On 7/31/07, Alan Robertson <[EMAIL PROTECTED]> wrote:
> Andrew Beekhof wrote:
> > On 7/30/07, Alan Robertson <[EMAIL PROTECTED]> wrote:
> >> Andrew Beekhof wrote:
> >>> On 7/24/07, Andrew Beekhof <[EMAIL PROTECTED]> wrote:
> >>>> On 7/24/07, Alan Robertson <[EMAIL PROTECTED]> wrote:
> >>>>> The Linux-HA team proudly announces feature and bug-fix release 2.1.1
> >>>>> of the Linux-HA (aka "Heartbeat", aka "OpenHA") software.
> >>>> Alan, can you please explain the following change-set which you made
> >>>> just before the release:
> >>>> http://hg.linux-ha.org/dev/rev/e46a432e3c50
> >>>>
> >>>> I am struggling to think of a legitimate reason for most of these
> >>>> messages to be changed from ERRORs to WARNINGs... even assuming they
> >>>> are related to Bug #1657
> >>>>
> >>>> In particular, I find the following change particularly concerning:
> >>>> if self.rsh(node, self["StartCmd"]) != 0:
> >>>> - self.log ("ERROR: Start command failed on node %s" %(node))
> >>>> + self.log ("Warn: Start command failed on node %s" %(node))
> >>> Still looking for some sort of response on this...
> >>>
> >>> With one exception, I would be most concerned if these changes were
> >>> required to allow testing to pass.
> >> Hadn't seen your reply yet. Sorry. I don't have time to read the main
> >> list at the moment. I think I backed those changes out.
> >
> > The changes are still present in /dev
> >
> >> I don't
> >> understand why expecting a message which always comes out if it succeeds
> >> to be there is a problem - as this is how it originally worked. Each
> >> time a message goes missing, it causes up to a 5 minute delay in the
> >> tests - so it's good to track down if we are still having troubles.
> >
> > This is not my concern.
> >
> > My concern, as Manfred correctly pointed out, is that you changed a
> > number of ERRORs to WARNings which would allow tests to appear to
> > succeed when in fact they failed.
> >
> > This was highlighted in the change I quoted:
> > if self.rsh(node, self["StartCmd"]) != 0:
> > - self.log ("ERROR: Start command failed on node %s" %(node))
> > + self.log ("Warn: Start command failed on node %s" %(node))
> >
> > StartCmd does not depend on any log messages and if it returns
> > something other than 0 we most certainly have a big problem.
>
> It was an ERROR only because I had changed it to an error. I'm pretty
> sure that this change was putting it back the way it was. This is the
> very thing that is needed for BSC to pass - that I'm fixing in another
> bug that you've commented on.
>
> When I wrote this code many years ago, this was originally an error.
Actually, that is incorrect.
This is the original message you had (dating back to revision 1018):
self.log ("Could not start %s on node %s"
% (self["Name"], node))
> I
> didn't research the change history for this, but knowing that this
> little kludge is necessary for BSC to pass, my suspicion is that it was
> you who made this a warning in order for BSC to pass on the CRM tests.
While I did make the message a warning, that actually _increased_ the
"severity".
> Sorry I had misunderstood the original request.
>
> It was me who had changed the warnings to ERRORS before release, then I
> reverted this change just before releasing, because it was necessary for
> BasicSanityCheck to pass
This was my concern.
You found some things that should have been ERRORs and changed them...
which is great. We advocate syslog-ng these days, which doesn't have
syslog's habit for losing messages, so it seems obvious that they
really should be treated more seriously than in the past.
But to then observe failures and change them back... right before a
release... without an explanation or a bug reference... and the bug
that was created later doesn't appear to be relevant to at least some
of the changes...
That I find troubling.
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
