Hi,
It seems like there is a disagreement between postgres and pgpool as to where the Unix domain sockets for talking to postgres ought to be found.
It's a situation like it can connect to postgres buy not pgpool, as it was directing all traffic to the local postgres node.
1) Check whether any of the postgres backends are cancelling query with 'statement time out' error.
2) If statement time out is not set in postgres.conf file then the application might have altered.
3) Check for any long running queries are being cancelled by Server.
Regards,
Pavan