MySQL Error 2002: Can't connect through socket fix

What's actually happening here

Error 2002 (HY000) means the MySQL client can't find the mysql.sock file. That file only exists while the MySQL server (mysqld) is running. So either:

The server isn't running at all
The server started but crashed right after
Your client is looking in the wrong directory for the socket

This usually hits you after a reboot, a failed update, or when you've just installed MySQL and haven't started it yet. I see it most often on Ubuntu 20.04+ and CentOS 7 boxes.

30-second fix: start the MySQL service

Run this immediately:

sudo systemctl start mysql

On some distros it's mariadb instead of mysql:

sudo systemctl start mariadb

Check if it's running:

sudo systemctl status mysql

If you see "active (running)" in green, you're done. The socket file appears at /var/run/mysqld/mysql.sock (or /var/lib/mysql/mysql.sock on older systems) the moment the process starts.

Why this works: systemd creates the socket file as part of the service start-up. If MySQL ran normally before a reboot, this single command brings it back.

If systemctl start fails with "Unit mysql.service not found" — you probably have MariaDB instead. Install it with sudo apt install mariadb-server or sudo yum install mariadb-server, then start it.

5-minute fix: check why MySQL won't start

If the server won't start, systemctl status will show "failed" or "inactive (dead)". Now we dig.

1. Look at the logs

sudo journalctl -u mysql --no-pager | tail -50

Or find the error log directly:

sudo tail -100 /var/log/mysql/error.log

Common causes I've hit:

Disk full — MySQL refuses to start. df -h /var/lib/mysql
Corrupted ibdata1 — InnoDB data file got messed up. You'll see "Corrupted page" or "innodb_force_recovery" references.
Port conflict — Another MySQL instance or process on port 3306. sudo ss -tlnp | grep 3306
Permission issue — The mysql user doesn't own /var/lib/mysql. sudo chown -R mysql:mysql /var/lib/mysql

2. Try safe mode

If the log mentions InnoDB corruption, start MySQL with recovery:

sudo mysqld_safe --skip-grant-tables &

Then in another terminal:

mysql -u root

If that works, you know the issue is in the data, not the config. Dump what you can with mysqldump and restore fresh.

3. Verify the socket path in your client

Sometimes MySQL is running, but your client looks in the wrong place. Check with:

mysqladmin variables | grep socket

That shows the actual socket path the server is using. Then connect explicitly:

mysql -S /var/run/mysqld/mysql.sock

If that works, your client's config (usually /etc/mysql/my.cnf or ~/.my.cnf) has a bad socket line. Fix it.

15+ minute fix: rebuild the socket file and fix the root cause

If the socket file is missing and starting the service doesn't create it, something is fundamentally broken. Here's the nuclear option.

1. Kill any stuck MySQL processes

sudo pkill -9 mysqld
sudo pkill -9 mysqld_safe
sleep 2
sudo rm -f /var/run/mysqld/mysql.sock /var/lib/mysql/mysql.sock

The rm -f is safe — if the socket is there, it's from a dead or orphaned process. A fresh start will recreate it.

2. Recreate the socket directory

sudo mkdir -p /var/run/mysqld
sudo chown mysql:mysql /var/run/mysqld

On some systems (like Arch or manual installs), this directory doesn't get created at boot. MySQL expects it to exist and fails silently.

3. Start MySQL with full debug output

sudo mysqld --user=mysql --console

This runs in the foreground and prints every error to your terminal. You'll see things like:

[ERROR] Can't start server: Bind on TCP/IP port: Address already in use — kill whatever's on port 3306
[ERROR] Fatal error: Can't open and lock privilege tables: Table 'mysql.user' doesn't exist — you need to reinitialize the database

4. Reinitialize the database (last resort)

If the error log says privilege tables are missing or corrupted, back up /var/lib/mysql (if you care about the data) and reinitialize:

sudo mv /var/lib/mysql /var/lib/mysql.bak
sudo mkdir -p /var/lib/mysql
sudo chown mysql:mysql /var/lib/mysql
sudo mysql_install_db --user=mysql --basedir=/usr --datadir=/var/lib/mysql

For MySQL 8.0+, the command is mysqld --initialize instead:

sudo mysqld --initialize --user=mysql

That creates a temporary root password printed to the log. Grab it with sudo grep 'temporary password' /var/log/mysql/error.log and log in immediately to change it.

The reason step 3 works: mysqld --initialize creates the mysql system tables (including the socket path config) from scratch. After this, systemctl start mysql will succeed and the socket file appears.

One last thing: check your `my.cnf` for typos

A common footgun I've seen — someone edits /etc/my.cnf and puts a stray character in the [mysqld] section. MySQL silently ignores the whole section if the syntax is wrong. Check for:

Missing semicolons on socket and pid-file lines
Blank lines with spaces
Uppercase MYSQL instead of mysqld in the section header

If you're not sure, rename the file and try a default start:

sudo mv /etc/my.cnf /etc/my.cnf.bak
sudo systemctl start mysql

If that works, your config was the problem. Restore it line by line.