@anonymous-piwik-user opened this Issue on March 8th 2013

There's broken parser since 1.11

LogFormat "%h %l %u %t \"%r\" %>s %O \"%{Referer}i\" \"%{User-Agent}i\" %I %O" combined - - [07/Mar/2013:08:46:03 +0100] "GET /19.pdf HTTP/1.0" 200 4324023 "-" "Apache-HttpClient/4.2.1 (java 1.5)" 276 4324023 - - [07/Mar/2013:05:12:44 +0100] "GET /index.htm HTTP/1.1" 206 33106 "http://refer" "Mozilla/5.0 (Windows NT 5.1) AppleWebKit/537.22 (KHTML, like Gecko) Chrome/25.0.1364.152 Safari/537.22" 437 33106


Direct Entry http://somesite/19.pdf
Website: refer http://somesite/index.htm


Direct Entry http://somesite/19.pdf%20HTTP/1.0%22%20200%204324023%20%22-%22%20%22Apache-HttpClient/4.2.1%20(java
Direct Entry http://somesite/
@anonymous-piwik-user commented on March 11th 2013


@mattab commented on March 11th 2013 Member

In 2debefdd0e8005fd1ab3ecab855cba070f717b5f: Refs #3805 Adding test and it's working OK so cannot replicate problem...

@mattab commented on March 11th 2013 Member

please submit failing test or more info as it seems to work https://github.com/piwik/piwik/commit/2debefdd0e8005fd1ab3ecab855cba070f717b5f#L0R14

@anonymous-piwik-user commented on March 11th 2013

Can you try import attached gentoo.log directly?

It fails for me with default import_logs.py. When I change -

        self.regex = re.compile(regex + '\s*$') # make sure regex includes end of line

back to what was there in 1.10.1 it works

        self.regex = re.compile(regex)

so I guess the problem is somewhere there, but I cannot figure out what.


@diosmosis commented on March 16th 2013 Member

@matt, I found the issue w/ this bug and uploaded a fix in https://github.com/piwik/piwik/compare/master...3805, can you review? Here's an explanation of the fix:

Seems that the change of order to the format exposed a couple bugs. First the change I committed before isn't flexible enough w/ log files (like gentoo.log which is ncsa_extended w/ two extra fields). Also, regex quantifiers are greedy, so now that common gets tested before ncsa_extended, 'common' matches w/ incorrect groups and ncsa_extended is never tried.

I've reverted my original change (the '\s*$' one) and modified the format autodetection logic to use the format that returns the most groups (ie, the one that matches and returns the most information). I've also modified the regexes to account for regex greediness. There are some extra tests and I fixed a bug w/ the S3 regex.

@mattab commented on March 16th 2013 Member

all looks good to me!

@diosmosis commented on March 23rd 2013 Member

@matt Ok, greediness was only an issue w/ my change, so I've reverted that and added some more tests. I've pushed it again so you can take a look if you want. Will commit tomorrow.

@diosmosis commented on March 23rd 2013 Member

In 62b43d8844699cb78f533643fa8c491d4fac0430: Fixes #3805, reverted change in log importer that looked for end-of-line after format regex match and modified format autodetection logic to pick the format based on whether the format matches and the number of groups returned in the match.


  • Added several more tests to log importer tests.py. Added tests for checking format of log files w/ extra junk info on log lines. Added individual tests for parsing regex format.
  • Modified log files used in ImportLogs test, added extra junk info to end of some lines.
  • Fixed failing test in tests.py for the S3 log file format.
This Issue was closed on May 27th 2013
Powered by GitHub Issue Mirror