rephacheck.py

#!/usr/bin/env python3

"""
Determine by voting which is the state of each node.
For this to work properly, you need to have an odd number of nodes.
"""

from collections import Counter
import psycopg2
import toml


def get_state(addr, port, node_id):
    try:
        # postgresql query
        con_args = {'host': addr, 'port': port, 'connect_timeout': TIMEOUT}
        con_args.update(CONNINFO)
        con = psycopg2.connect(**con_args)
        cur = con.cursor()
        query = 'SELECT active, type FROM repmgr.nodes WHERE node_id = {};'
        cur.execute(query.format(node_id))
        data = cur.fetchone()
        cur.close()
        # return result
        return data
    except Exception:
        # an error occured, so return false by default
        return (False, 'unknown')


def get_quorum_state(node_id):
    # init vars
    votes = []
    # ask each node for the state of `node_id`
    for node in NODES.values():
        active, role = get_state(node['addr'], node['port'], node_id)
        # if node considered active take vote, otherwise fence it
        if active:
            votes.append(role)
        else:
            votes.append('fenced')
    # determines voting result
    results = Counter(votes)
    state = results.most_common(1)[0][0]
    # catch inconsistent state case
    if state == 'standby' and 'primary' in votes:
        state = 'fenced'
    # return result
    return state


if __name__ == '__main__':
    with open('/etc/postgresql/11/main/rephacheck.conf') as rephaconf:
        conf = toml.load(rephaconf)

    NODES = conf.get('nodes')
    CONNINFO = conf.get('conninfo')
    TIMEOUT = conf.get('connect_timeout', 3)
    CURRENT = conf.get('local_node_id')

    state = get_quorum_state(CURRENT)
    print(state)