Nagios Postgres replication delay check.

check_replication_delay

#!/usr/bin/env bash

##
# Postgres replication delay check.
#
# @author Jay Taylor [@jtaylor]
#
# @date 2013-07-08
#

while getopts “u:p:d:w:c:” OPTION; do
    case $OPTION in
        u)
            user=$OPTARG
            ;;
        p)
            password=$OPTARG
            ;;
        d)
            dbname=$OPTARG
            ;;
        w)
            warning=$OPTARG
            ;;
        c)
            critical=$OPTARG
            ;;
    esac
done

if [ -z "${user}" ]; then
    echo 'error: missing required parameter: -u [user]' 1>&2
    exit 1
fi
if [ -z "${password}" ]; then
    echo 'error: missing required parameter: -p [password]' 1>&2
    exit 1
fi
if [ -z "${dbname}" ]; then
    echo 'error: missing required parameter: -d [dbname]' 1>&2
    exit 1
fi
if [ -z "${warning}" ]; then
    echo 'error: missing required parameter: -w [warningThresholdSeconds]' 1>&2
    exit 1
fi
if [ -z "${critical}" ]; then
    echo 'error: missing required parameter: -c [criticalThresholdSeconds]' 1>&2
    exit 1
fi

psqlResult=$(psql "host=localhost user=${user} password=${password} dbname=${dbname}" -c 'SELECT now() - pg_last_xact_replay_timestamp()' 2>&1)
exitCode=$?
if [ $exitCode -ne 0 ]; then
    echo "psql exited with non-zero code: $exitCode"
    exit 3
fi

value=$(echo $psqlResult | grep -o '[0-9:\.]\+' | head -n1 | sed 's/00\([:\.]\)/0\1/g' | \
    python -c "import sys; sys.stdout.write('{0}'.format(round(sum((float(x) if '.' in x else int(x)) * 60 ** i for i,x in enumerate(reversed(sys.stdin.read().strip().split(':')))), 2)))")

echo "replication delay=${value}s"

status=$(python -c "import sys; value=float(${value}); sys.stdout.write('OK' if value < float(${warning}) else 'WARNING' if value < float(${critical}) else 'CRITICAL')")

if [[ $status = 'OK' ]]; then
    exit 0
elif [[ $status = 'WARNING' ]]; then
    exit 1
elif [[ $status = 'CRITICAL' ]]; then
    exit 2
fi