Instantiated systemd unit (/lib/systemd/system/mysqld@.service) uses mysqld-wait-ready script to check status of launched mysqld daemon. Which internally uses /usr/bin/my_print_defaults to read socket address from config(defaults) file. So if config file (/etc/mysql/my%I.cnf) redefines socket path mysqld-wait-ready will use incorred (readed from default config file /etc/mysql/my.cnf) value of socket path. As result, it will or report that mysqld daemon does not start (after 300 seconds timeout) or it will read check "other" mysqld instance status (accessible via socket path defined into /etc/mysqld/my.cnf) file. Proposed fix into mysqld-wait-ready script: diff -Nau a/mysqld-wait-ready b/mysqld-wait-ready --- a/mysqld-wait-ready 2021-04-27 23:44:51.949666241 +0300 +++ b/mysqld-wait-ready 2021-04-27 23:45:36.758275459 +0300 @@ -8,6 +8,7 @@ # Service file passes us the daemon's PID (actually, mysqld_safe's PID) daemon_pid="$1" +defaults_file="$2" # extract value of a MySQL option from config files # Usage: get_mysql_option SECTION VARNAME DEFAULT @@ -15,7 +16,7 @@ # We use my_print_defaults which prints all options from multiple files, # with the more specific ones later; hence take the last match. get_mysql_option(){ - result=`/usr/bin/my_print_defaults "$1" | sed -n "s/^--$2=//p" | tail -n 1` + result=`/usr/bin/my_print_defaults ${defaults_file:+--defaults-file="$defaults_file"} "$1" | sed -n "s/^--$2=//p" | tail -n 1` if [ -z "$result" ]; then # not found, use default result="$3" And change PostExec into unit file to the: ExecStartPost=/usr/libexec/mysqld-wait-ready $MAINPID /etc/mysql/my%I.cnf Reproducible: Always
