Re: [studs-user] Truncated WEB-INF/work/ServletContext.ser

[Dan A. <dan...@gm...>]

On 7/28/05, Johan Mj=F6nes <joh...@on...> wrote:
> Hi all!
>=20
> I've noticed that WEB-INF/work/ServletContext.ser file sometimes gets
> truncated to zero bytes, but still exists.
>=20
> This results in the following error:
>=20
> : Fatal error: Call to a member function getStartup() on a non-object in =
C:\develsoft\Apache Group\Apache2\htdocs\basic\WEB-INF\lib\stratus\connecto=
r\HttpProcessor.php on line 80
>         at .RootException(C:\develsoft\Apache Group\Apache2\htdocs\basic\=
WEB-INF\lib\horizon\lang\RootException.php:67)
>         at .fatalErrorWatchdog(C:\develsoft\Apache Group\Apache2\htdocs\b=
asic\WEB-INF\lib\stratus\connector\HttpProcessor.php:328)
>         at [PHP].main(:)
>=20
> context is now of type bool (and is false since an eval() of a zero lengt=
h string will return false). ContextConfig does some minor validity checkin=
g (if context is null), but it should really verify that the object was pro=
perly loaded. Also, if object is not properly loaded, I don't throwing an e=
xception is exciting is really what you want. Log it and then try to recrea=
te the Context (by reloading the configuration).
>=20
> What do you think? :)
>=20
> However, this isn't the problem. The problem is WHY this file gets trunca=
ted? I haven't done any major investigation so far, I wanted to see if you =
had any suggestions or ideas first.
>=20
> Best Regards,
>=20
> Johan
>=20

Okay, first a few questions.  What platform are you running (I know it
is some variant of windows) and what is the file system.  Clearly this
functionality is working fine on Linux because the demo application
has never become corrupted and it is hit in parallel quite often.  It
feels to me like a file locking issue.  In general, the servlet
context is written by:

open file in append mode
lock
truncate
unlock
close

I have added a couple of santiy checks in the code.  First, if
ServletContext.ser is empty, I treat it as non-existing.  If the lock
fails, then I don't write the servlet context.  Finally, I do a check
for a valid StandardContext object rather than null.  However, if the
StandardContext is not valid, I do throw an exception rather than
loading a new version.  I figured if the ServletContext.ser file has
invalid data, we should not treat that as okay, but rather throw an
exception since we need to depend on that file as being valid for
smoooth operation of the application.

If the problem is with locking, what I suggest is that you run a
little test script (actually two scripts) that test basic file locking
on your platform and see if that works.  If it doesn't, perhaps we
need to use a more generic form of file locking in Studs.

/dan
--=20
Open Source Advocacy
http://www.mojavelinux.com