Skip to content

Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥

fix(nuq-postgres): aggressive checkpoint tuning to prevent queue stalls (#2679)

firecrawl-spring[bot]7b4acdb5f7e3
diff --git a/apps/nuq-postgres/nuq.sql b/apps/nuq-postgres/nuq.sql
index 4195aa199..e551d2003 100644
--- a/apps/nuq-postgres/nuq.sql
+++ b/apps/nuq-postgres/nuq.sql
@@ -3,10 +3,32 @@ CREATE EXTENSION IF NOT EXISTS pg_cron;
-- Checkpoint tuning: spread I/O to reduce stalls during heavy WAL activity
-- These settings help prevent prefetch queries from returning 0 jobs during checkpoints
-ALTER SYSTEM SET checkpoint_completion_target = 0.95;
-ALTER SYSTEM SET max_wal_size = '4GB';
-ALTER SYSTEM SET bgwriter_lru_maxpages = 500;
-ALTER SYSTEM SET bgwriter_delay = '100ms';
+
+-- Checkpoint settings: reduce frequency and spread I/O
+ALTER SYSTEM SET checkpoint_completion_target = 0.9; -- Spread checkpoint I/O over 90% of interval
+ALTER SYSTEM SET checkpoint_timeout = '15min'; -- Longer intervals between time-based checkpoints
+ALTER SYSTEM SET max_wal_size = '16GB'; -- Much larger WAL before forced checkpoint (was 4GB)
+ALTER SYSTEM SET min_wal_size = '4GB'; -- Keep WAL pre-allocated to avoid allocation stalls
+
+-- Aggressive background writer: pre-flush dirty pages to reduce checkpoint burst
+ALTER SYSTEM SET bgwriter_lru_maxpages = 1000; -- Flush up to 1000 pages per round (was 500)
+ALTER SYSTEM SET bgwriter_lru_multiplier = 4.0; -- More aggressive dirty page estimation
+ALTER SYSTEM SET bgwriter_delay = '50ms'; -- Run twice as often (was 100ms)
+ALTER SYSTEM SET bgwriter_flush_after = '512kB'; -- Force OS flush after 512kB written
+
+-- I/O concurrency for SSD/cloud storage (hyperdisk)
+ALTER SYSTEM SET effective_io_concurrency = 200; -- Parallel I/O operations for prefetch
+ALTER SYSTEM SET maintenance_io_concurrency = 100; -- I/O concurrency for maintenance ops
+
+-- WAL settings for better write performance
+ALTER SYSTEM SET wal_buffers = '64MB'; -- Larger WAL buffer (default is too small)
+ALTER SYSTEM SET wal_writer_delay = '10ms'; -- Flush WAL more frequently to avoid bursts
+ALTER SYSTEM SET wal_writer_flush_after = '1MB'; -- Flush after 1MB of WAL
+
+-- Reduce fsync overhead
+ALTER SYSTEM SET commit_delay = 10; -- Microseconds to wait for group commit
+ALTER SYSTEM SET commit_siblings = 5; -- Min concurrent transactions for commit_delay
+
SELECT pg_reload_conf();
CREATE SCHEMA IF NOT EXISTS nuq;