Re: (no subject)

On Wed, 29 Jan 2003 [email protected] wrote:

> Previously in isconf 2i when we wanted to synchronize
> a set of files from the gold server to the client we
> would (in the makefile) "cd /" and then rsync one or
> more  representative root trees from the gold server
> (common tree, os specific tree, etc..)
>
> Although the representative root trees may indeed be
> controlled in CVS, if a client is rebuilt, it has
> the possibility of functioning different because
> the fileset it will rsync when repeating steps
> may reflect contents that was historically not
> there at the initial build and which in some way
> conflicts with the early stanzas in the makefile.

Yep.

> If I understand correctly, filesets was introduced
> to move this 'cd /; rsync from goldhost' functionality
> into the makefile state-engine paradigm so that
> a strict sequencing of when a set of files is introduced
> onto the client.

Not exactly...

> What confuses me is that once filesets are added
> to a client, the client then appears to rsync the
> filesets everytime isconf is run.
>
> If these filesets are updated in any manner on
> the goldhost, it seems, at least to some extent,
> we are back to the original dilema.

That is exactly the case.  Repeat after me: if you are pulling files down
from a central server, then you are breaking deterministic ordering.  If
you are using cron jobs, you are breaking deterministic ordering.  If you
allow a host to be multiple host types, you are potentially breaking
deterministic ordering.  If you add a stanza which requires a reboot, but
allow other stanzas to execute while waiting for available downtime to
reboot, you are breaking deterministic ordering.

Basically, I consider it impossible to maintain a dynamic environment in a
completely deterministically ordered manner.  I still plan on elaborating
all of this into a paper, but I haven't yet.

Currently, if you use filesets for any reason other than initialization of
the files on the local system with no later updates, then yes, you are
breaking deterministic ordering.  I haven't yet found this to be a
problem, and I don't really expect to, but it does break the rules.

Caveat emptor.

Luke

-- 
I hate to advocate drugs, alcohol, violence, or insanity to anyone, but
they've always worked for me.                     -- Hunter S. Thompson


-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com