MySQL Error 2006

Error number

2006

SQLSTATE

HY000

Symbol

CR_SERVER_GONE_ERROR

Product

MySQL Server

Versions

MySQL client libraries and connectors for MySQL 5.7, 8.0 and 8.4 (PHP mysqlnd reports it for most lost connections)

Category

Connections & Network

Last reviewed

24 October 2026

What it means

Error 2006 means the connection to the server was already closed when the client tried to use it, so the statement was never run (or its result never arrived).

The client library raises 2006 (CR_SERVER_GONE_ERROR) when it cannot send a statement over the connection or the connection turns out to be closed. The server did not reject the statement; the connection was gone. The most common reason, in the MySQL manual's words, is that the server timed out and closed an idle connection: a connection that sends nothing for wait_timeout seconds (28800, eight hours, by default) is closed by the server, and the client only notices when it next sends a query. Other reasons are a KILL of the connection, a server restart or crash, a statement larger than max_allowed_packet (1153), and client-side read or write timeouts.

Which number you see depends on the client library and platform. With PHP mysqli and PDO (mysqlnd) we got 2006 for an idle timeout, for a connection killed in the middle of a query, for a read timeout and for the statement after an oversized packet. The mysql 8.0.46 command-line client on Windows reported 2013 (Lost connection during query) for the same idle timeout, which the manual explains: on Windows the client often gets the error when reading the answer rather than when writing the query. Treat 2006 and 2013 the same way in application code: the connection is unusable and must be reopened.

Common causes

• The connection stayed idle longer than wait_timeout: a long-lived script, a worker that sleeps between jobs, or a pool that hands out a connection the server has already closed.
• Something ended the connection: an administrator or monitoring tool used KILL, the server restarted or crashed, or a failover moved the database.
• The statement was bigger than the server's max_allowed_packet, for example a large BLOB or a huge multi-row INSERT; the server closes the connection after error 1153.
• A client-side timeout fired, such as MYSQLI_OPT_READ_TIMEOUT or a driver's socket timeout shorter than the query.
• A firewall, NAT gateway or load balancer between client and server dropped an idle TCP connection without either side knowing.
• Several processes share one connection (a connection opened before fork()), or the application uses a connection after closing it.

Diagnose it

Read-only These statements only read metadata (information_schema, performance_schema and SHOW statements) or your own data. They do not change data, privileges or configuration, so they are safe to run first.

SHOW VARIABLES LIKE 'wait_timeout';
SHOW VARIABLES LIKE 'interactive_timeout';
SHOW VARIABLES LIKE 'max_allowed_packet';
SHOW VARIABLES LIKE 'net_%_timeout';

SHOW GLOBAL STATUS LIKE 'Uptime';

SELECT id, user, host, db, command, time AS seconds_in_state
FROM performance_schema.processlist
WHERE command = 'Sleep'
ORDER BY time DESC
LIMIT 20;

SELECT error_number, error_name, sum_error_raised, last_seen
FROM performance_schema.events_errors_summary_global_by_error
WHERE error_number IN (1153, 4031);

SHOW GLOBAL STATUS LIKE 'Aborted_clients';

How to fix it

Each fix is labelled with what it changes. Try read-only checks and the least invasive fix first, and test changes outside production.

SET SESSION wait_timeout = 3600;  -- this connection only, in seconds

Examples

An idle connection closed by wait_timeout

Run with PHP mysqli (mysqlnd); PDO gave SQLSTATE[HY000]: General error: 2006 MySQL server has gone away. The mysql 8.0.46 client on Windows reported ERROR 2013 for the same idle timeout.

SET SESSION wait_timeout = 2;  -- this connection only (the default is 28800)
-- the application now sends nothing for 4 seconds
SELECT 1;

ERROR 2006 (HY000): MySQL server has gone away

The connection is killed while a query runs

Session A ran through PHP mysqli and received 2006 as soon as its connection was killed. The mysql command-line client reports this case as 2013 Lost connection to MySQL server during query.

-- Session A: a long query
SELECT SLEEP(10);

-- Session B (for example an administrator), while A is waiting:
KILL CONNECTION 1234;  -- A's connection id from SHOW PROCESSLIST

ERROR 2006 (HY000): MySQL server has gone away

A client read timeout shorter than the query

Captured with PHP 8.2 mysqli (mysqlnd) against MySQL 8.0.46: the query failed after 2 seconds with errno 2006, shown here in the client's ERROR format. The lab cannot set client options, so this one is not rerun automatically.

<?php
$db = mysqli_init();
$db->options(MYSQLI_OPT_READ_TIMEOUT, 2);  // seconds
$db->real_connect('127.0.0.1', 'app', $password, 'shop');
$db->query('SELECT SLEEP(6)');

ERROR 2006 (HY000): MySQL server has gone away

How to prevent it

Design for connections that disappear: keep them short-lived or pooled with an idle limit below wait_timeout, check them after long idle periods, and handle 2006 and 2013 with a reconnect and a deliberate retry of the whole transaction. Never share one connection between forked processes.

Keep statements well under max_allowed_packet, send large data in chunks, and watch the server: alert on restarts (Uptime), on a rising Aborted_clients count and on error 1153 in performance_schema, so the cause is found rather than hidden by a longer timeout.

Common questions

Learn more on CodeWithSQL

• SQL transactions and ACID explained
• Best MySQL monitoring tools

Official documentation

• MySQL server has gone away
• MySQL client error message reference
• Server system variables (wait_timeout, max_allowed_packet)
• Automatic reconnection control (C API)

Last technically reviewed 24 October 2026; examples and diagnostics run on MySQL 8.0.46. Message templates checked against the server's error catalogue (perror) for that version.