Naive is-alive check for your servers
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
icanhasserver/status-check.sh

123 lines
3.0 KiB

#!/bin/bash
# Check status of a list of servers
#
# Any servers in urls that yield a bad response are collated and sent via email
# to all recipients.
# recipients are sourced from a separate file in the cwd; it is expected to
# have a single email address per line.
#
# If WATCHDOG_LOG is set, all output will be redirected there.
#
# Dependencies
# - curl: Polling the servers
# - msmtp: Sending email; requires separate configuration
# NOTE: Currently only useful for services running over HTTP
#set -eux
main()
{
local recipients_file="./recipients"
local urls_file="./urls"
# TODO: How to check these sensibly?
#local services_file="./services"
if [ ! -f ${recipients_file} ]; then
echo "[ERROR] Please create a file '${recipients_file}'"
exit 1
fi
if [ ! -f ${urls_file} ]; then
echo "[ERROR] Please create a file '${urls_file}'"
exit 1
fi
# See above TODO
#if [ ! -f ${services_file} ]; then
# echo "[ERROR] Please create a file '${services_file}'"
# exit 1
#fi
# Source array contents from files, ignoring commented lines
local recipients=($(cat ${recipients_file} | sed 's/^#.*$//'))
local urls=($(cat ${urls_file} | sed 's/^#.*$//'))
# See above TODO
#local services=($(cat ${services_file} | sed 's/^#.*$//'))
declare -A FLAGGED
if [ -n "${WATCHDOG_LOG}" ]; then
rotate_log WATCHDOG_LOG
# Redirect all output to WATCHDOG_LOG but preserve a link to stdout should we
# still want to use it
exec 3>&1 1>>${WATCHDOG_LOG} 2>&1
fi
echo ">> $(date)"
echo "== Checking:"
for uri in "${urls[@]}"; do \
full_uri="https://${uri}"
echo -n " ${full_uri} ... ";
stat="$(curl -s -o /dev/null -w "%{http_code}%{stdout}" ${full_uri})"
echo "${stat}"
# All other HTTP responses should not technically be errors
if ((${stat} >= 400 && ${stat} <= 599)); then
FLAGGED[${full_uri}]=${stat}
fi
done
ret=0
if ((${#FLAGGED[@]} > 0)); then
echo "== Error(s) detected. Sending mail to:"
for rcp in ${recipients[@]}; do
echo " ${rcp}"
done
send_mail FLAGGED
ret=1
else
echo "== OK"
fi
exit ${ret}
}
rotate_log()
{
local -n log=$1
local max_log=4096 # 4M as we get size in K below
touch ${log}
local log_size=$(du -k "${log}" | cut -f 1)
if ((${log_size} > ${max_log})); then
# Clobber any previous backup
mv -f ${log} ${log}.1
touch ${log}
fi
}
send_mail()
{
local -n error_list=$1
local error_string=""
for i in "${!error_list[@]}"; do
# Use bash's awful syntax to escape any / in the server string
code=${i//\//\\\/}
server=${error_list[$i]}
error_string+=" ${code}: ${server}\n"
done
cat mail_template \
| sed "s/{ERR_LIST}/${error_string}/g" \
| msmtp -a omn_server_watch ${recipients[@]} \
|| echo "[ERROR] Failed to send mail!"
}
main