pcp
[Top] [All Lists]

Re: [pcp] PCP Updates: qa fallout from ipv6/unix sockets for pmlogger an

To: Dave Brolley <brolley@xxxxxxxxxx>
Subject: Re: [pcp] PCP Updates: qa fallout from ipv6/unix sockets for pmlogger and pmlc
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Thu, 20 Feb 2014 18:24:27 -0500 (EST)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <530612EC.8020206@xxxxxxxxxx>
References: <52FE5058.4030702@xxxxxxxxxx> <53023D4E.1060504@xxxxxxxxxx> <y0mmwhoqu69.fsf@xxxxxxxx> <757832688.10280462.1392753861578.JavaMail.zimbra@xxxxxxxxxx> <896174788.10421447.1392770006295.JavaMail.zimbra@xxxxxxxxxx> <5304D039.9010708@xxxxxxxxxx> <1347098955.12246278.1392874951684.JavaMail.zimbra@xxxxxxxxxx> <530612EC.8020206@xxxxxxxxxx>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: AD8K2hhiKC0yPzGDdoHqT7kCzCuTgg==
Thread-topic: PCP Updates: qa fallout from ipv6/unix sockets for pmlogger and pmlc
Hi Dave,

----- Original Message -----
> On 02/20/2014 12:42 AM, Nathan Scott wrote:
> >
> > ----- Original Message -----
> >> I think that ~/.pcp/pmlogger would be a good place.
> > The problem (which I failed to get across :) is that local pmloggers
> > are already writing to that location.  We would end up mixing up log
> > folios, per-host subdirectories, and so on with these new PID files.
> > Which would likely end in occassional wierd namespace conflicts, and
> > generally make management of these files more difficult.
> >
> > Hence, the suggestion of ~/.pcp/tmp/pmlogger mirroring (a little, wrt
> > the suffix) the layout we're using for the system loggers port map.
> The actual full name of the socket would be
> ~/.pcp/pmlogger/pmlogger.<pid>.socket. I don't see how this could end up
> with a namespace conflict. Could you perhaps give an example? There is
> some redundancy there. It could be shortened to
> ~/.pcp/pmlogger/<pid>.socket. Have said that, I'm not attached to my
> choice, and it would be easy to add the extra tmp component.

As long as the name (the basename(1) I mean) is the same as the system
variant, its fine by me - although as you mention, yeah, the pmlogger.
prefix is redundant under all schemes so I guess I'd prefer it gone.

Wrt "namespace conflict", yeah, no specific scenario - just a general
vibe that separating permanent files (archives, folios and so on) from
temporary rundir type files is a Good Thing generally.  One example off
the top of my head would be usability in pmchart.  When someone browses
their local archives in the GUI, looking at ~/.pcp/pmchart/* it'd be a
shame for them to have to sort through socket files in there too which
would have nothing to do with the task at hand.  *shrug*... just seems
to make sense to keep 'em apart.

I suppose if we managed to lose track of these socket files somehow, &
ended up leaving some lying around after pmlogger exited, its easier to
cleanup when its a separate directory too.

> Overnight testing has revealed no qa issues.

Mhmmm, is test 750 passing for you atm?  :)  I accidentally included my
hostname in its output, so it should be failing everywhere else.  Sorry
'bout that - will be fixed shortly ... but do you have any other common
failures though?  (that fail without these new changes, I mean).

cheers.

--
Nathan

<Prev in Thread] Current Thread [Next in Thread>