scripts/cluster/apt.check.update.sh

371 lines
10 KiB
Bash
Raw Normal View History

#!/bin/sh
# This script will check if any APT upgrade is available and
# will prepare the host in order to apply upgrade with another script
# 1. Create a temp file
# 2. Disable SGE queue
# This script can be call by a cronjob (eg. weekly)
# Another script should try to apply upgrades also with cron (eg. hourly)
# Vars {{{
readonly PROGNAME=$(basename "${0}")
readonly PROGDIR=$(readlink -m $(dirname "${0}"))
readonly ARGS="${*}"
readonly NBARGS="${#}"
2020-09-24 15:18:09 +02:00
[ -z "${DEBUG}" ] && DEBUG=1
## Export DEBUG for sub-script
export DEBUG
readonly APT_TMP_FILE="/tmp/.apt.upgrade"
## Colors
readonly PURPLE='\033[1;35m'
readonly RED='\033[0;31m'
readonly RESET='\033[0m'
readonly COLOR_DEBUG="${PURPLE}"
# }}}
usage() { # {{{
cat <<- EOF
usage: $PROGNAME [-a|-d|-h|-t]
Verify if any APT package upgrade is available and
try to prepare the host by:
* Disabling SGE queue
EXAMPLES:
- Verify upgrade and prepare the current host
${PROGNAME}
2020-09-24 15:18:09 +02:00
- Prepare the current host (if pending upgrade) only after 6 runs
if sge_slots_used is under 75% of total SGE slots.
${PROGNAME} --attempts 6
2020-09-24 15:18:09 +02:00
OPTIONS:
-a,--attempt,--attempts,-t,--tries INT_MAX_ATTEMPTS
Number of time this script must be run to disable the host
if sge_slots_used is under 75% of total SGE slots.
2020-09-24 15:18:09 +02:00
-d,--debug
Enable debug messages.
-h,--help
Print this help message.
EOF
}
# }}}
debug_message() { # {{{
local_message="${1}"
## Print message if DEBUG is enable (=0)
[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6b\e[m\n' "DEBUG ${PROGNAME}: ${local_message}"
return 0
}
# }}}
define_vars() { # {{{
## If sge_hostname wasn't defined (environment variable,…)
if [ -z "${sge_hostname}" ]; then
## Use local host for sge_hostname
sge_hostname="$(hostname -f)"
fi
## If max_attempts wasn't defined
if [ -z "${max_attempts}" ]; then
## Allow maximum 3 attempts before disable a host
max_attempts="3"
fi
## Script used to disable SGE queue(s)
sge_disable_host_queue_script="${PROGDIR}/sge.disable.host.queue.sh"
## Get the number of total SGE slots
sge_slots=$(qhost -h "${sge_hostname:=/dev/null}" -q -xml \
| grep --max-count=1 -- "'slots'" \
| sed 's;.*<queuevalue.*>\(.*\)</queuevalue>;\1;')
## Get the number of SGE used slots
sge_slots_used=$(qhost -h "${sge_hostname:=/dev/null}" -q -xml \
| grep --max-count=1 -- "'slots_used'" \
| sed 's;.*<queuevalue.*>\(.*\)</queuevalue>;\1;')
}
# }}}
2020-06-15 13:49:25 +02:00
is_apt_upgrade_absent() { # {{{
## Count the number of upgradable packages and substract 1 for the header
2020-06-25 08:50:17 +02:00
local_apt_upgrade_number="$(apt list --upgradable 2>/dev/null \
2020-06-04 16:52:10 +02:00
| wc -l \
| awk '{print $1-1}')"
case "${local_apt_upgrade_number}" in
0 ) ## No available upgrade
return_apt_upgrade_absent="0"
;;
* ) ## Upgrade seems available
return_apt_upgrade_absent="1"
;;
esac
## Simple debug message to valid current variable
debug_message "is_apt_upgrade_absent \
APT upgrade available for this system: ${RED}${local_apt_upgrade_number:=/dev/null}${COLOR_DEBUG}."
return "${return_apt_upgrade_absent}"
}
# }}}
2020-06-15 13:49:25 +02:00
is_apt_upgrade_present() { # {{{
## Count the number of upgradable packages and substract 1 for the header
2020-06-25 08:50:17 +02:00
local_apt_upgrade_number="$(apt list --upgradable 2>/dev/null \
2020-06-04 16:52:10 +02:00
| wc -l \
| awk '{print $1-1}')"
case "${local_apt_upgrade_number}" in
0 ) ## No available upgrade
return_apt_upgrade_present="1"
;;
* ) ## Upgrade seems available
return_apt_upgrade_present="0"
;;
esac
## Simple debug message to valid current variable
debug_message "is_apt_upgrade_present \
APT upgrade available for this system: ${RED}${local_apt_upgrade_number:=/dev/null}${COLOR_DEBUG}."
return "${return_apt_upgrade_present}"
}
# }}}
is_sge_slots_more_than_percentage() { # {{{
local_percentage="${1}"
## Get the expected percentage of total SGE slots
local_sge_slots_percentage=$(echo "${sge_slots}" \
| awk -v percentage="0.${local_percentage}" '{ print int($1 * percentage) }')
if [ "${sge_slots_used}" -ge "${local_sge_slots_percentage}" ]; then
## Used slots is greater or equal than expected percentage
return_sge_slots_percentage="0"
## Simple debug message to valid current variable
debug_message "is_sge_slots_more_than_percentage \
Used slots has reached ${RED}${local_percentage}%${COLOR_DEBUG} of total slots: ${RED}${sge_slots_used:=/dev/null}${COLOR_DEBUG}/${sge_slots}."
else
return_sge_slots_percentage="1"
## Simple debug message to valid current variable
debug_message "is_sge_slots_more_than_percentage \
Used slots did not reach ${RED}${local_percentage}%${COLOR_DEBUG} of total slots: ${RED}${sge_slots_used:=/dev/null}${COLOR_DEBUG}/${sge_slots}."
fi
return "${return_sge_slots_percentage}"
}
# }}}
is_pending_upgrade_more_than_attempts() { # {{{
local_attempts="${1}"
2020-11-30 17:50:27 +01:00
local_line_size="26"
local_max_file_size="$(( ${local_attempts} * ${local_line_size} ))"
debug_message "is_pending_upgrade_more_than_attempts (${local_attempts}) \
Check if ${APT_TMP_FILE} has a size bigger than ${RED}${local_max_file_size:=/dev/null}${COLOR_DEBUG} bytes (${local_attempts} attempts * ${local_line_size} bytes size for one line)."
if [ $(find "${APT_TMP_FILE}" -type f -size +"${local_max_file_size}"c 2>/dev/null) ]; then
## Temp file seems to exist for more than the maximum attempts
return_pending_upgrade_more_than_attempts="0"
debug_message "is_pending_upgrade_more_than_attempts \
There is pending upgrade(s) for more than ${RED}${local_attempts:=/dev/null}${COLOR_DEBUG} attempts."
else
return_pending_upgrade_more_than_attempts="1"
debug_message "is_pending_upgrade_more_than_attempts \
NO pending upgrades for more than ${RED}${local_attempts:=/dev/null}${COLOR_DEBUG} attempts."
fi
return "${return_pending_upgrade_more_than_attempts}"
}
# }}}
is_sge_slots_empty() { # {{{
if [ "${sge_slots_used}" -eq "0" ]; then
## Used slots is null
return_sge_slots_empty="0"
else
return_sge_slots_empty="1"
fi
## Simple debug message to valid current variable
debug_message "is_sge_slots_empty \
SGE slots currently in use: ${RED}${sge_slots_used:=/dev/null}${COLOR_DEBUG}."
return "${return_sge_slots_empty}"
2020-09-24 14:56:27 +02:00
}
# }}}
is_pending_job_empty() { # {{{
## Count the number of pending jobs (qw state only)
## Excluding root's jobs
local_pending_jobs="$(qstat -s p -u '*' \
| grep --count --perl-regexp -- "(?=.*?\bqw\b)((?!root).)*$")"
case "${local_pending_jobs}" in
0 ) ## Pending jobs list is empty
return_pending_job_empty="0"
;;
* ) ## Some jobs are waiting
return_pending_job_empty="1"
;;
esac
## Simple debug message to valid current variable
debug_message "is_pending_job_empty \
Pending jobs for the compute cluster: ${RED}${local_pending_jobs:=/dev/null}${COLOR_DEBUG}."
return "${return_pending_job_empty}"
}
# }}}
main() { # {{{
## Define all vars
define_vars
2020-09-24 14:56:27 +02:00
## If NO APT package upgrade is available {{{
### Ensure to remove any temp file related to APT upgrades
### Exit
is_apt_upgrade_absent \
&& rm -f -- "${APT_TMP_FILE}" \
&& exit 0
2020-09-24 14:56:27 +02:00
## }}}
2020-09-25 15:08:51 +02:00
## If APT package upgrade(s) is available AND
is_apt_upgrade_present \
2020-09-25 15:08:51 +02:00
&& {
## If SGE used slots is more than 75% of total slots {{{
## APT package upgrade is available
### Create a temp file
### Disable SGE queue
is_sge_slots_more_than_percentage "75" \
&& touch "${APT_TMP_FILE}" && echo "APT upgrade is available." >> "${APT_TMP_FILE}" \
&& sh "${sge_disable_host_queue_script}" \
&& exit 0
## }}}
## If pending upgrade since max_attempts (default: 3) {{{
2020-09-25 15:08:51 +02:00
## APT package upgrade is available
### Create a temp file
### Disable SGE queue
is_pending_upgrade_more_than_attempts "${max_attempts}" \
2020-09-25 15:08:51 +02:00
&& touch "${APT_TMP_FILE}" && echo "APT upgrade is available." >> "${APT_TMP_FILE}" \
&& sh "${sge_disable_host_queue_script}" \
&& exit 0
## }}}
## If SGE used slots is NULL {{{
## APT package upgrade is available
### Create a temp file
### Disable SGE queue
is_sge_slots_empty \
&& touch "${APT_TMP_FILE}" && echo "APT upgrade is available." >> "${APT_TMP_FILE}" \
&& sh "${sge_disable_host_queue_script}" \
&& exit 0
## }}}
## If pending job list is empty {{{
## APT package upgrade is available
### Create a temp file
### Disable SGE queue
is_pending_job_empty \
&& touch "${APT_TMP_FILE}" && echo "APT upgrade is available." >> "${APT_TMP_FILE}" \
&& sh "${sge_disable_host_queue_script}" \
&& exit 0
## }}}
## After all just add content to a temp file (+1 attempt) {{{
2020-09-25 15:08:51 +02:00
touch "${APT_TMP_FILE}" && echo "APT upgrade is available." >> "${APT_TMP_FILE}" \
&& debug_message "main Add content to temp file for pending upgrade(s)." \
&& exit 0
## }}}
}
}
# }}}
2020-09-24 15:18:09 +02:00
# Manage arguments # {{{
# This code can't be in a function due to argument management
2020-09-24 15:18:09 +02:00
if [ ! "${NBARGS}" -eq "0" ]; then
manage_arg="0"
## If the first argument is not an option
if ! printf -- '%s' "${1}" | grep -q -E -- "^-+";
then
## Print help message and exit
printf '%b\n' "${RED}Invalid option: ${1}${RESET}"
printf '%b\n' "---"
usage
exit 1
fi
# Parse all options (start with a "-") one by one
while printf -- '%s' "${1}" | grep -q -E -- "^-+"; do
case "${1}" in
-a|--attempt|--attempts|-t|--tries ) ## Fix the maximum attempts
## Move to the next argument
shift
## Define max_attempts with this argument
max_attempts="${1}"
;;
-d|--debug ) ## debug
2020-09-24 15:18:09 +02:00
DEBUG=0
;;
-h|--help ) ## help
2020-09-24 15:18:09 +02:00
usage
## Exit after help informations
exit 0
;;
* ) ## unknow option
2020-09-24 15:18:09 +02:00
printf '%b\n' "${RED}Invalid option: ${1}${RESET}"
printf '%b\n' "---"
usage
exit 1
;;
esac
debug_message "Arguments management \
${RED}${1}${COLOR_DEBUG} option managed."
## Move to the next argument
shift
manage_arg=$((manage_arg+1))
done
debug_message "Arguments management \
${RED}${manage_arg}${COLOR_DEBUG} argument(s) successfully managed."
else
debug_message "Arguments management \
No arguments/options to manage."
fi
# }}}
main
2020-06-15 13:49:25 +02:00
exit 255