bundles/backup-client: rework backup generation

bundles/backup-client/files/check_backup_last_run

@@ -1,42 +1,28 @@
-#!/usr/bin/env python3
+#!/bin/bash
 
-from os.path import getmtime, isfile
-from sys import exit
-from time import time
+statusfile="/var/tmp/backup.monitoring"
 
-statusfile = '/var/tmp/backup.monitoring'
-if not isfile(statusfile):
-    print('Status file not found')
-    exit(3)
+if [[ ! -r "$statusfile" ]]
+then
+    echo "cannot read $statusfile"
+    exit 3
+fi
 
-mtime = getmtime(statusfile)
-now = time()
+. "$statusfile"
 
-if now-mtime > 60*60*24*2:
-    print('Status file is older than 2 days!')
-    exit(3)
+if [[ -z "$msg" ]] || [[ -z "$status" ]] || [[ -z "$timestamp" ]]
+then
+    echo "status file is corrupt, cannot read status"
+    exit 3
+fi
 
-with open(statusfile, 'r') as f:
-    status = f.read().splitlines()
+two_days_ago=$(($(date +%s) - 86400*2))
 
-exitcode = status[0].strip()
+if [[ $timestamp -lt $two_days_ago ]]
+then
+    echo "last saved status is older than two days"
+    exit 2
+fi
 
-if exitcode == 'ok':
-    print('OK')
-    exit(0)
-elif exitcode == 'rsync_error':
-    print('rsync produced some errors, exit codes were:')
-    for line in status[1:]:
-        print(line)
-    exit(2)
-elif exitcode == 'hook':
-    print('run-parts /etc/backup-pre-hook.d failed with exit code {}'.format(status[1]))
-    exit(2)
-elif exitcode == 'abort_no_key':
-    print('no ssh key found in /etc/backup.priv!')
-    exit(1)
-else:
-    # garbage in file
-    for line in status:
-        print(line)
-    exit(3)
+echo "$msg"
+exit "$status"

bundles/backup-client/files/generate-backup

@@ -1,77 +1,121 @@
 #!/bin/bash
 
-statusfile=/var/tmp/backup.monitoring
+statusfile="/var/tmp/backup.monitoring"
+logdir="/var/log/backup-client"
+lock="/tmp/backup-client-is-running"
 ssh_login="${username}@${server}"
-ssh_cmnd="ssh -o IdentityFile=/etc/backup.priv -o StrictHostKeyChecking=accept-new -p ${port}"
+ssh_opts="-o IdentityFile=/etc/backup.priv -o StrictHostKeyChecking=accept-new -p ${port}"
 nodename="${node.name}"
 
 <%text>
+try="${1:-<unknown>}"
 [[ -n "$DEBUG" ]] && set -x
-NL=$'\n'
-
-if ! [[ -f /etc/backup.priv ]]
-then
-    echo "/etc/backup.priv does not exist" | logger -t backup-client -p user.error
-    echo "abort_no_key" > "$statusfile"
-    exit 1
-fi
-
-run-parts --exit-on-error -- /etc/backup-pre-hooks.d
-exitcode=$?
-if [[ $exitcode != 0 ]]
-then
-    echo "run-parts /etc/backup-pre-hooks.d exited $exitcode" | logger -t backup-client -p user.error
-    echo "hook $exitcode" > "$statusfile"
-    exit 1
-fi
 
 do_backup() {
-    rsync_errorcodes_for_this_path=""
-    backup_has_successfully_run="no"
-
-    for try in {1..5}
-    do
-        echo "Backup for '$1', try $try ..." | logger -t backup-client -p user.info
+    echo "==> starting backup for '$1'"
 
     # Compress level 1 is a good compromise between speed and cpu usage.
     rsync --compress-level=1 -aAP --numeric-ids --delete --relative \
         --rsync-path="/usr/bin/rsync --fake-super" \
-            -e "$ssh_cmnd" \
+        -e "ssh $ssh_opts" \
         "$1" "$ssh_login":backups/
 
     # Exit code 24 means some files have vanished during rsync.
     # I don't know why, but this is very common, apparently?
     exitcode=$?
-        echo "Backup for '$1' try $try exited $exitcode" | logger -t backup-client -p user.info
+    echo "==> backup for '$1' exited $exitcode"
     if [[ $exitcode != 0 ]] && [[ $exitcode != 24 ]]
     then
-            rsync_errorcodes_for_this_path+=" $exitcode"
-            sleep 30
-        else
-            backup_has_successfully_run="yes"
-            break
-        fi
-    done
-
-    if [[ "$backup_has_successfully_run" != "yes" ]]
-    then
-        echo "Backup for '$1' did not succeed!" | logger -t backup-client -p user.error
-        rsync_errors+="${NL}${1}${rsync_errorcodes_for_this_path}"
+        rsync_errors+=" $1 ($exitcode)"
     fi
 }
 
-rsync_errors=""
+on_exit() {
+    rmdir "$lock"
+    echo "*** END BACKUP RUN $(date '+%F %T %z') ***"
+}
 
-$ssh_cmnd $ssh_login "sudo /usr/local/bin/rotate-single-backup-client $nodename"
+prepare_and_cleanup_logdir() {
+    # rsync logs tend to get very large. That's why we pipe them through
+    # gzip when writing. Because we're running multiple tries, we cannot
+    # rely on logrotate to rotate the logs, we have to do it ourselves.
+    # Of course that means we have to clean up after ourselves, too.
+    mkdir -p "$logdir"
+    find "$logdir" -type f -mtime +14 -name "*.log" -delete
+    find "$logdir" -type f -mtime +14 -name "*.gz" -delete
+}
+
+save_result_for_monitoring() {
+    code=$1
+    msg=$2
+    printf "status=%q\n" "$code" > "$statusfile"
+    printf "msg=%q\n" "$msg" >> "$statusfile"
+    printf "timestamp=%q\n" "$(date +%s)" >> "$statusfile"
+}
+
+if ! mkdir "$lock" >/dev/null 2>&1
+then
+    save_result_for_monitoring 2 "could not get lock"
+    exit 1
+fi
+trap "on_exit" EXIT
+
+# redirect stdout and stderr to logfile
+prepare_and_cleanup_logdir
+logfile="$logdir/backup--$(date '+%F--%H-%M-%S')--$$.log.gz"
+echo "All log output will go to $logfile" | logger -it backup-client
+exec > >(gzip >"$logfile")
+exec 2>&1
+
+# this is where the real work starts
+ts_begin=$(date +%s)
+
+echo "*** BEGIN BACKUP RUN $(date '+%F %T %z') ***"
+echo "This is attempt $try"
+echo "using ssh options [$ssh_opts]"
+echo "using ssh login [$ssh_login]"
+
+if ! [[ -f /etc/backup.priv ]]
+then
+    save_result_for_monitoring 2 "/etc/backup.priv does not exist"
+    exit 100
+fi
+
+for i in /etc/backup-pre-hooks.d/*
+do
+    [[ -x "$i" ]] || continue
+
+    echo "Running pre-hook '$i'"
+    if ! $i
+    then
+        save_result_for_monitoring 2 "pre-hook '$i' failed to run"
+        exit 1
+    fi
+done
+
+rsync_errors=""
 </%text>
 
 % for path in sorted(paths):
 do_backup "${path}"
 % endfor
 
+<%text>
 if [[ -n "$rsync_errors" ]]
 then
-    echo "rsync_error$rsync_errors" > "$statusfile"
-else
-    echo "ok" > "$statusfile"
+    save_result_for_monitoring 2 "rsync failed: $rsync_errors"
+    exit 1
 fi
+
+ssh $ssh_opts $ssh_login "sudo /usr/local/bin/rotate-single-backup-client $nodename" </dev/null
+ssh_error=$?
+if [[ $ssh_error -ne 0 ]]
+then
+    save_result_for_monitoring 2 "rotating backups failed with status code $ssh_error"
+    exit 1
+fi
+
+ts_end=$(date +%s)
+echo "Success"
+save_result_for_monitoring 0 "Backup finished at $(date '+%F %T %z') (took $((ts_end - ts_begin)) seconds)"
+</%text>

bundles/backup-client/files/generate-backup-with-retries

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Try generating a backup multiple times. If one attempt succeeds, we're
+# done. If not, there will be logs for every attempt, plus monitoring
+# will read the result of the last backup.
+for try in {1..3}
+do
+    generate-backup "$try"
+    exitcode=$?
+
+    if [[ $exitcode -eq 100 ]]
+    then
+        # fatal error, cannot recover
+        exit 1
+    elif [[ $exitcode -eq 0 ]]
+    then
+        # successful backup
+        exit 0
+    else
+        sleep 60
+    fi
+done

bundles/backup-client/files/logrotate.conf

@@ -1,10 +0,0 @@
-/var/log/backup-client/*.log {
-    compress
-    copytruncate
-    daily
-    dateext
-    missingok
-    notifempty
-    rotate 14
-    sharedscripts
-}

bundles/backup-client/items.py

@@ -18,12 +18,25 @@ else:
     backup_paths = node.metadata.get('backups/paths', set())
 
 if node.metadata.get('backups/exclude_from_backups', False):
-    files['/etc/backup.priv'] = {
+    # make sure nobody tries to do something funny
+    for file in [
+        '/etc/backup.priv',
+        '/usr/local/bin/generate-backup',
+        '/usr/local/bin/generate-backup-with-retries',
+        '/var/tmp/backup.monitoring', # status file
+    ]:
+        files[file] = {
             'delete': True,
         }
+
 else:
     backup_target = repo.get_node(node.metadata.get('backup-client/target'))
 
+    files['/etc/backup.priv'] = {
+        'content': repo.vault.decrypt_file(join('backup', 'keys', f'{node.name}.key.vault')),
+        'mode': '0400',
+    }
+
     files['/usr/local/bin/generate-backup'] = {
         'content_type': 'mako',
         'context': {
@@ -35,9 +48,8 @@ else:
         'mode': '0700',
     }
 
-    files['/etc/backup.priv'] = {
-        'content': repo.vault.decrypt_file(join('backup', 'keys', f'{node.name}.key.vault')),
-        'mode': '0400',
+    files['/usr/local/bin/generate-backup-with-retries'] = {
+        'mode': '0700',
     }
 
 files['/usr/local/share/icinga/plugins/check_backup_last_run'] = {
@@ -45,15 +57,13 @@ files['/usr/local/share/icinga/plugins/check_backup_last_run'] = {
 }
 
 files['/etc/logrotate.d/backup-client'] = {
-    'source': 'logrotate.conf',
+    'delete': True,
 }
 
 directories['/etc/backup-pre-hooks.d'] = {
     'purge': True,
 }
 
-directories['/var/log/backup-client'] = {}
-
 for hname, hcontent in node.metadata.get('backup-client/pre-hooks', {}).items():
     files[f'/etc/backup-pre-hooks.d/50-{hname}'] = {
         'content': '#!/bin/sh\n\n' + hcontent,

bundles/backup-client/metadata.py

@@ -20,7 +20,7 @@ def cron(metadata):
     return {
         'cron': {
             # spread backups between 00:00 and 04:59 UTC
-            'backup': '{} {} * * *    root    /usr/local/bin/generate-backup > /var/log/backup-client/backup.log 2>&1'.format(
+            'backup': '{} {} * * *    root    /usr/local/bin/generate-backup-with-retries'.format(
                 (node.magic_number % 60),
                 (node.magic_number % 4),
             ),