httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Behlendorf <br...@organic.com>
Subject More on logging
Date Thu, 08 Jun 1995 17:32:15 GMT
On Thu, 8 Jun 1995, Randy Terbush wrote:
> > FIFO - as in, first child to tell it to log something, gets logged  ?
> 
> FIFO - as in named pipe. One concern that RST raised is giving
> the parent too much to do.  I agree with that concern. I think
> that giving the parent the responsibility to manage the
> transaction log for each child should not be too much of a load.
> The "write to FIFO" would mean that it would simply dump that
> information down the pipe *after* the child sends the "logit"
> request back to the parent.
> 
> I still believe that any hostname lookups, formating, logging to
> database, etc. should be handled by an external process reading from
> that pipe. This would also make it relatively easy to configure a
> CLF format and write it to a flat file for backward
> compatibility.
> 
> WRT the suggestion of the status monitor and an "internal"
> protocol to request that information, it would make sense
> to have the children using this same protocol to exchange the
> logging information, no?

Yes. Brilliant.  Also, if the logging process did *not* receive a "logit" 
after some length of time (Rob suggested ETIME?) then all that info would 
be written to the error log, making debugging a snap.

Can the logging process keep track of which child is sending it a 
message, if they all have the same pipe open?  How would this be done in 
Perl?  If not, some sort of unique ID needs to be transmitted so that the 
logging process can assemble the record back together.  

> > I still think the children should send pieces of loggable info, as 
> > and when they feel like, and in no particular order. This'll make
> > configuration of logging *much* easier in the long term.
> > IMO, CLF II should define how the info is passed... preferably in
> > some abbreviated form..
> 
> I agree with this model. I would however like to make the issue
> of whether or not to send headers configureable. Should be easy
> to toggle.

I'm concerned about overhead, but we need a bake-off to know for sure.

	Brian

--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
brian@organic.com  brian@hyperreal.com  http://www.[hyperreal,organic].com/


Mime
View raw message