httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Ford <david+c...@blue-labs.org>
Subject Re: Parent death should force children suttee
Date Wed, 30 Jan 2002 21:22:44 GMT
I don't know about other sites, but the servers I manage tend to start 
showing oddness rather quickly when something goes wrong.  I would 
rather have a clean and quick restart than have a degradation in 
service.  With that said, I don't know of many people who would rely on 
this feature.

As a side note, a "feature" I relied on many many moons ago (short story 
follows), I accidently deleted the httpd binary.  This was back in the 
days when virtual hosting wasn't heard of yet and it was a patched set 
of source we'd tailored to do virtuals that had been lost in a harddrive 
crash.  We ran that server for 6 months without an httpd binary.  When 
power was lost, we shut down -every- other service and machine to make 
sure this one kept going.

Eventually the Apache group released code that supported virtuals, we 
recompiled and upgraded.  But until then, that was our little wonder. 
 We shoved that in the face of the M$ people for years (we were 
proponents of Linux).

This was also the machine that set a few records back in the day for 
webhosting and bandwidth served.

Ok, fireside chat time over.  Yes, I think there is a Proper (tm) way to 
do this that can be graceful and prompt as well as reacting consistently 
so that monitoring services don't need to be rewritten to support quirks.

David

Bill Stoddard wrote:

>
>>Any important site that relies on uptime -should- have some thought
>>already put into it.  If something fails, some procedure immediately
>>takes steps to fix it, i.e. a watchdog that restarts apache.
>>
>
>Yep.
>
>>If you don't have something like this in place and you are relying so
>>heavily on uptime, you are foolishly dancing on the edge of the cliff.
>>
>Yep!!
>
>>I vote for doing The Right Thing.  It isn't very hard to restart apache.
>> But I do recommend a N second delay or some method to gracefully
>>terminate as many connections as possible so the world doesn't see any
>>problems.
>>
>>David
>>
>
>Okay, Uncle! I know when I've been bested :-)
>
>If this change is made, it should have flashing lights attached to it in the CHANGES file.
>There may be people relying on this "feechur" without much thinking about it. Yea, you
can
>say they deserve what they get if this is their only line of defense, but good people
make
>mistakes occasionally. And Apache has an good track record for reliability and robstness,
>so some folks may just take it for granted.
>
>Bill
>


Mime
View raw message