CentOS - Jan 2010 - Logrotate in CentOS 5.4 more brutal (to httpd at least) than in 5.3?

I've just updated a few CentOS 5.3 servers to 5.4. One of them were a Apache
Webserver. Doing a diff/check on the new ".rpmnew" config files that
are made, I saw that the logrotate command for apache was changed. In 5.3 it did
a reload, but in 5.4 it does a hard kill:

CentOS 5.3:

/var/log/httpd/*log {
    missingok
    notifempty
    sharedscripts
    postrotate
        /sbin/service httpd reload > /dev/null 2>/dev/null || true
    endscript
}


CentOS 5.4:

/var/log/httpd/*log {
    missingok
    notifempty
    sharedscripts
    postrotate
        /bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2> /dev/null
|| true
    endscript
}

Isn't this a bit harsh in 5.4? Wouldn't the 5.3 method be better?

-Christopher Thorjussen

On Friday 15 January 2010, Christopher Thorjussen wrote:
> I've just updated a few CentOS 5.3 servers to 5.4. One of them were a
> Apache Webserver. Doing a diff/check on the new ".rpmnew" config
files that
> are made, I saw that the logrotate command for apache was changed. In 5.3
> it did a reload, but in 5.4 it does a hard kill:
5.4 does a "kill -HUP" which is what "service httpd reload"
does too. The -HUP
signal causes http to re-read stuff rather than shutting down.

Still, I don't really understand why 5.4 doesn't use "service
reload" which
seems more "correct" (and the init.d script does a little bit of
checking
before the kill -HUP too).

*shrug*
 Peter
> CentOS 5.3:
...
>     postrotate
>         /sbin/service httpd reload > /dev/null 2>/dev/null || true
> CentOS 5.4:
...
>     postrotate
>         /bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2>
/dev/null ||
> true endscript
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part.
URL:
<http://lists.centos.org/pipermail/centos/attachments/20100115/69d23564/attachment-0001.sig>

On Fri, Jan 15, 2010 at 11:35 AM, Christopher Thorjussen
<Christopher.Thorjussen at carrot.no> wrote:
> I've just updated a few CentOS 5.3 servers to 5.4. One of them were a
Apache Webserver. Doing a diff/check on the new ".rpmnew" config files
that are made, I saw that the logrotate command for apache was changed. In 5.3
it did a reload, but in 5.4 it does a hard kill:
>
> CentOS 5.3:
>
> /var/log/httpd/*log {
> ? ? ? ?/sbin/service httpd reload > /dev/null 2>/dev/null || true
> CentOS 5.4:
>
> /var/log/httpd/*log {
> ? ? ? ?/bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2>
/dev/null || true
strange, because in my systems it does service reload
logrotate-3.7.4-9

On Fri, Jan 15, 2010 at 4:35 AM, Christopher Thorjussen
<Christopher.Thorjussen at carrot.no> wrote:
> I've just updated a few CentOS 5.3 servers to 5.4. One of them were a
Apache Webserver. Doing a diff/check on the new ".rpmnew" config files
that are made, I saw that the logrotate command for apache was changed. In 5.3
it did a reload, but in 5.4 it does a hard kill:
>
> CentOS 5.3:
>
> /var/log/httpd/*log {
> ? ?missingok
> ? ?notifempty
> ? ?sharedscripts
> ? ?postrotate
> ? ? ? ?/sbin/service httpd reload > /dev/null 2>/dev/null || true
> ? ?endscript
> }
>
>
> CentOS 5.4:
>
> /var/log/httpd/*log {
> ? ?missingok
> ? ?notifempty
> ? ?sharedscripts
> ? ?postrotate
> ? ? ? ?/bin/kill -HUP `cat /var/run/httpd.pid 2>/dev/null` 2>
/dev/null || true
> ? ?endscript
> }
>
> Isn't this a bit harsh in 5.4? Wouldn't the 5.3 method be better?
>
> -Christopher Thorjussen

The "kill" command is misnamed and has undoubtedly cause this kind of
confusion since its inception.  What the kill command really does is
send a signal to the process.  By default it just happens to be the
TERM signal.  If you give it another signal, like HUP, it will send
that one instead.

It's up to the individual program to determine what to do when
receiving certain signals.  In the case of HUP, many programs choose
to reload their configurations.  "dd" is interesting in that it prints
out status information when it receives a USR1 signal.

I agree that it should be using "service" instead of kill though.

On Fri, Jan 15, 2010 at 11:47:40AM +0100, Peter Kjellstrom
wrote:
> 5.4 does a "kill -HUP" which is what "service httpd
reload" does too. The -HUP
> signal causes http to re-read stuff rather than shutting down.
> Still, I don't really understand why 5.4 doesn't use "service
reload" which
> seems more "correct" (and the init.d script does a little bit of
checking
> before the kill -HUP too).
That said, *both* of them are too brutal, in that they'll kill any open
connections. Signal USR1 (or service httpd graceful) is much nicer, since it
lets any open connections complete. The downside is that these old
connections might get written to the _rotated_ log instead of the new one,
but to me that's a small price to pay.


-- 
Matthew Miller           mattdm at mattdm.org         
<http://mattdm.org/>