Apache, Nginx and the like log every request your web server processes, unless you’ve configured them not to. Whilst statistics packages such as Awstats, Webalizer, Google Analytics and friends provide a useful overview there’s nothing that beats the raw data for being able to answer your own questions. Here, I introduce an awk script you can use to get your own analysis going.

There’s a useful article on The Art of Web, that covers the basics of awk and gives some good ideas on what you might do with it: Analysing Apache log files. However, if the URL contains spaces, or the user id field is wrapped by double-quotes, the status code is no longer $9… I ran into this problem for a small number of requests, but wanted to include them.

The following awk script handles the above cases and filters the log into a number of other (compressed) files, one for each HTTP status code and request method. This makes it much easier to then analyse just the 403 errors for example and by using tabs to separate the filtered format of the files, we can make it much easier to get accurate results from subsequent awk runs.

You’re welcome to use this script too. Download it: filter.awk or fork it on github: lapsedtheorist/awk-for-apache-nginx-logs.

If you have a directory full of logs with date-base filenames and want to investigate everything that happened in January, for example, this may be of use:

for f in /path/to/logs/access.log-201201*; do \
    if [ "${f:(-3)}" = ".gz" ]; then gunzip -c $f; else cat $f; fi; \
done | awk -f filter.awk

The above deals with both gzipped and plain text files at the same time. Do update the path and the filename to suit your server, of course.

#	Web server log file analysis & filtering
#	v1.2; Oct 2012
#	Ben Carpenter
#	http://www.bencarpenter.co.uk/awk-for-apache-nginx-logs
#	This awk script processes lines from a log format that matches the
#	'combined' log often used by the Apache and Nginx web servers. If your log
#	file format is different, amend accordingly, but for reference this is the
#	combined format this script expects by default:
#		%h %l %u %t "%r" %>s %b "%{Referer}i" "%{User-agent}i"
#		%h		Remote host
#		%l		Remote logname (ignored)
#		%u		Remote user (ignored)
#		%t		Date and time of the request
#		%r		First line of the request, typically "GET /something HTTP/1.1"
#		%>s		Status
#		%b		Size of response in bytes
#	It tries to be efficient on resources, so there's minimal progress messages
#	and no system commands in the main loop other than writing to a file based
#	on the status code. The output files are written in a simplified
#	tab-separated format, error corrected for some strange things like spaces
#	in URLs and double quotes for the userid. This revised format is easier to
#	pass reliably through other awk scripts when filtering for specific data,
#	etc. The file format is:
#		IP, Date/Time, Method, URL, Status, Size, Referer, User Agent
#	You should be able to send a large (>1GB) amount of log data through this
#	script quite comfortably. This works well for me, but usual clauses apply
#	(use it at your own risk, etc.). Bug reports and suggestions for
#	improvements are very welcome
	FS="( \"|\" )"
	printf "%s", intro

	split($1, a, " ")
	# It seems some browsers/bots set the 'user' part to the blank string,
	# double quoted, which is therefore something that can foul our detection
	# for the status code, unless we explicitly look for it
	if($2!="") {
		datetime=a[4]" "a[5]
		split($3, c, " ")
	} else {
		split($3, b, " ")
		datetime=b[2]" "b[3]
		split($5, c, " ")
	if(match(code, /^[0-9]+$/)==0) {
		# This status code, whatever it is, isn't a number so let's set it to
		# UNKNOWN so it's obvious in the analysis that this is a dud

	# Analyse the request
	n=split(request, detail, " ")
	if(match(method, /^[A-Z]+$/)==0) {
		# This request method, whatever it is, doesn't 'look like' a request
		# method, so let's set it to UNKNOWN so it's obvious in the analysis
		# that this is a dud

	# We want the URL, but we need to handle the case where the URL contains
	# one or more space characters, even though they shouldn't be there
	for(i=2; i<n; i++) {
		url=(url" "detail[i])
	url=substr(url, 2)

	# Create and add to a file for each status code
	printf "%s\t%s\t%s\t%s\t%d\t%d\t%s\t%s\n", \
		ip, datetime, method, url, code, size, referer, useragent > file

	# Create and add to a file for each request method
	printf "%s\t%s\t%s\t%s\t%d\t%d\t%s\t%s\n", \
		ip, datetime, method, url, code, size, referer, useragent > file

	for(l=0; l<length(intro); l++) {
		printf "\b"
	printf "%d requests filtered\n", \

	# Write out some useful summary data
	printf "\n%-8s\t%11s\t%6s\t%s\n", \
			"status", "occurrences", "%", "output\tfile"
	for(code in statuses) {
		printf "%-8d\t%11d\t%6.2f\t", \
			code, statuses[code], (100*statuses[code]/total)
		# Close and compress each file, because they can be large
		system("gzip -f "file)
		system("du -sh "file".gz")
	printf "\n%-8s\t%11s\t%6s\t%s\n", \
		"method", "occurrences", "%", "output\tfile"
	for(method in methods) {
		printf "%-8s\t%11d\t%6.2f\t", \
			method, methods[method], (100*methods[method]/total)
		# Close and compress each file, because they can be large
		system("gzip -f "file)
		system("du -sh "file".gz")

	printf "\n"