Mailing List Archive: drbd / 2.4.. getting better...

drbd / 2.4.. getting better...

Aug 30, 2001, 5:59 PM

Post #1 of 3 (566 views)

I have a 17gb drive mirrored, it got through syncing the whole thing,
etc... (at 4MB/s)
could copy a few files onto it, etc... (biggest about 133MB)

when coppying a 400mb file i started to get the following messages flying
down the screen (on primary)

drbd0:Epoch set size wrong, found 6 expected 322
drbd0:transfer log too small

(im going from memory here... the message might have been worded slighly
different...)
kept going for maybee a minute or so while copying the file in, then
paniced...

i then was able to run "drbsetup /dev/nb0 primary" on the second machine..
and mount the device...
all the files seemed ok.. (except the one it crashed while copying of
course...)

Paul

Re: drbd / 2.4.. getting better... [ In reply to ]

philipp.reisner at example

Aug 31, 2001, 1:07 AM

Post #2 of 3 (558 views)

Permalink

This is not a real bug, it is just a bit un-tuned :)
Simply increase the size of the transfer log by using something like
this:

...
net {
sync-rate=4M
tl-size=512
^^^^^
}
...

On my user mode linux boxes I have to set it to 1024, but I think
this is due to the high latency of UML's block device driver.

256 was big enough in 2.2.x times, but 2.4.x has bigger request queue
sizes... (and the size of these is determined at boot time. I think
it is derived from memory size)

PS: one tl-entry takes 8 bytes of main memory..

-Philipp

* Paul Botelho <paul@example.com> [010831 02:59]:
> I have a 17gb drive mirrored, it got through syncing the whole thing,
> etc... (at 4MB/s)
> could copy a few files onto it, etc... (biggest about 133MB)
>
> when coppying a 400mb file i started to get the following messages flying
> down the screen (on primary)
>
> drbd0:Epoch set size wrong, found 6 expected 322
> drbd0:transfer log too small
>
> (im going from memory here... the message might have been worded slighly
> different...)
> kept going for maybee a minute or so while copying the file in, then
> paniced...
>
> i then was able to run "drbsetup /dev/nb0 primary" on the second machine..
> and mount the device...
> all the files seemed ok.. (except the one it crashed while copying of
> course...)
>
> Paul
>
>
> _______________________________________________
> DRBD-devel mailing list
> DRBD-devel@example.com
> http://lists.sourceforge.net/lists/listinfo/drbd-devel

Re: drbd / 2.4.. getting better... [ In reply to ]

paul at example

Aug 31, 2001, 9:59 AM

Post #3 of 3 (562 views)

Permalink

thanks.... its working great now... (actually realized that as i was
typeing out the error message "transfer log too small" duh... increase
xfer log size...

anyone know what the fastest sync-rate you can get away with on a 100baseT
loopback connection between the servers?

-Paul.

On 2001.08.31 01:07 Philipp Reisner wrote:
> This is not a real bug, it is just a bit un-tuned :)
> Simply increase the size of the transfer log by using something like
> this:
>
> ...
> net {
> sync-rate=4M
> tl-size=512
> ^^^^^
> }
> ...
>
> On my user mode linux boxes I have to set it to 1024, but I think
> this is due to the high latency of UML's block device driver.
>
> 256 was big enough in 2.2.x times, but 2.4.x has bigger request queue
> sizes... (and the size of these is determined at boot time. I think
> it is derived from memory size)
>
> PS: one tl-entry takes 8 bytes of main memory..
>
> -Philipp
>
> * Paul Botelho <paul@example.com> [010831 02:59]:
> > I have a 17gb drive mirrored, it got through syncing the whole thing,
> > etc... (at 4MB/s)
> > could copy a few files onto it, etc... (biggest about 133MB)
> >
> > when coppying a 400mb file i started to get the following messages
> flying
> > down the screen (on primary)
> >
> > drbd0:Epoch set size wrong, found 6 expected 322
> > drbd0:transfer log too small
> >
> > (im going from memory here... the message might have been worded
> slighly
> > different...)
> > kept going for maybee a minute or so while copying the file in, then
> > paniced...
> >
> > i then was able to run "drbsetup /dev/nb0 primary" on the second
> machine..
> > and mount the device...
> > all the files seemed ok.. (except the one it crashed while copying of
> > course...)
> >
> > Paul
> >
> >
> > _______________________________________________
> > DRBD-devel mailing list
> > DRBD-devel@example.com
> > http://lists.sourceforge.net/lists/listinfo/drbd-devel
>
>