cvs bytestats

cvs bytestats

[Christopher Faylor]

Do we *really* need to track all of the I/O activity for CVS?
/var/log/syslog has thousands of messages like this:

sourceware cvs: error setting KEEPALIVE: Socket operation on non-socket

which is due to the fact that anoncvs is being run in a pipe so that we
can track the amount of I/O performed by cvs.

I turned this off once before but I think Ian and Jason thought the data
was useful.  Is this the kind of data that is useful when there is a
problem?  Because I doubt that anyone is looking at this now.

I noticed this while setting up cvs on the new system so I thought I'd
revisit the decision.

cgf

Re: cvs bytestats

[Ian Lance Taylor]

Christopher Faylor <cgf@redhat.com> writes:

> Do we *really* need to track all of the I/O activity for CVS?
> /var/log/syslog has thousands of messages like this:
> 
> sourceware cvs: error setting KEEPALIVE: Socket operation on non-socket
> 
> which is due to the fact that anoncvs is being run in a pipe so that we
> can track the amount of I/O performed by cvs.
> 
> I turned this off once before but I think Ian and Jason thought the data
> was useful.  Is this the kind of data that is useful when there is a
> problem?  Because I doubt that anyone is looking at this now.
> 
> I noticed this while setting up cvs on the new system so I thought I'd
> revisit the decision.

I think the CVS byte data is useful.  I don't think it is useful to
have a syslog message from CVS--I didn't even realize that CVS would
use syslog for something like that.  There is probably some
syslog.conf adjustment you can make to stop those messages, or just
change CVS.

But if you want to get rid of the CVS byte data, I won't fight to keep
it.

Ian

Re: cvs bytestats

[Jason Molenda]

On Sat, Jan 11, 2003 at 05:02:29PM -0500, Christopher Faylor wrote:

> 
> I turned this off once before but I think Ian and Jason thought the data
> was useful.  Is this the kind of data that is useful when there is a
> problem?  Because I doubt that anyone is looking at this now.
> 

I think it's useful for establishing trends and baselines.  When
the system starts to melt down - bandwidth is congested, disks fill
up, processor maxes out - it's easy to look over the trends and
see what is "normal" and how "normal" has changed over time.  

When you're looking for an acute problem, an anoncvs abuser, it's
nice to have it running, but that's not an argument for necessarily
leaving it running continuously.

Those are the reasons I like having it, but I won't argue strenuously
against disabling it if it causes some problem.  I'd be willing to
disable the message logging in cvs, as per Ian's suggestion, and
install an updated binary.

Jason

Re: cvs bytestats

[Christopher Faylor]

On Sat, Jan 11, 2003 at 03:52:50PM -0800, Jason Molenda wrote:
>On Sat, Jan 11, 2003 at 05:02:29PM -0500, Christopher Faylor wrote:
>
>> 
>> I turned this off once before but I think Ian and Jason thought the data
>> was useful.  Is this the kind of data that is useful when there is a
>> problem?  Because I doubt that anyone is looking at this now.
>> 
>
>I think it's useful for establishing trends and baselines.  When
>the system starts to melt down - bandwidth is congested, disks fill
>up, processor maxes out - it's easy to look over the trends and
>see what is "normal" and how "normal" has changed over time.  
>
>When you're looking for an acute problem, an anoncvs abuser, it's
>nice to have it running, but that's not an argument for necessarily
>leaving it running continuously.
>
>Those are the reasons I like having it, but I won't argue strenuously
>against disabling it if it causes some problem.  I'd be willing to
>disable the message logging in cvs, as per Ian's suggestion, and
>install an updated binary.

It's trivial to disable the message but I wonder if using bytestats
in this way is also causing other problems.  We have a lot of hung
cvs connections and I was wondering if this is a symptom of that.
I can't see why it would be since bytestats also sets SO_KEEPALIVE
unless bytestats isn't noticing the connection going away.  Ian,
how would I go about regenerating bytestats?

Anyway, I regenerated cvs and got rid of the syslog message.

cgf

Re: cvs bytestats

[Ian Lance Taylor]

Christopher Faylor <cgf@redhat.com> writes:

> It's trivial to disable the message but I wonder if using bytestats
> in this way is also causing other problems.  We have a lot of hung
> cvs connections and I was wondering if this is a symptom of that.
> I can't see why it would be since bytestats also sets SO_KEEPALIVE
> unless bytestats isn't noticing the connection going away.  Ian,
> how would I go about regenerating bytestats?

As you probably know, the source is in
/sourceware/cvspserver/bytestats.c.  The header files and libraries
which it uses are in /usr/local/src/qmail-support/ucspi-tcp-0.88.

I guess I should have written a Makefile entry to build it.  I can do
that if you like.

Ian

Re: cvs bytestats

[Christopher Faylor]

On Sun, Jan 12, 2003 at 09:30:57AM -0800, Ian Lance Taylor wrote:
>Christopher Faylor <cgf@redhat.com> writes:
>
>> It's trivial to disable the message but I wonder if using bytestats
>> in this way is also causing other problems.  We have a lot of hung
>> cvs connections and I was wondering if this is a symptom of that.
>> I can't see why it would be since bytestats also sets SO_KEEPALIVE
>> unless bytestats isn't noticing the connection going away.  Ian,
>> how would I go about regenerating bytestats?
>
>As you probably know, the source is in
>/sourceware/cvspserver/bytestats.c.  The header files and libraries
>which it uses are in /usr/local/src/qmail-support/ucspi-tcp-0.88.
>
>I guess I should have written a Makefile entry to build it.  I can do
>that if you like.

No need.  I don't think we'll be rebuilding bytestats too often.  I
just wanted to add some debugging statements to it.

cgf

Re: cvs bytestats

[David Edelsohn]

>>>>> Christopher Faylor writes:

Chris> Anyway, I regenerated cvs and got rid of the syslog message.

	Did you make any other changes when you regenerated cvs?  Was the
cvs stat optimization removed or corrupted?  "cvs update" of the gcc
sources has gotten *much* slower for me in the past week -- around the
time of the bytestat change and the archive regeneration.

David