[pgpool-hackers: 687] pgpool-II vs. SIGKILL

[Tatsuo Ishii]

Hi,

Currently if pgpool-II is killed by SIGKILL (kill -9), sockets for
pgpool-II and pcp are remained and next time pgpool-II won't start
because of the target socket file is already there.

I think we should fix this because this is very annoying under no
operator attended environment.

I think we could steel PostgreSQL's idea:

1) have a pid file (pgpool-II already has, pcp not)

2) when starting up, try to bind to the socket anyway. If it fails,
   then check the pid file.

3) if there's no pid files, then we unlink the socket files and bind
   again.

4) if there's pid files, send signal to the process recorded in the
   pid files. If the process there, give up because there must be
   already pgpool-II process and it is running. If the process is not
   there, we do #3.

Comments or objections?
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp