echo

echo

[Frank Ch. Eigler]

echo

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 01:16:12PM -0400, Frank Ch. Eigler wrote:
> echo

There exists a test-list@sourceware.  There is a web archive.  Please
don't send this sort of thing to actual lists.

Re: echo

[Jim Kingdon]

> There exists a test-list@sourceware.  There is a web archive.  Please
> don't send this sort of thing to actual lists.

...ACTUAL LISTS....Actual Lists...actual lists....

Actually Frank's message was quite clever; here's the relevant part for
the subtlety-challenged:

    To: overseers@sourceware.cygnus.com, overseers@sources.redhat.com
    Cc: overseers@sources.redhat.com, overseers@sourceware.cygnus.com

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 01:47:30PM -0400, Jim Kingdon wrote:

> 
>     To: overseers@sourceware.cygnus.com, overseers@sources.redhat.com
>     Cc: overseers@sources.redhat.com, overseers@sourceware.cygnus.com


OK OK, it's early in the morning for me, it's not my fault.  I saw
all those addrs and figured they must be having another mail routing
problem @ RH and it was an oddly formatted test message.

J

PS-  As a bonus piece of information to make up for it, I'll note
that I was on-line when Cagney sent in the note about cvs-tmp.  I
checked a few minutes after he reported the problem and there were
a couple hundred megs free already.

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 10:52:53AM -0700, Jason Molenda wrote:

> PS-  As a bonus piece of information to make up for it, I'll note
> that I was on-line when Cagney sent in the note about cvs-tmp.  I
> checked a few minutes after he reported the problem and there were
> a couple hundred megs free already.


BTW I learned a trick to confuse cvs and fill up cvs temp directories
a while back:  Touch all the files in your check-out, then do a
cvs update.

cvs records the timestamp of files as it checks them out, and if
the timestamp on the file is newer than the one in the CVS/Entries
file, cvs thinks the file might have been changed.  cvs stupidly
uploads all of the files in your checkout all at once, so with a
sufficiently large repository, you can fill up the temp space pretty
easily.  (as opposed to a normal update, where the temp directories
are used one-directory-at-a-time and then cleaned up after they
have been used)

The src or gcc repositories are probably large enough to trigger
this problem on sourceware.  The cvs-tmp directory was not set up
to handle the biggest possible checkout that can be done - it was
set up to handle the normal one-directory-at-a-time usage pattern.
We had a similar problem here at Yahoo, where I had the temp
directory running out of a RAM disk (huge speedup BTW).  We have
a large enough repository that someone did a nice DoS on the server
by the touch/update scheme I outlined above - I had to drop the
RAM disk approach because of it.

J

Re: echo

[Jim Kingdon]

> cvs stupidly uploads all of the files in your checkout all at once

This kind of falls out of other aspects of the CVS design, which there
are good reasons for.  But I don't think overseers is the place to
discuss the design tradeoffs (since it is far from a simple change to
cvs even if there was agreement about changing it).

Of course, as sysadmins we do need to be aware of what CVS does....

> I had the temp directory running out of a RAM disk (huge speedup BTW).

I bet the speedup is bigger on *BSD than Linux, since Linux would be
more aggressive about using that RAM for buffer cache of inodes and
such.

Re: echo

[Jim Kingdon]

> cvs stupidly uploads all of the files in your checkout all at once

This kind of falls out of other aspects of the CVS design, which there
are good reasons for.  But I don't think overseers is the place to
discuss the design tradeoffs (since it is far from a simple change to
cvs even if there was agreement about changing it).

Of course, as sysadmins we do need to be aware of what CVS does....

> I had the temp directory running out of a RAM disk (huge speedup BTW).

I bet the speedup is bigger on *BSD than Linux, since Linux would be
more aggressive about using that RAM for buffer cache of inodes and
such.

echo

[Frank Ch. Eigler]

echo

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 01:47:30PM -0400, Jim Kingdon wrote:

> 
>     To: overseers@sourceware.cygnus.com, overseers@sources.redhat.com
>     Cc: overseers@sources.redhat.com, overseers@sourceware.cygnus.com


OK OK, it's early in the morning for me, it's not my fault.  I saw
all those addrs and figured they must be having another mail routing
problem @ RH and it was an oddly formatted test message.

J

PS-  As a bonus piece of information to make up for it, I'll note
that I was on-line when Cagney sent in the note about cvs-tmp.  I
checked a few minutes after he reported the problem and there were
a couple hundred megs free already.

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 10:52:53AM -0700, Jason Molenda wrote:

> PS-  As a bonus piece of information to make up for it, I'll note
> that I was on-line when Cagney sent in the note about cvs-tmp.  I
> checked a few minutes after he reported the problem and there were
> a couple hundred megs free already.


BTW I learned a trick to confuse cvs and fill up cvs temp directories
a while back:  Touch all the files in your check-out, then do a
cvs update.

cvs records the timestamp of files as it checks them out, and if
the timestamp on the file is newer than the one in the CVS/Entries
file, cvs thinks the file might have been changed.  cvs stupidly
uploads all of the files in your checkout all at once, so with a
sufficiently large repository, you can fill up the temp space pretty
easily.  (as opposed to a normal update, where the temp directories
are used one-directory-at-a-time and then cleaned up after they
have been used)

The src or gcc repositories are probably large enough to trigger
this problem on sourceware.  The cvs-tmp directory was not set up
to handle the biggest possible checkout that can be done - it was
set up to handle the normal one-directory-at-a-time usage pattern.
We had a similar problem here at Yahoo, where I had the temp
directory running out of a RAM disk (huge speedup BTW).  We have
a large enough repository that someone did a nice DoS on the server
by the touch/update scheme I outlined above - I had to drop the
RAM disk approach because of it.

J

Re: echo

[Jim Kingdon]

> There exists a test-list@sourceware.  There is a web archive.  Please
> don't send this sort of thing to actual lists.

...ACTUAL LISTS....Actual Lists...actual lists....

Actually Frank's message was quite clever; here's the relevant part for
the subtlety-challenged:

    To: overseers@sourceware.cygnus.com, overseers@sources.redhat.com
    Cc: overseers@sources.redhat.com, overseers@sourceware.cygnus.com

Re: echo

[Jason Molenda]

On Fri, Sep 08, 2000 at 01:16:12PM -0400, Frank Ch. Eigler wrote:
> echo

There exists a test-list@sourceware.  There is a web archive.  Please
don't send this sort of thing to actual lists.