modules/postgres/templates/backup_server/postgres-make-base-backups.erb

   1 #!/bin/bash
   2
   3 # vim:syn=sh:
   4 # vim:ts=4:
   5 # vim:et:
   6
   7
   8 # run a bunch of full postgresql backups
   9 #  if given a host:port, run this backup,
  10 #  else run all defined once if they have not run recently
  11 # vim:syn=sh:
  12
  13
  14 # Copyright 2014 Peter Palfrader
  15 #
  16 # Permission is hereby granted, free of charge, to any person obtaining
  17 # a copy of this software and associated documentation files (the
  18 # "Software"), to deal in the Software without restriction, including
  19 # without limitation the rights to use, copy, modify, merge, publish,
  20 # distribute, sublicense, and/or sell copies of the Software, and to
  21 # permit persons to whom the Software is furnished to do so, subject to
  22 # the following conditions:
  23 #
  24 # The above copyright notice and this permission notice shall be
  25 # included in all copies or substantial portions of the Software.
  26 #
  27 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  28 # EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  29 # MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  30 # NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
  31 # LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  32 # OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  33 # WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  34
  35
  36 MIN_WAIT=$(( 60*60*4 ))
  37 MIN_WAIT_SUCCESS=$(( 60*60*24*7 ))
  38 MAX_WAIT_SUCCESS=$(( 60*60*24*10 ))
  39 STATEDIR=/var/lib/dsa/postgres-make-base-backups
  40
  41 ####
  42 set -u
  43
  44 if [ "$(id -u)" = 0 ]; then
  45     echo >&2 "Do not run me as root.  Probably you want sudo -u debbackup."
  46     exit 1
  47 fi
  48
  49 SELF="`basename "$0"`[$$]"
  50 DELTA_WAIT_SUCCESS=$(( MAX_WAIT_SUCCESS - MIN_WAIT_SUCCESS ))
  51 MYHOSTNAME=$(hostname -f)
  52
  53 if [ -t 0 ]; then
  54     verbose=1
  55 else
  56     verbose=0
  57 fi
  58
  59 log() {
  60     [ "$verbose" -gt 0 ] && echo "$*"
  61     logger -p daemon.info -t "$SELF" "$*"
  62 }
  63 format_timedelta() {
  64     local secs="$1"; shift
  65     if [ "$secs" -ge 86400 ]; then
  66         printf '%d+%02d:%02d:%02d\n' $(($secs/3600/24)) $(($secs/3600%24)) $(($secs/60%60)) $(($secs%60))
  67     else
  68         printf '%02d:%02d:%02d\n' $(($secs/3600)) $(($secs/60%60)) $(($secs%60))
  69     fi
  70 }
  71
  72
  73 if [ "${1:-}" = "-h" ] || [ "${1:-}" = "--help" ]; then
  74     echo "Usage: $0 [<host>:<port>]"
  75     exit 0
  76 fi
  77
  78 if [ "$#" -gt 0 ]; then
  79     forcehostport="$1"
  80     shift
  81 else
  82     forcehostport=""
  83 fi
  84
  85 mkdir -p "$STATEDIR"
  86
  87 # get a lock, but only if we did not force the run
  88 if [ -z "$forcehostport" ]; then
  89     exec 200< "$STATEDIR"
  90     if ! flock -w 0 -e 200; then
  91         log "Cannot acquire lock on $STATEDIR."
  92         exit 0
  93     fi
  94 fi
  95
  96 while read host port username  cluster version; do
  97     [ "${host#\#}" = "$host" ] || continue
  98     [ -z "$host" ] && continue
  99
 100     flagfile="$STATEDIR/$host-$port.last-attempt"
 101     flagfilesuccess="$STATEDIR/$host-$port.last-success"
 102     if [ -n "$forcehostport" ]; then
 103         if [ "$forcehostport" != "$host:$port" ]; then
 104             log "Skipping $host:$port $version/$cluster because this run is limited to $forcehostport."
 105             runme=0
 106         else
 107             log "Forcing $host:$port $version/$cluster run."
 108             runme=1
 109         fi
 110     else
 111         if ! [ -e "$flagfile" ]; then
 112             runme=1
 113             log "Planning to run $host:$port $version/$cluster because no flag file exists."
 114         else
 115             now=$(date +%s)
 116             mtime="$(stat --printf "%Y" "$flagfile")"
 117             delta=$(( now - mtime ))
 118             if [ "$delta" -lt "$MIN_WAIT" ]; then
 119                 runme=0
 120                 log "Skipping $host:$port $version/$cluster because last attempt was only $(format_timedelta "${delta}") (< $(format_timedelta "${MIN_WAIT}")) ago."
 121             else
 122                 if ! [ -e "$flagfilesuccess" ]; then
 123                     runme=1
 124                     log "Planning to run $host:$port $version/$cluster because no success flag exists."
 125                 else
 126                     mtime="$(stat --printf "%Y" "$flagfilesuccess")"
 127                     delta=$(( now - mtime ))
 128                     if [ "$delta" -lt "$MIN_WAIT_SUCCESS" ]; then
 129                         runme=0
 130                         log "Skipping $host:$port $version/$cluster because last success was only $(format_timedelta "${delta}") (< $(format_timedelta "${MIN_WAIT_SUCCESS}")) ago."
 131                     elif [ "$delta" -gt "$MAX_WAIT_SUCCESS" ]; then
 132                         runme=1
 133                         log "Planning to run $host:$port $version/$cluster because last success was $(format_timedelta "${delta}") (>= $(format_timedelta "${MAX_WAIT_SUCCESS}")) ago."
 134                     else
 135                         # get a "randomish" but stable value for this backup run
 136                         val=$(echo "$MYHOSTNAME-$host-$port-$mtime" | sha256sum | head -c 8)
 137                         val=$((16#$val))
 138                         rnd_cuttoff=$(($val % $DELTA_WAIT_SUCCESS))
 139                         age_after_min=$((delta - MIN_WAIT_SUCCESS))
 140                         if [ "$age_after_min" -lt "$rnd_cuttoff" ]; then
 141                             runme=0
 142                             log "Skipping $host:$port $version/$cluster because random computer says wait ([$(format_timedelta "${age_after_min}") < $(format_timedelta "${rnd_cuttoff}") (< $(format_timedelta "${DELTA_WAIT_SUCCESS}"))] + $(format_timedelta "${MIN_WAIT_SUCCESS}"))."
 143                         else
 144                             runme=1
 145                             log "Planning to run $host:$port $version/$cluster because random computer says so ($(format_timedelta "${age_after_min}") >= $(format_timedelta "${rnd_cuttoff}"))."
 146                         fi
 147                     fi
 148                 fi
 149             fi
 150         fi
 151     fi
 152
 153     if [ "$runme" -gt 0 ]; then
 154         touch "$flagfile"
 155         exec 201< "$flagfile"
 156         if flock -w 0 -e 201; then
 157             log "Running $host:$port $version/$cluster."
 158             /usr/local/bin/postgres-make-one-base-backup "$host" "$port" "$username" "$cluster" "$version"
 159             rc=$?
 160             log "Base backup for $host:$port $version/$cluster exited with rc $rc."
 161             [ "$rc" = 0 ] && touch "$flagfilesuccess"
 162             flock -u 201
 163         else
 164             log "Cannot acquire lock on $flagfile, skipping $host:$port $version/$cluster."
 165         fi
 166     fi
 167 done << EOF
 168 bmdb1.debian.org        5435    debian-backup           main            9.6
 169 bmdb1.debian.org        5440    debian-backup           debsources      9.6
 170 sallinen.debian.org     5473    debian-backup           snapshot        9.6
 171 #
 172 # puppet notice:  this is just a partial file.  The tail EOF comes
 173 # from a different concat fragment