Pgpool + Postgres integration

[paunin]

Now Pgpool and Repmgr know nothing about each over. But there is a chance when a new master elected pgpool still points to old master.

To Be Done:

• Pgpool should receive signals from repmgr(event in repmgr?) about new master and automatically exclude old (if it's still online)
• For short period of time (before helthcheck for false_master is done or it's disabled or no quorum) - splitbrain situation is possible, so pgpool should be ready for that and ignore minor master and it's slaves
• Pgpool should be able to accept signals about slaves are back (let say after restart)
• Avoid SSH servers in all containers (pcp protocol?...)

[alivespirit]

Regarding point №3, I've added this to has_enough_backends.sh in my k8s deploy:

BACKENDS_COUNT=`echo $BACKENDS |awk -F ',' '{print NF}'`
if [[ "$HEALTHY_BACKENDS" -ne "$BACKENDS_COUNT" ]]; then
  for (( node=0; node<$BACKENDS_COUNT; node++ )); do
    node_state=`pcp_node_info -h localhost -U $PCP_USER -w -n $node | awk '{print $5}'`
    if [[ "$node_state" != "up" ]]; then
      pcp_attach_node -h localhost -U $PCP_USER -w -n $node || echo ">>> Node $node attach failed!"
    fi
  done
fi

This way if some node was marked as down and then went back online, liveness probe will mark it again as up.