "SELECT o.id, o.created_at, oi.quantity\n FROM order_items oi\n JOIN orders o ON o.id = oi.order_id\n WHERE oi.product_id = $1\n ORDER BY o.created_at DESC\n LIMIT $2", "calls": 150, "total_exec_time": 11084.462094...
Drops idx_order_items_product_id and drives product-lookup traffic. The join seq-scans 2.5M rows on every call.
plan_regression
Plan regression from stale statistics
statistics
stale_stats
Statistics on orders are stale after a bulk load. The planner's row estimate for status = 'processing' is off by orders of magnitude, so it picks the wrong plan. Fix is ANALYZE orders, not a new index.
database
add an index on orders.status
false
{ "statements": [ { "queryid": 6567515492305370000, "query": "SELECT id, user_id, total_cents, created_at\n FROM orders\n WHERE status = $1 AND created_at > now() - interval $2\n ORDER BY created_at DESC\n LIMIT $3", "calls": 100, "total_exec_time": 7.046200999999999, "me...
Inserts 300k orders with a brand-new status value and blocks autovacuum, so pg_statistic never learns the value exists.
bloat
Table bloat — autovacuum disabled
vacuum
bloat
autovacuum is disabled on events, so dead tuples from UPDATE churn are never reclaimed. Fix is re-enabling autovacuum and vacuuming.
database
the table is just large
false
{ "statements": [ { "queryid": 6567515492305370000, "query": "SELECT id, user_id, total_cents, created_at\n FROM orders\n WHERE status = $1 AND created_at > now() - interval $2\n ORDER BY created_at DESC\n LIMIT $3", "calls": 30, "total_exec_time": 46.683979999999984, "me...
Disables autovacuum on events and runs repeated mass UPDATEs, leaving several hundred thousand dead tuples that never get reclaimed.
lock_contention
Lock contention — idle in transaction
locking
lock_contention
A session is idle-in-transaction holding row locks on orders. Blocked writers are waiting on it. Fix targets the holder, not the waiters.
session
kill the slow queries
false
{ "statements": [ { "queryid": 455725342067307140, "query": "SELECT id, status, total_cents, created_at\n FROM orders\n WHERE user_id = $1\n ORDER BY created_at DESC\n LIMIT 25", "calls": 30, "total_exec_time": 8.986045000000003, "mean_exec_time": 0.29953483333333336, ...
Leaves a session idle-in-transaction holding row locks on 1000 orders rows, plus three writers that block behind it.
n_plus_1
ORM N+1 query pattern
application
n_plus_1
Application-side N+1. order_items is queried once per order row instead of batched. High call count, low mean time, trivial rows per call. Fix is in the application, not the database.
application
nothing is slow, the database is healthy
false
{ "statements": [ { "queryid": 3395054879552150500, "query": "SELECT id, product_id, quantity FROM order_items WHERE order_id = $1", "calls": 1000, "total_exec_time": 59.843165000000084, "mean_exec_time": 0.059843165000000156, "rows": 2640, "shared_blks_read": 182, ...
Runs 20 pages x 50 rows of list-then-per-row-lookup traffic. No schema change: 1000 fast queries where 20 would do.