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 SELF="`basename "$0"`[$$]"
  45 DELTA_WAIT_SUCCESS=$(( MAX_WAIT_SUCCESS - MIN_WAIT_SUCCESS ))
  46 MYHOSTNAME=$(hostname -f)
  47
  48 if [ -t 0 ]; then
  49     verbose=1
  50 else
  51     verbose=0
  52 fi
  53
  54 log() {
  55     [ "$verbose" -gt 0 ] && echo "$*"
  56     logger -p daemon.info -t "$SELF" "$*"
  57 }
  58 format_timedelta() {
  59     local secs="$1"; shift
  60     if [ "$secs" -ge 86400 ]; then
  61         printf '%d+%02d:%02d:%02d\n' $(($secs/3600/24)) $(($secs/3600%24)) $(($secs/60%60)) $(($secs%60))
  62     else
  63         printf '%02d:%02d:%02d\n' $(($secs/3600)) $(($secs/60%60)) $(($secs%60))
  64     fi
  65 }
  66
  67
  68 if [ "${1:-}" = "-h" ] || [ "${1:-}" = "--help" ]; then
  69     echo "Usage: $0 [<host>:<port>]"
  70     exit 0
  71 fi
  72
  73 if [ "$#" -gt 0 ]; then
  74     forcehostport="$1"
  75     shift
  76 else
  77     forcehostport=""
  78 fi
  79
  80 mkdir -p "$STATEDIR"
  81
  82 # get a lock, but only if we did not force the run
  83 if [ -z "$forcehostport" ]; then
  84     exec 200< "$STATEDIR"
  85     if ! flock -w 0 -e 200; then
  86         log "Cannot acquire lock on $STATEDIR."
  87         exit 0
  88     fi
  89 fi
  90
  91 while read host port username  cluster version; do
  92     [ "${host#\#}" = "$host" ] || continue
  93     [ -z "$host" ] && continue
  94
  95     flagfile="$STATEDIR/$host-$port.last-attempt"
  96     flagfilesuccess="$STATEDIR/$host-$port.last-success"
  97     if [ -n "$forcehostport" ]; then
  98         if [ "$forcehostport" != "$host:$port" ]; then
  99             log "Skipping $host:$port $version/$cluster because this run is limited to $host:$port."
 100             runme=0
 101         else
 102             log "Forcing $host:$port $version/$cluster run."
 103             runme=1
 104         fi
 105     else
 106         if ! [ -e "$flagfile" ]; then
 107             runme=1
 108             log "Planning to run $host:$port $version/$cluster because no flag file exists."
 109         else
 110             now=$(date +%s)
 111             mtime="$(stat --printf "%Y" "$flagfile")"
 112             delta=$(( now - mtime ))
 113             if [ "$delta" -lt "$MIN_WAIT" ]; then
 114                 runme=0
 115                 log "Skipping $host:$port $version/$cluster because last attempt was only $(format_timedelta "${delta}") (< $(format_timedelta "${MIN_WAIT}")) ago."
 116             else
 117                 if ! [ -e "$flagfilesuccess" ]; then
 118                     runme=1
 119                     log "Planning to run $host:$port $version/$cluster because no success flag exists."
 120                 else
 121                     mtime="$(stat --printf "%Y" "$flagfilesuccess")"
 122                     delta=$(( now - mtime ))
 123                     if [ "$delta" -lt "$MIN_WAIT_SUCCESS" ]; then
 124                         runme=0
 125                         log "Skipping $host:$port $version/$cluster because last success was only $(format_timedelta "${delta}") (< $(format_timedelta "${MIN_WAIT_SUCCESS}")) ago."
 126                     elif [ "$delta" -gt "$MAX_WAIT_SUCCESS" ]; then
 127                         runme=1
 128                         log "Planning to run $host:$port $version/$cluster because last success was $(format_timedelta "${delta}") (>= $(format_timedelta "${MAX_WAIT_SUCCESS}")) ago."
 129                     else
 130                         # get a "randomish" but stable value for this backup run
 131                         val=$(echo "$MYHOSTNAME-$host-$port-$mtime" | sha256sum | head -c 8)
 132                         val=$((16#$val))
 133                         rnd_cuttoff=$(($val % $DELTA_WAIT_SUCCESS))
 134                         age_after_min=$((delta - MIN_WAIT_SUCCESS))
 135                         if [ "$age_after_min" -lt "$rnd_cuttoff" ]; then
 136                             runme=0
 137                             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}"))."
 138                         else
 139                             runme=1
 140                             log "Planning to run $host:$port $version/$cluster because random computer says so ($(format_timedelta "${age_after_min}") >= $(format_timedelta "${rnd_cuttoff}"))."
 141                         fi
 142                     fi
 143                 fi
 144             fi
 145         fi
 146     fi
 147
 148     if [ "$runme" -gt 0 ]; then
 149         touch "$flagfile"
 150         exec 201< "$flagfile"
 151         if flock -w 0 -e 201; then
 152             log "Running $host:$port $version/$cluster."
 153             /usr/local/bin/postgres-make-one-base-backup "$host" "$port" "$username" "$cluster" "$version"
 154             rc=$?
 155             [ "$rc" = 0 ] && touch "$flagfilesuccess"
 156             flock -u 201
 157         else
 158             log "Cannot acquire lock on $flagfile, skipping $host:$port $version/$cluster."
 159         fi
 160     fi
 161 done << EOF
 162 seger.debian.org        5432    debian-backup           dak             9.6
 163 bmdb1.debian.org        5435    debian-backup           main            9.6
 164 bmdb1.debian.org        5436    debian-backup           wannabuild      9.6
 165 bmdb1.debian.org        5440    debian-backup           debsources      9.6
 166 fasolo.debian.org       5433    debian-backup           dak             9.6
 167 sibelius.debian.org     5433    debian-backup           snapshot        9.4
 168 sallinen.debian.org     443     debian-backup           snapshot        9.6
 169 <%- if @hostname != "backuphost" -%>
 170 moszumanska.debian.org  5432    debian-backup           main            9.1
 171 <%- end -%>
 172 #
 173 # puppet notice:  this is just a partial file.  The tail EOF comes
 174 # from a different concat fragment