Analyzing web logs with grep

If you have a website, your logs are of course a great source of information. For complete, in depth analysis, you can't beat Google Analytics, but you can actually get a lot of quick and useful stats from the command line with just "grep".

Please don't tell me your website isn't running on Linux or BSD or that you don't have shell access. It is completely unnecessary and foolish to be running a website on Microsoft, and if your problem is that your host won't give you shell access, you need to find another host.

Let's start with the simplistic approach: I want to extract log entries for yesterday's There's something about a Muntz TV post. Obviously "grep" can fo that; I just cd over to my logs directory and run:

 grep muntz_tv.html access_log
 

But that gives me too much:

 72.74.93.123 - - [16/Jan/2008:23:06:11 +0000] "GET /foo-web/muntz_tv.html HTTP/1.1" 200 21849 "-" "Mozilla/5.0 (Macintosh; U; Intel Mac OS X; en-US; rv:1.8.1.11) Gecko/20071127 Firefox/2.0.0.11"
 168.75.65.67 - - [16/Jan/2008:23:06:12 +0000] "GET /foo-web/muntz_tv.html HTTP/1.1" 200 21855 "-" "Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.0;.NET CLR 1.0.3705; ContextAd Bot 1.0)"
 66.249.72.165 - - [16/Jan/2008:23:06:13 +0000] "GET /foo-web/muntz_tv.html HTTP/1.1" 200 21868 "-" "Mediapartners-Google"
 72.74.93.123 - - [16/Jan/2008:23:06:15 +0000] "GET /foo-web.js HTTP/1.1" 200 3590 "http://aplawrence.com/foo-web/muntz_tv.html" "Mozilla/5.0 (Macintosh; U; Intel Mac OS X; en-US; rv:1.8.1.11) Gecko/20071127 Firefox/2.0.0.11"
 72.74.93.123 - - [16/Jan/2008:23:06:15 +0000] "GET /images/147.jpg HTTP/1.1" 200 2288 "http://aplawrence.com/foo-web/muntz_tv.html" "Mozilla/5.0 (Macintosh; U; Intel Mac OS X; en-US; rv:1.8.1.11) Gecko/20071127 Firefox/2.0.0.11"
 ..
 

For one thing, one of those IP addresses is mine. For another, I'm not interested in the "GET /foo-web.js" lines that are other files loaded with these accesses. So my next attempt improves things a bit:

 grep "muntz_tv.html HTTP"  access_log | sed /72.74.93.123/d
 

The addition of " HTTP" eliminates everything that wasn't actually a GET of the actual page, and the "sed" eliminates my ip address from the results (not my actual ip, by the way - just for illustrative purposes).

 168.75.65.67 - - [16/Jan/2008:23:06:12 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 21855 "-" "Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.0;.NET CLR 1.
 0.3705; ContextAd Bot 1.0)"
 66.249.72.165 - - [16/Jan/2008:23:06:13 +0000] "GET /foo-web/muntz_tv.html HTTP/
 1.1" 200 21868 "-" "Mediapartners-Google"
 128.30.52.13 - - [16/Jan/2008:23:09:07 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 22175 "-" "W3C_Validator/1.575"
 128.30.52.13 - - [16/Jan/2008:23:09:32 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 22428 "-" "W3C_Validator/1.575"
 128.30.52.13 - - [16/Jan/2008:23:10:37 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 22346 "-" "W3C_Validator/1.575"
 128.30.52.13 - - [16/Jan/2008:23:11:47 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 22056 "-" "W3C_Validator/1.575"
 128.30.52.13 - - [16/Jan/2008:23:13:14 +0000] "GET /foo-web/muntz_tv.html HTTP/1
 .1" 200 22004 "-" "W3C_Validator/1.575"
 ..
 

Of course I could pipe that through "wc -l" to get a count:

 $ grep "muntz_tv.html HTTP"  access_log | sed /72.74.93.123/d
   88
 

But that also includes a lot of repeated ip's. Those are easy to get rid of:

 $ grep "muntz_tv.html HTTP"  access_log | sed '/72.74.93.123/d;s/- - .*//' | sort -u |  wc -l
   59
 

Adding "s/- - .*//" to the "sed" eliminates everything on the line except the leading IP address, and sending that through "sort -u" eliminates duplicates.

What if we want to know how many of those people visited other pages? That's easy enough:

 grep "HTTP.*muntz_tv.html" access_log
 

is the base of it, because that gives us lines where "muntz_tv.html" is in the Referrer field: the visitor clicked on some link on that page. However, I need more. I'll still need to delete my own IP address and get rid of duplicates, but I also need to get rid of lines like this:

 66.199.242.10 - - [17/Jan/2008:13:31:17 +0000] "GET /foo-web/muntz_tv.html HTTP/1.0" 200 20384 "http://aplawrence.com/foo-web/muntz_tv.html" "Mozilla/4.0 (compatible; MSIE 6.0; Windows 98; Win 9x 4.90)"
 

That's someone who did a GET of the page from the page - in other words, a reload. To eliminate those, I add another "sed":

 grep "HTTP.*muntz_tv.html" access_log | sed '/muntz_tv.html HTTP/d'
 

Putting it all together:

 grep "HTTP.*muntz_tv.html" t | sed '/muntz_tv.html HTTP/d;/72.74.93.123/d' | grep "html HTTP"
 

That gives us the visitors who went to another page from there. I can pipe that through the "sed s'/- -.*//' and "wc -l" also.

How about getting the visitors who came to a page here from another page here? That sounds a little tricky, but t's actually not too bad:

 grep "HTTP.*aplawrence.com" access_log | grep "html HTTP" | sed '/72.74.93.123/d'
 

As you can see, simple tools can give useful information very quickly.