Reduce Supabase disk IO

2026-05-29 17:22:33 +08:00
parent 9e3a4e2f45
commit 2269aaefd7
63 changed files with 4637 additions and 410 deletions
@@ -0,0 +1,98 @@
+-- Supabase Disk IO diagnostics for PolyWeather.
+-- Run sections independently if your project does not expose every stats view.
+
+-- 1) Tables with high sequential scan pressure.
+select
+  schemaname,
+  relname,
+  seq_scan,
+  seq_tup_read,
+  idx_scan,
+  n_tup_ins,
+  n_tup_upd,
+  n_tup_del,
+  n_dead_tup
+from pg_stat_user_tables
+where schemaname = 'public'
+order by seq_tup_read desc
+limit 30;
+
+-- 2) Largest public tables and indexes.
+select
+  relname,
+  pg_size_pretty(pg_total_relation_size(relid)) as total_size,
+  pg_size_pretty(pg_relation_size(relid)) as table_size,
+  pg_size_pretty(pg_indexes_size(relid)) as index_size
+from pg_catalog.pg_statio_user_tables
+where schemaname = 'public'
+order by pg_total_relation_size(relid) desc
+limit 30;
+
+-- 3) Slow/high-read statements, if pg_stat_statements is available.
+select
+  calls,
+  round(total_exec_time::numeric, 2) as total_exec_ms,
+  round(mean_exec_time::numeric, 2) as mean_exec_ms,
+  rows,
+  shared_blks_read,
+  shared_blks_hit,
+  temp_blks_read,
+  temp_blks_written,
+  left(query, 500) as query_sample
+from pg_stat_statements
+order by shared_blks_read desc, total_exec_time desc
+limit 30;
+
+-- 4) Dead tuple pressure that can trigger heavier autovacuum work.
+select
+  schemaname,
+  relname,
+  n_live_tup,
+  n_dead_tup,
+  last_vacuum,
+  last_autovacuum,
+  last_analyze,
+  last_autoanalyze
+from pg_stat_user_tables
+where schemaname = 'public'
+order by n_dead_tup desc
+limit 30;
+
+-- 5) Indexes that still read the most disk blocks.
+select
+  s.schemaname,
+  s.relname,
+  s.indexrelname,
+  s.idx_scan,
+  s.idx_tup_read,
+  s.idx_tup_fetch,
+  io.idx_blks_read,
+  io.idx_blks_hit,
+  pg_size_pretty(pg_relation_size(s.indexrelid)) as index_size
+from pg_stat_user_indexes s
+join pg_statio_user_indexes io
+  on io.indexrelid = s.indexrelid
+where s.schemaname = 'public'
+order by io.idx_blks_read desc, s.idx_scan desc
+limit 50;
+
+-- 6) Large non-unique indexes with no recorded scans since stats reset.
+-- Treat this as a candidate list only; confirm with production traffic history
+-- before dropping anything.
+select
+  s.schemaname,
+  s.relname,
+  s.indexrelname,
+  s.idx_scan,
+  pg_size_pretty(pg_relation_size(s.indexrelid)) as index_size,
+  i.indisunique,
+  i.indisprimary
+from pg_stat_user_indexes s
+join pg_index i
+  on i.indexrelid = s.indexrelid
+where s.schemaname = 'public'
+  and s.idx_scan = 0
+  and not i.indisprimary
+  and not i.indisunique
+order by pg_relation_size(s.indexrelid) desc
+limit 30;