On Tue, Jul 14, 2009 at 09:01:55PM +1000, Mark Goodwin wrote:
> Martin Hicks wrote:
>>> How about this patch. It rejigs things a lot, but essentially:
>>>
>>> - No config file will be created unless the '-w' flag is passed to
>>> pmdaib. (this is a change)
>>>
>>> - By default, if no config file is found, only local ports are monitored.
>>> (before we would always write out a config file containing only local
>>> ports)
>>>
>>> - If a config file exists then it is the preferred source of information
>>> and perhaps not even local ports will be monitored. (before you could
>>> have edited the automatically created file to remove the local ports)
>>>
>>>
>>> So in the case of a user who is using a custom config file (either to
>>> name the instances differently, or to monitor remote ports) who replaces
>>> an HCA, it will be up to them to edit the config file that they created.
>>>
>>> For those who are using the ib pmda in its simplest mode with no config
>>> file, no changes will be necessary when an HCA is replaced but they have
>>> no guarantee of static device naming if they add an additional HCA.
>
> Seems like a reasonable compromise to me. Instance naming in merged
> archives spanning reboots will sometimes be screwed up, but you still
> have the option to avoid that if it really matters. When you look at the
> flaming hoops the scsi and device-mapper folks have jumped thru to
> achieve persistent naming, a more sophisticated scheme isn't warranted;
> set-and-forget performance monitoring is more important in most cases.
I think I'm going to change things so the '-w' writes the config file
and exits immediately.
I'll also write up a manpage. This PMDA is fairly complicated and I had
a few "oh, it does that too?" moments while I was doing this work.
mh
|