worker threads for postgresql are not terminated properly on PROXYSQL RESTART; #5086
Description
proxysql - current HEAD 3.0.3-64-g0d2e9ef_DEBUG
While connecting to the proxysql postgresql query port, the psql client seems to hang, and needs to be stopped by CTLR-C.
# psql --version
psql (PostgreSQL) 17.5 (Debian 17.5-1)
# psql -Uroot -h127.0.0.1 -p6133
^C
#
Reproducible in about 50% of tries.
Expected result is the password prompt:
# psql -Uroot -h127.0.0.1 -p6133
Password for user root:
To reproduce, start proxysql, and issue a PROXYSQL RESTART; command.
listing the bound 6133 ports shows 4, before restart:
# netstat -ntpl | grep 6133
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
listing the bound 6133 ports shows 8, after restart:
# netstat -ntpl | grep 6133
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
tcp 0 0 0.0.0.0:6133 0.0.0.0:* LISTEN 161161/./proxysql
Conclusion:
postgresql working threads are not being terminated properly by the PROXYSQL RESTART; command, and a new set of threads is started.
Zombie threads are getting 50% of the connections which then hang.