Hi Philip,
I look into setting you up as a comitter!
cheers,
Rob
On 31 Jul 2005, at 20:35, Philip Dodds wrote:
Glad I could help, I wasn't trying to suggest it was a Windows
problem but a difference in the JVM :)
Btw, how you do you apply for commit rights?
P
-----Original Message-----
From: Holger Hoffstätte [mailto:[EMAIL PROTECTED]
Sent: Sunday, July 31, 2005 2:11 PM
To: [email protected]
Subject: Re: [servicemix-dev] Fun with InstallationService
Philip,
thanks for looking into this - I didn't get to it yet. I applied
both of
your patches and PXE now deploys happily as well. :)
This problem looks like it’s a bug in the renameTo method in the JVM,
on Windows if you rename a directory to another directory and they
both
exist then it doesn't do anything. In ServiceMix this means that
the
That's not strictly a Windows JVM bug but rather just the
underlying NTFS
file system semantics - on UNIX rename(2) replaces a directory
(rather its
inode) if it exists, while NTFS does not, so in a way it's a
servicemix
platform dependency. To be safe I'd always copy/delete, that way
you also
gain some kind of transactional behaviour when something goes wrong
along
the way.
If services in the final location are monitored/started
asynchonously (had
this in on of my own systems), copy to servicedir/newservice.tmp,
rename
when finished (atomic operation) and let the monitor ignore everything
that ends with .tmp so that it doesn't try to start a half-copied
service.
The other think you might want to look at it trapping it missing jars
that are being referenced by the component so that invalid
descriptors
don't end up with class not found exceptions or missing bundled jars.
That's a nice idea. I'd also like to see some more DEBUG logging
during
the process so that problems with classpaths, embedded jars etc.
whatever
can be found easier.
Thanks!
Holger
