report-state.sh

#!/usr/bin/env bash

set -o pipefail

archives_dir='archives'
default_server='unknown-host'
services_url='https://raw.githubusercontent.com/GreatMedivack/files/master/list.out'

log() {
    echo 1>&2 "$script_name: $*"
}

abort() {
    log 'ERROR: aborting due to failed command.'
    exit 1
}

# Select rows from $services_file, using the argument as an awk(1) condition.
# Then print the names of matching services,
# with trailing ids (-??????????-?????) removed.
select_services() {
    local condition=$1
    awk "$condition {print \$1}" "$services_file" | sed -E 's/-[^-]{10}-[^-]{5}$//'
}

# main():

# To access the script's $0 from log().
script_name=$0

if [[ $# -eq 0 ]]; then
    log "WARN: server name not provided, using '$default_server' as the default."
    server=$default_server
elif [[ $# -eq 1 ]]; then
    server=$1
else
    log "ERROR: too many arguments (expected 0 or 1, but got $#)."
    exit 1
fi

log "INFO: getting the current date..."
current_date=$(date '+%d_%m_%Y') || abort

services_file="${server}_${current_date}_list.out"
log "INFO: downloading the list of services into '$services_file' ..."
wget "$services_url" -O "$services_file" || abort

failed_services_file="${server}_${current_date}_failed.out"
log "INFO: saving failed services to '$failed_services_file' ..."
# shellcheck disable=SC2016 # $3 is an awk variable, not a bash variable.
select_services '($3 == "Error" || $3 == "CrashLoopBackOff")' > "$failed_services_file" || abort

running_services_file="${server}_${current_date}_running.out"
log "INFO: saving running services to '$running_services_file' ..."
# shellcheck disable=SC2016 # $3 is an awk variable, not a bash variable.
select_services '$3 == "Running"' > "$running_services_file" || abort

report_file="${server}_${current_date}_report.out"
log "INFO: generating the report into '$report_file' ..."
# Need to pre-calculate the values outside of the heredoc,
# in order for abort() to work properly.
n_running=$(wc -l < "$running_services_file") || abort
n_failed=$(wc -l < "$failed_services_file") || abort
username=$(id -un) || abort
cat << EOF > "$report_file" || abort
- Количество работающих сервисов: $n_running
- Количество сервисов с ошибками: $n_failed
- Имя системного пользователя: $username
- Дата: $current_date
EOF
chmod a+r "$report_file" || abort

# It's not clear whether $services_file should be there, added it just in case.
archived_files=("$services_file" "$failed_services_file" "$running_services_file" "$report_file")
if [[ ! -e "$archives_dir" ]]; then
    log "WARN: archives directory '$archives_dir' doesn't exist, creating it..."
    mkdir "$archives_dir" || abort
fi
archive_file="$archives_dir/${server}_$current_date"
log "INFO: archiving gathered files into '$archive_file' ..."
# IMO, this should be checked at the very beginning of the script to avoid
# needless work, but I've implemented the task verbatim just in case.
if [[ -e "$archive_file" ]]; then
    log "WARN: file already exists, skipping this step."
else
    tar -czf "$archive_file" "${archived_files[@]}" || abort
fi

log "INFO: deleting gathered files..."
rm "${archived_files[@]}" || abort

log "INFO: checking archive integrity..."
tar -tzf "$archive_file" > /dev/null || abort

log "INFO: done, exiting."