[pgpool-general-jp: 1143] Improvement of watchdog monitoring method.
nagata ＠ sraoss.co.jp
2013年 3月 1日 (金) 16:37:06 JST
I'm thinking to fix pgpool watchdog to use heartbeat signal instead of
In current version, watchdog monitors other pgpools by using SQL query
such as 'SELECT 1'. However, this requires a connection to backend DB.
Therefore if there are connections larger than max_connections, watchdog
fails to establish connection and treats pgpool-II as down. In addition,
if all backend DB are down, watchdog treats pgpool-II as down.
The idea for an improvement is that watchdog monitors other pgpool by
using heartbeat signal. Watchdog sends hearbeat to others periodically.
If there are no signal for a certain interval, watchdog treats the pgpool
as down. I think of using UDP for low overhead.
Yugo Nagata <nagata ＠ sraoss.co.jp>