httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Behlendorf <>
Subject Re: graceful restarts & log file rotation again
Date Tue, 09 Sep 1997 00:00:53 GMT
At 03:12 PM 9/8/97 -0700, you wrote:
>See htdocs/manual/stopping.html, this has been documented already :) 
>You probably had one child busy still writing to it, processing some
>hugely long request or something.  Solutions include reliable piped logs
>(ugh, yeah, that thing dean still hasn't finished), or the log file
>locking I've mentioned a few times. 

I did read it; it says wait about 15-20 minutes as a child may still be
active and will need to write a transaction.  However, this was 12 hours
later, and according to the scoreboard there were no children in "G" state.
 So, that feels like a bug.  Certainly someone may have been transferring
the CVS repository from Antarctica and that would have been going since
last night; but I didn't see anything on inspection.  I'll try looking at
it again tonight.

I have a few ideas on things that might help: when a -USR1 is sent to the
parent and the children are told to go into graceful restart, perhaps 

1) the logfiles could close and the new file descriptors for the new logs
could be passed to the children?

2) httpd children could be told to turn off persistant HTTP, thus ensuring
that the current request is the very last one served?

I forget, does your reliable piped logs solution involve sending the log
messages to the parent and having the parent do the writing?


"Why not?" - TL  - -

View raw message