httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rob Hartill <hart...@ooo.lanl.gov>
Subject logging proposal
Date Sat, 27 May 1995 12:31:01 GMT

Something I thought up last night...

  I still believe that in order to make logging truely configurable,
which is what people are basically asking for now, we can't get bogged
down in fixing a standard string which Apache outputs.

  this proposal extends the earlier idea of sending a MIME-header-like 
list of loggable data, e.g

URL: /
Response: 200
Content-length: 1234
Remote-Host: foo.bar.com
etc

  Here's how it could work,


Apache child processes send loggable info to another process, each piece
of info will have an abbreviated key and the child process number,

e.g.  child #4 sends the Remote-Host..

4:RH:foo.bar.com

  The new idea here, is that Apache sends info as and when it discovers
it, and the logging process just collects it until it gets a special
"log it" input from the Apache process, e.g.


4:CL: 100                   (content-length aka bytes sent)
4:LOGIT


  The logging process can (for example) use a record structure to store
the info as it arrives, and moreover, it can ignore any fields it just
isn't interested in.

  By sending the log info piece by piece, the logging process
can process all the requests concurrently and probably with a more
even workload - it'll have a steady stream of input instead of bursts.

  In the event of a problem, e.g. the logging process not receiving the
LOGIT input, it can detect the problem since it'll be asked to overwrite
existing data. On detecting that, it can wipe the slate clean and start
again.


  Another special input can be used to inform the logging process of
abbreviation mappings, e.g.

-1:ABBR:RH=REMOTE-HOST,UA=USER-AGENT,RM=REQUEST_METHOD

  When the logging process sees these, it can update it's mappings.



Anyway, think it over.

--
Rob Hartill                           
http://nqcd.lanl.gov/~hartill/

Mime
View raw message