Terminate backend when query_timeout is reached #1954

Activity

1. rinarakaki commented on Oct 15, 2026

   

   rinarakaki

   on Oct 15, 2026

   More actions

   A consequence worth adding: after query_timeout fails a query that was already sent, the client stays busy with it. So the next query waits behind it, and fails with the same timeout without ever running. Reproduced on pg 8.23.1, the latest, against Postgres 18; master (0980cef) is the same:

   const { Client } = require("pg");
   
   const client = new Client({ connectionString: process.env.DATABASE_URL, query_timeout: 1000 });
   await client.connect();
   const start = Date.now();
   await client.query("select pg_sleep(3)").catch((error) => console.log(error.message, Date.now() - start)); // Query read timeout 1004
   await client.query("select 1").catch((error) => console.log(error.message, Date.now() - start)); // Query read timeout 2008

   Only after the server finishes the first query does the client answer again. So a ROLLBACK sent after a timed-out statement in a transaction waits out its own timeout too, about twice the timeout in all.

   In pipeline mode, a timed-out query that was sent destroys the connection for this reason (else if (this.pipeline) in client.js). Without pipelining it doesn't, and the client stays tied to the running query. We work around it by discarding the client once a query times out, sending nothing after it (client.release(error)), so the server ends the session.