Ryan Schmidt wrote:
> On 08.03.2005, at 07:24, Ryan Schmidt wrote:
>>> We serve through an Apache 2 server that's also used for many other
>>> things, and it happens probably at least weekly that some
>>> configuration
>>> needs to be changed, which then means Apache has to be restarted. This
>>> is done at times when people could certainly be committing to the
>>> repository. As long as we always do a graceful restart, it seems to me
>>> that it should not be possible for this to cause any problems for
>>> Subversion. Do I see that correctly? What Subversion problems could be
>>> caused by doing a normal non-graceful Apache restart? Would an ongoing
>>> transaction simply be aborted, or do we risk repository corruption?
>>>
>>> We're using Subversion 1.1.2 on Gentoo Linux with a single FSFS
>>> repository. Client access is exclusively through Apache -- no svnserve
>>> or direct file:/// access (although we do have a post-commit hook
>>> which
>>> does various things with svn and svnlook with direct file:/// URLs).
>
>
> On 08.03.2005, at 16:46, Peter Kahn wrote:
>
>> Restarting actually isn't necessary. You can perform a graceful via
>> the
>> apachectl. For all new connections, the new settings will be applied.
>> This allows you to avoid a restart. Please be sure of two things:
>>
>> 1. you are performing the action as the appropriate user
>> if apache or svn_user started apache and now you are acting
>> upon as root, that could lead to permissions issues
>>
>> 2. you apachectl is the right one.
>> If you have multiple installation of apache, then you may perform
>> the graceful upon the wrong one. This can lead to a variety of
>> confusing
>> situations.
>>
>> I recommend performing a which apachectl
>> and then double checking the version of apachectl
>
>
> As I said, I do a graceful restart, and I wanted to confirm that this
> would cause no problems.
>
> I do not believe you are correct with your statements about the root
> user. I believe that (re)starting Apache as the root user should not be
> a problem, since Apache switches itself to the apache user. In fact, I
> believe Apache must be started as root. Anyway we have never had any
> problems with this in the past.
>
> We do not use apache2ctl directly; we use the init.d scripts provided
> with Gentoo Linux. There should be no problem with it using the wrong
> version of the script, since AFAIK there should be only one on the
> server.
>
> As I wrote, I also wanted to know what would happen if an administrator
> other than myself did not use the graceful restart mechanism. Would
> that cause any problems? Would that corrupt the repository? As I said,
> we're using FSFS, not BDB, so I'm hopeful that the worst that could
> happen would be that someone's commit would be interrupted and they'd
> have to try it a second time, but I wanted to ask to make sure. If
> repository corruption is possible by restarting Apache non-gracefully,
> then I must make this extremely clear to our other administrators or
> even block this command.
>
> Thank you for your input.
There is absolutely no problem with doing a cold (non-graceful) restart.
There is, after all, very little difference between a cold and a graceful
restart, except in the relative timing of apache children being stopped and
started.
Max.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Sun Mar 13 01:56:19 2005