5.5.54 fails to start on Debian due refusal to create /var/run/mysqld

[mig5]

Hi,

In 5.5.53-rel38.5-1, there is a portion of /usr/bin/mysqld_safe that looks like this:

---
safe_mysql_unix_port=${mysql_unix_port:-${MYSQL_UNIX_PORT:-/var/run/mysqld/mysqld.sock}}
# Make sure that directory for $safe_mysql_unix_port exists
mysql_unix_port_dir=`dirname $safe_mysql_unix_port`
if [ ! -d $mysql_unix_port_dir ]
then
  if [ ! -h $mysql_unix_port_dir ]; then
    mkdir $mysql_unix_port_dir
    chown $user $mysql_unix_port_dir
    chmod 755 $mysql_unix_port_dir
  fi
fi
---

In 5.5.54-rel38.6-1, that part of the script now looks like this:

---
# Check that directory for $safe_mysql_unix_port exists
mysql_unix_port_dir=`dirname $safe_mysql_unix_port`
if [ ! -d $mysql_unix_port_dir ]
then
  log_error "Directory '$mysql_unix_port_dir' for UNIX socket file don't exists."
  exit 1
fi
---

It no longer tries to create /var/run/mysqld (in my case). This is a bit of a crisis, as it prevents MySQL from starting.

Even worse, since apparently MySQL waits forever (at least on Debian Wheezy) to try and start on boot, a reboot of the machine will make it fail to boot. I had to reboot into recovery mode and fix the script. For some reason my Jessie machines finished booting and I resolved by mkdir /var/run/mysqld && chown mysql /var/run/mysqld by hand.

And as a minor note, 'don't exists' isn't good English. It should be 'doesn't exist'.

[EvgeniyPatlan (evgeniy-patlan)]

Hi.
Thanks for your report, but I wasn't able to reproduce this issue. I have checked on wheezy, jessie, xenial, yakkety
Could you please provide a bit more extended STR and may be logs from installation step.
These logs might shed some light on the situation as the needed directory should be created during package post-installation actions.

[Oleg Motienko (motienko)]

Same issue for me.
Ubuntu 16.04

percona-server-server-5.5
5.5.54-rel38.6-1.xenial
amd64

After installing of deb package it works, but after rebooting the Ubuntu it doesn't start.

[Oleg Motienko (motienko)]

This is git commit of this change

https://github.com/percona/percona-server/commit/da96f8dee0e10c80a2f7eea88d3a7eac94494a32#diff-144aa2f11374843c969d96b7b84247ea

[SWick (swick)]

same issue here with Trusty. After Upgrade AND a reboot Percona doesn't start anymore.
I looked at the Init Script from 5.7 and used this part as a workaround:

--- /tmp/mysql 2017-02-01 23:06:34.270319050 +0100
+++ /etc/init.d/mysql 2017-02-01 23:06:59.449692853 +0100

case "${1:-''}" in
    'start')
    sanity_checks;
    # Start daemon
    log_daemon_msg "Starting MySQL (Percona Server) database server" "mysqld"

    if mysqld_status check_alive nowarn; then

       log_progress_msg "already running"
       log_end_msg 0

    else

+ MYSQLRUN=/var/run/mysqld
+
+ if [ ! -d "${MYSQLRUN}" -a ! -L "${MYSQLRUN}" ];
+ then
+ mkdir ${MYSQLRUN}
+ chown mysql:mysql ${MYSQLRUN}
+ chmod 755 ${MYSQLRUN}
+ fi

[mig5 (mig5)]

@EvgeniyPatlan the issue is that on most distros, /var/run (which is often a symlink to /run) is mounted as tmpfs.

So it doesn't matter of your post-install script makes the /var/run/mysqld - that dir is volatile and does not persist across reboots, and hence why the issue only occurs on fresh boot up/reboot.

I would suggest to look at how other scripts do it - most handle it in the init script like mysqld_safe did, and I believe systemd might have other methods available, but you probably have to support both since Debian Wheezy uses initd by default.

See http://askubuntu.com/questions/303120/how-folders-created-in-var-run-on-each-reboot

[Hrvoje Matijakovic (hrvojem)]

New packages have been pushed to repositories with incremented package version (5.5.54-rel38.6-2 for deb and 5.5.54-rel38.6.2 for rpm) to address this.

[SWick (swick)]

fix works fine. thank you

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-399