Hi, On Wed, Nov 11, 2009 at 10:45:13AM +0100, Lars Marowsky-Bree wrote: > On 2009-11-11T10:14:29, Florian Haas <[email protected]> wrote: > > > > This is a common mistake, we remove them when we find them, but please > > > don't add new ones ;-) > > > > Maybe I'm missing something, but I don't follow that. If that is indeed > > a mistake, then please replace "common" with "ubiquitous" as I believe > > at least checking for binaries and erroring out if they are not present > > is something that almost all resource agents do. > > > > Examples (by no means exhaustive): > > > > - mysql > > - pgsql > > - Route > > - VirtualDomain > > - Filesystem > > > > Some of these check for binaries inside validate, some in a part of the > > script that is unconditionally executed every time it's invoked > > (including during validate). > > > > So please, either revisit that policy, or fix all those RAs. > > The point being that it doesn't even need to be a policy; it is simply > an observation. > > "Filesystem" is quite probably fine, because it just relies on standard > system binaries, and none of the binary paths are configurable. (Same > applies to Route/VirtualDomain.) Further, it is quite unlikely that > Filesystem needs to have a special cluster fs mounted, since one would > use Filesystem to do that ;-) > > For mysql, pgsql, apache or others with configurable paths, it is > possible (and sometimes implemented as such) that they reference a > binary/file on cluster-managed storage. That won't be yet mounted on a > normal start-up, and thus, if the probe tries to verify the path, fail. > > Simply put: if you're checking for dependencies possibly provided by > other resources, these won't be present at probe time. (Kind of obvious, > really.)
In this case, the CRM, which knows that dependencies are not started, should ignore the INSTALLED error. But that's probably an extra burden on an already complex piece of software. Or we provide some facility to all resource agents which would deal with the binaries on a missing filesystem situation. I mean, this is a issue common to all resource agents. > So yes, all RAs which unconditionally check for files which could > realistically be on cluster managed storage need to be fixed. > > (pgsql would benefit immensely from using check_binary directly, too; > but that script further believes that the binary not present means that > the service is stopped. Ah, well.) Recently, I opened a bugzilla for this issue: http://developerbugs.linux-foundation.org/show_bug.cgi?id=2204 which is still open. Thanks to Lars for the exhaustive account on the matter (which one can read in the bugzilla too). Dejan > > Regards, > Lars > > -- > Architect Storage/HA, OPS Engineering, Novell, Inc. > SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) > "Experience is the name everyone gives to their mistakes." -- Oscar Wilde > > _______________________________________________________ > Linux-HA-Dev: [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev > Home Page: http://linux-ha.org/ _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
