#!/bin/bash
# snapshot.sh
# Usage: set up a root cron to run once per <CRASH_THRESHOLD minutes
# current snapshots are in ~/.snapshot.cur
# snapshots from the last crash are in ~/.snapshot.crash.last
# snapshots from all previous crashes archived in ~/.snapshot.crash.date.time
# (the time listed will be the first run of this script after the crash)
# ---
# Written by: Jonathan Kelly (jkelly@rackspace.com)
# v1.02 (2-14-07)
# ---

# if the server has been up for less than CRASH_THRESHOLD minutes when called
# assume it has crashed (increase this if you want a longer interval between
# cron jobs)
CRASH_THRESHOLD=11

# retain this number of snapshots before rotating them out
RETENTION=6

# first we make sure the ~/.snapshot.cur exists
if [[ ! -d ~/.snapshot.cur ]]; then
    mkdir ~/.snapshot.cur
fi

# if the server has been up for less than CRASH_THRESHOLD minutes
# we need process the last set of logs before proceeding
uptime=`uptime|awk {'print $3'}`
uptime_unit=`uptime|awk {'print $4'}`
if [[ $uptime_unit == "min," && $uptime -lt $CRASH_THRESHOLD && ! -e ~/.snapshot.crash.processed ]]; then
    # ~/.snapshot.crash.last is a symlink to the directory containing the last crash
    # so we can remove it if it exists
    if [[ -e ~/.snapshot.crash.last ]]; then
        rm -f ~/.snapshot.crash.last        
    fi
    # move the current snapshot dir to ~/.snapshot.crash.$crash_date and symlink to newest crash
    crash_date=`date +%d.%b.%H:%M:%S`
    mv ~/.snapshot.cur ~/.snapshot.crash.$crash_date
    ln -s ~/.snapshot.crash.$crash_date ~/.snapshot.crash.last
    mkdir ~/.snapshot.cur
    # touch ~/.snapshot.crash.processed to ensure a single crash is only processed once
    touch ~/.snapshot.crash.processed
# remove ~/.snapshot.crash.processed once we're past the CRASH_THRESHOLD
elif [[ $uptime_unit == "min," && $uptime -gt $CRASH_THRESHOLD && -e ~/.snapshot.crash.processed ]]; then
    rm -f ~/.snapshot.crash.processed
fi


# the rotation
# delete the last file in the rotation if needed
if [[ -e ~/.snapshot.cur/snapshot.$RETENTION ]]; then
    rm -f ~/.snapshot.cur/snapshot.$RETENTION
fi

# increment the rest of the snapshots
i=$((RETENTION-1))
while [ $i -gt 0 ]
do
    if [[ -e ~/.snapshot.cur/snapshot.$i ]]; then
        mv ~/.snapshot.cur/snapshot.$i ~/.snapshot.cur/snapshot.$((i+1))
    fi
    i=$((i-1))
done

# SNAPSHOT TIME
# Change this stuff to whatever you want to end up in the snapshots
# make sure to redirect the output you want to ~/.snapshot.cur/snapshot.1

uptime > ~/.snapshot.cur/snapshot.1
echo "---free -m---" >> ~/.snapshot.cur/snapshot.1
free -m >> ~/.snapshot.cur/snapshot.1
echo "---ps auxwwf---" >> ~/.snapshot.cur/snapshot.1
ps auxwwf >> ~/.snapshot.cur/snapshot.1
export LANG="english"
echo "---pstree---" >> ~/.snapshot.cur/snapshot.1
pstree >> ~/.snapshot.cur/snapshot.1
unset LANG
echo "---top -b---" >> ~/.snapshot.cur/snapshot.1
top -b -n1 >> ~/.snapshot.cur/snapshot.1