apache-to-cloudwatch.sh

#!/bin/bash
MINUTES=10
#
# this script pulls the last $MINUTES minutes out of the apache logs
# so cron it to run every $MINUTES minutes
#
SERVER_ALIAS=$(hostname -s)

NAMESPACE="Apache"
METRIC_NAME="RequestsPerMinute"

WORKING_FOLDER="/var/log/apache2"
FILE_NAME="other_vhosts_access.log"
FULL_PATH="$WORKING_FOLDER/$FILE_NAME"
# 
# because cron is grossly inconsistent about which environment variables it persists
#
source $HOME/.rvm/scripts/rvm
PATH=$PATH:$HOME/.rvm/bin 

export JAVA_HOME=/usr/lib/jvm/java-6-sun
export JRE_HOME=$JAVA_HOME

export AWS_CREDENTIAL_FILE=$HOME/.aws-credentials
export AWS_CLOUDWATCH_HOME=$HOME/CloudWatch-1.0.13.4
export PATH=$PATH:$AWS_CLOUDWATCH_HOME/bin
# 
# awk: the date out of the log file
# cut: the year, month, day, hour, minute
# uniq: count up the uniq occurences
# awk: join the dates and their counts so the loop doesn't treat them as separate items in the list
# tail: just the last 10 items (latest 10 minutes)
#
SNIPPETS=$(awk '{print $5}' $FULL_PATH | cut -d: -f1-3 | uniq -c | awk '{print $1 $2}' | tail -n $MINUTES)
#
# this will return lines that look like this:
#
# 35[19/Jan/2013:06:29
# 
for snippet in $SNIPPETS; do
  # 
  # split the date and the count
  # 
  JUNK=($(echo $snippet | awk -F[ '{print $1, $2}'))
  COUNT=${JUNK[0]}
  RAW_DATE=${JUNK[1]}
  # 
  # convert the date format to what cloudwatch wants
  # 2013-01-21T13:00:00.000Z
  # 
  CONVERTED_DATE=$(ruby -rtime -e "puts Time.strptime('$RAW_DATE', '%d/%b/%Y:%H:%M').strftime('%Y-%m-%dT%H:%M:00.000Z')")
  # 
  # submit the count to cloudwatch
  # 
  CMD="mon-put-data --namespace $NAMESPACE --metric-name $METRIC_NAME --timestamp $CONVERTED_DATE --value $COUNT --dimensions VolumeId=$SERVER_ALIAS"
  echo $CMD
  $($CMD)
  # 
  # FIXME: ignore the last item (minute) in the list as it could be partial data
  # 
done