clfmerge - merge Common-Log Format web logs based on time-stamps

clfmerge [--help | -h] [-b size] [-d] [file names]

The clfmerge program is designed to avoid using sort to merge multiple web log files. Web logs for big sites consist of multiple files in the >100M size range from a number of machines. For such files it is not practical to use a program such as gnusort to merge the files because the data is not always entirely in order (so the merge option of gnusort doesn't work so well), but it is not in random order (so doing a complete sort would be a waste). Also the date field that is being sorted on is not particularly easy to specify for gnusort (I have seen it done but it was messy).

This program is designed to simply and quickly sort multiple large log files with no need for temporary storage space or overly large buffers in memory (the memory footprint is generally only a few megs).

It will take a number (from 0 to n) of file-names on the command line, it will open them for reading and read CLF format web log data from them all. Lines which don't appear to be in CLF format (NB they aren't parsed fully, only minimal parsing to determine the date is performed) will be rejected and displayed on standard-error.

If zero files are specified then there will be no error, it will just silently output nothing, this is for scripts which use the find command to find log files and which can't be counted on to find any log files, it saves doing an extra check in your shell scripts.

If one file is specified then the data will be read into a 1000 line buffer and it will be removed from the buffer (and displayed on standard output) in date order. This is to handle the case of web servers which date entries on the connection time but write them to the log at completion time and thus generate log files that aren't in order (Netscape web server does this - I haven't checked what other web servers do).

If more than one file is specified then a line will be read from each file, the file that had the earliest time stamp will be read from until it returns a time stamp later than one of the other files. Then the file with the earlier time stamp will be read. With multiple files the buffer size is 1000 lines or 100 * the number of files (whichever is larger). When the buffer becomes full the first line will be removed and displayed on standard output.

-b buffer-size

Specify the buffer-size to use, if 0 is specified then it means to disable the sliding-window sorting of the data which improves the speed.

-d

Set domain-name mangling to on. This means that if a line starts with as the name of the site that was requested then that would be removed from the start of the line and the GET / would be changed to GET http://www.company.com/ which allows programs like Webalizer to produce good graphs for large hosting sites. Also it will make the domain name in lower case.

0 No errors

1 Bad parameters

2 Can't open one of the specified files

3 Can't write to output

This program, its manual page, and the Debian package were written by Russell Coker <russell@coker.com.au>.

clfsplit(1),clfdomainsplit(1)