Ok, i found the timeouts problem:

/* we now proceed to read from the client until we get EOF, or until
 * MAX_SECS_TO_LINGER has passed.  the reasons for doing this are
 * documented in a draft:
 *
 * http://www.ics.uci.edu/pub/ietf/http/draft-ietf-http-connection-00.txt
 *
 * in a nutshell -- if we don't make this effort we risk causing
 * TCP RST packets to be sent which can tear down a connection before
 * all the response data has been sent to the client.
 */
#define SECONDS_TO_LINGER 2

so this is not so easy that i thought ;-/

Can you tell more about "are other ways of avoiding the restarts"?

On 13 Sty, 19:41, grassoalvaro <[email protected]> wrote:
> > FWIW, there are other ways of avoiding the restarts by preallocating
> > daemon process groups in advance and not using VirtualHost but other
> > means of virtual hosting.
>
> Hmm, this is interesting, any example (docs)?
>
> On 13 Sty, 12:28, Graham Dumpleton <[email protected]> wrote:
>
> > On 13 January 2011 22:14, grassoalvaro <[email protected]> wrote:
>
> > > Thanks. I'll try to modify Apache source code to change that timeout
> > > and then make some tests.
>
> > Have fun trying to work out where to change it. It seems to differ
> > between MPMs and I never quite got my head around it completely, or at
> > least I have forgotten exactly how it works. Overall It is not
> > something I would recommend and increasing the timeout will likely
> > only cause other problems. That is, rather than have requests be
> > killed, you will cause new requests to stall for a lot longer and
> > sites will simply appear to not be responding.
>
> > > Event if mod_wsgi wan't create for shared hosting - many people use it
> > > for this. And it works well (except Apache reloading ;-)) Solution to
> > > use one Apache for one client is good but takes too many memory.
>
> > FWIW, there are other ways of avoiding the restarts by preallocating
> > daemon process groups in advance and not using VirtualHost but other
> > means of virtual hosting.
>
> > Graham
>
> > > Anywany, thanks for information.
>
> > > On 11 Sty, 23:30, Graham Dumpleton <[email protected]> wrote:
> > >> On 11 January 2011 23:06, grassoalvaro <[email protected]> wrote:
>
> > >> > Thanks for your very useful description.
> > >> > Do you know Is there any good reason why apache is killing process
> > >> > after 3 seconds? Why no 15 for example?
>
> > >> It is a hard wired timeout period it applies to Apache child processes
> > >> when a 'restart' is applied. When it does a graceful restart, it
> > >> handles its own child process differently and lets them run to
> > >> completion, but in that graceful restart mode it still kills off
> > >> processes after fixed 3 seconds if they are what it regards as other
> > >> processes, which is what it regards mod_wsgi daemon processes.
>
> > >> Note that mod_wsgi was originally intended for self hosting of your
> > >> own web applications with daemon mode not actually a part of the
> > >> original goals as far as features. The model of how it is implemented
> > >> isn't necessarily suited to doing shared hosting, especially where
> > >> changes are going to be made often to Apache configuration.
>
> > >> Graham
>
> > >> > On 11 Sty, 10:17, Graham Dumpleton <[email protected]> wrote:
> > >> >> On 11 January 2011 10:27, grassoalvaro <[email protected]> wrote:
>
> > >> >> > Ok, short description is not enought.
>
> > >> >> > My app is an ordering app for shared hosting accounts (for Python
> > >> >> > mod_wsgi) and it's based on mod_wsgi also.
> > >> >> > App is calling some API method (server-app) which is restarting 
> > >> >> > apache
> > >> >> > after httpd modifications.
> > >> >> > The problem is that I can of course use sleep && httpd graceful in
> > >> >> > background, but this will kill other (customer apps) connections 
> > >> >> > after
> > >> >> > sleep.
> > >> >> > So, is there any method do reload config without killing any 
> > >> >> > mod_wsgi
> > >> >> > connections?
>
> > >> >> Connections are transient and not permanent. For a well written site
> > >> >> connections would be maintained for less than a second.
>
> > >> >> The way the process restart works for daemon processes, regardless of
> > >> >> whether you do an Apache restart or graceful restart, is that when
> > >> >> daemon process is signalled to restart, it will stop accepting any
> > >> >> more connections. If there are no active connections for current
> > >> >> requests it will shutdown process immediately and then restart. Any
> > >> >> connections that come during that time effectively queue up with
> > >> >> Apache child processes and/or within port 80 socket listener queue.
> > >> >> When processes restart, the held over requests will be handled.
>
> > >> >> In the case that the daemon process still had active requests, then
> > >> >> rather than daemon process shutting down and restarting immediately it
> > >> >> waits, giving the current requests an opportunity to finish first at
> > >> >> what point the process is then shutdown and restarted. The way Apache
> > >> >> works however, is that it gives a maximum of three seconds grace to
> > >> >> the mod_wsgi daemon processes and if they have not shutdown by then,
> > >> >> because of the requests not finishing, Apache will forcibly kill them
> > >> >> off and they will be restarted.
>
> > >> >> The problem you saw was because your request handler was likely
> > >> >> blocked in some way on waiting for the Apache process to restart and
> > >> >> that very act meant the request didn't complete within the three
> > >> >> seconds allowing the process to shutdown cleanly.
>
> > >> >> Now, as I said, for a well written site, requests should be sub
> > >> >> second, only long requests like large file uploads might be affected.
> > >> >> So, for typical case, the restart wouldn't usually be noticed and
> > >> >> wouldn't affect any connections because outstanding requests would
> > >> >> finish within the three seconds dictated by Apache.
>
> > >> >> Of course, if you are constantly restarting Apache because of
> > >> >> configuration changes, then you increase the risk that you will hit a
> > >> >> requests that doesn't finish within the three seconds and exit in time
> > >> >> to allow process to restart cleanly.
>
> > >> >> Whether you can avoid doing the restarts I cannot comment because have
> > >> >> no idea what changes you are enacting to the Apache configuration
> > >> >> which triggers you wanting to restart it. If the changes relate to
> > >> >> mod_wsgi daemon process provisioning then for some cases you may be
> > >> >> able to avoid it, but it means setting up things in a particular way.
> > >> >> Overall, you are often better off having a nginx front end and simply
> > >> >> giving each customer their own Apache instance behind that. That way
> > >> >> only customer to whom the changes apply are affected and no one else.
> > >> >> This is how WebFaction and some other quality Python hosting companies
> > >> >> work.
>
> > >> >> > Regards.
>
> > >> >> > On 11 Sty, 00:15, Graham Dumpleton <[email protected]> 
> > >> >> > wrote:
> > >> >> >> On 11 January 2011 07:53, grassoalvaro <[email protected]> 
> > >> >> >> wrote:
>
> > >> >> >> > Hi.
>
> > >> >> >> > I have application deployed on apache + mod_wsgi which is 
> > >> >> >> > changing
> > >> >> >> > httpd.conf and then reloading apache (httpd graceful).
> > >> >> >> > And here is the problem: application is making request changing 
> > >> >> >> > httpd
> > >> >> >> > config and reloading him and while request is in progress the 
> > >> >> >> > process
> > >> >> >> > is killing:
>
> > >> >> >> > Premature end of script headers: start.wsgi
> > >> >> >> > (2)No such file or directory: mod_wsgi (pid=31624): Unable to 
> > >> >> >> > connect
> > >> >> >> > to WSGI daemon process 'pylons_app' on 
> > >> >> >> > '/var/logs/wsgi.31409.0.1.sock'
> > >> >> >> > after multiple attempts.,
>
> > >> >> >> > So it is possible to reload _only_ apache config without 
> > >> >> >> > reloading
> > >> >> >> > whole mod_wsgi?
>
> > >> >> >> The short answer is no.
>
> > >> >> >> > Or any other idea how to deal with this problem?
>
> > >> >> >> How are you triggering the Apache restart? The daemon processes
> > >> >> >> themselves will not have sufficient privileges to do it itself, so 
> > >> >> >> you
> > >> >> >> must be using a separate suid script or invoking a remote API for 
> > >> >> >> some
> > >> >> >> service that does it on your behalf.
>
> > >> >> >> If it is a command line script, then invoke it as a background 
> > >> >> >> command
> > >> >> >> with a time delay before it. Ie.
>
> > >> >> >>   os.system('(sleep 1; /some/path/restart-apache)&')
>
> > >> >> >> If it is via a remote API using XML-RPC or something, instead of 
> > >> >> >> doing
> > >> >> >> it from web application, make it a script executed as above.
>
> > >> >> >> A further way is to create a background thread which calls the API.
>
> > >> >> >> In all ways, you are attempting to avoid the problem of the request
> > >> >> >> handler blocking on waiting for the restart command to complete.
>
> > >> >> >> Anyway, will help if you explain how you are triggering the Apache 
> > >> >> >> restart.
>
> > >> >> >> Graham
>
> > >> >> > --
> > >> >> > You received this message because you are subscribed to the Google 
> > >> >> > Groups "modwsgi" group.
> > >> >> > To post to this group, send email to [email protected].
> > >> >> > To unsubscribe from this group, send email to 
> > >> >> > [email protected].
> > >> >> > For more options, visit this group 
> > >> >> > athttp://groups.google.com/group/modwsgi?hl=en.
>
> > >> > --
> > >> > You received this message because you are subscribed to the Google 
> > >> > Groups "modwsgi" group.
> > >> > To post to this group, send email to [email protected].
> > >> > To unsubscribe from this group, send email to 
> > >> > [email protected].
> > >> > For more options, visit this group 
> > >> > athttp://groups.google.com/group/modwsgi?hl=en.
>
> > > --
> > > You received this message because you are subscribed to the Google Groups 
> > > "modwsgi" group.
> > > To post to this group, send email to [email protected].
> > > To unsubscribe from this group, send email to 
> > > [email protected].
> > > For more options, visit this group 
> > > athttp://groups.google.com/group/modwsgi?hl=en.
>
>

-- 
You received this message because you are subscribed to the Google Groups 
"modwsgi" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/modwsgi?hl=en.

Reply via email to