Hi,
On Mon, Feb 04, 2008 at 04:16:15PM +0900, [EMAIL PROTECTED] wrote:
Content-Description: Mail message body
> Hi All,
>
> I think that there are some problems in the apache RA.
> My computer environment is as follows
>
> # /usr/sbin/httpd -v
> Server version: Apache/2.2.3
> Server built: Aug 6 2007 07:22:24
> #
> # /usr/lib64/heartbeat/heartbeat -V
> 2.1.3
> #
>
> Please see the problems below:
>
> 1)
> When there is no "PidFile" tag in the "httpd.conf" file, httpd will use
> the value in "/var/log/httpd/httpd.pid".
> It seems default for httpd.
> But apache RA doesn't know about it.
> As a result, RA start becomes the error.
The user should really set the PidFile tag. The default for
various apache releases may be different and it'd probably be an
issue in case of a transition to another
distribution/platform/release.
> 2)
> When there is no "ServerRoot" tag in the "httpd.conf" file, httpd will
> use the value in "/var/${PidFile}".
> It seems to be default for httpd.
> But apache RA doesn't know about it.
> As a result, RA start becomes the error.
Hmm, I thought that ServerRoot was not optional.
> 3)
> When there is no "Listen" tag in the "httpd.conf" file, Listen number
> will be "8443".
> Because "httpd.conf" file is included in "/etc/httpd/conf.d/*.conf".
> and there is "Listen 8443" set in the "/etc/httpd/conf.d/nss.conf".
> If there is no "Listen" tag in the "httpd.conf" file, apache RA will use
> "Listen 8443" in the "/etc/httpd/conf.d/nss.conf".
> But the real Listen number is "443".
> As a result, RA monitor becomes the error.
Looks like our poor man's parser is failing in this case.
> 4)
> In apache, start_apache is using monitor_apache.
> As a result, if the monitor is NG because of the OCF parameters, the
> start will be NG too, although httpd has started normally.
If the monitor op doesn't function properly, the cluster is
unfortunately screwed anyway. The fact that this start may be
reported as success now won't help in the long term.
> I think it is a little strange and confusing.
> And I have checked the other RAs and found that among the all 46 RAs,
> there are 33 RAs does not use monitor in start, but pidfile.
> So I am wondering if we can use the same policy in start_apache.
The argument that monitor is not used in the start operation of
other RAs, even if they are in majority, is not enough. I think
that it is a feature that the start operation reports start
_only_ when apache has really been started. Hmm, hope it's
possible to follow what I just wrote. BTW, that's something which
I built in and that was out of a real need. Namely, it may happen
that a probe (or monitor) is scheduled by CRM very quickly after
the start operation. If that monitor runs, and that may happen,
before the apache is ready to serve the requests, the CRM won't
be extremely happy about it.
> I made a patch file to solve these problems above.
> Please give me some comments.
For 1), 2), and 3) I'd suggest that people configure their
apaches and not rely on the defaults. If they don't and we allow
that, I'm afraid that it's going to be back with a revenge some
day. After all, they should be trying to create an HA solution
which should imply careful consideration and setup of all
components.
As for 4), I'd already given argumentation above.
> If it is OK, can we put it into the next release 2.1.4?
Many thanks for your efforts and comments, but I'm afraid that
you should provide more convincing arguments.
Cheers,
Dejan
> Best Regards,
> Hiroshi Akamatsu.
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
