Blog
Backfill Cost
A postgres backfill on a large table doubled its heap in range batches without VACUUM, while id % 10 batches at fillfactor 90 kept 96.1% of updates HOT.
Sep 26, 2026Read article →Missing Query Stats
Forty one-off deletes disappeared from pg_stat_statements while the same session read the schema. What the view merges, and what it never records at all.
Sep 22, 2026Read article →Pooled Session Leaks
In PgBouncer transaction mode, SET, temp tables and advisory locks stay on the backend and the next client reads them. The measured matrix of what leaks.
Sep 18, 2026Read article →Agent Kill Switch
The Postgres role for AI agent sessions usually carries statement_timeout and a read-only default. Both are USERSET, so the session lifts them itself.
Sep 11, 2026Read article →Session Replication Role
Setting session_replication_role to replica drops foreign keys and most triggers, leaving CHECK, NOT NULL, UNIQUE and RLS enforcing. The verified matrix.
Sep 4, 2026Read article →Pg_dump Data-Only Restore
A pg_dump data-only restore can break on a schema holding zero rows, and psql exits 0 while it happens. The verified failure map, with what each fix costs.
Aug 28, 2026Read article →Postgres Sequence Out of Sync
A Postgres sequence out of sync with its table breaks the first insert after a data load. Why it happens on serial and identity columns, and how to realign it.
Aug 21, 2026Read article →Synthetic vs Real Data
Synthetic data vs real data has no single winner; the answer flips with the job. For application databases, Seedfast generates rows that fit the schema.
Jul 24, 2026Read article →Training Data for AI Agents
Training data for AI agents that run software is an environment, not a labeled dataset. Seedfast seeds its database with connected, realistic rows.
Jul 24, 2026Read article →