Heh, I didn't think a thread entitled "AWS summits" would be so GCE focused, or otherwise something I should focus on given my employer. Well done proving me wrong. :)
Yeah we've noticed some timing issues around kpartx too; gcimagebundle (one of the tools we install) has started to use the -s option to kpartx and also calls sync after umount. Those make it more fully synchronous.
I think we recently added some sleeps too, but I forget if the flakiness which motivated that occurred on wheezy or not (we do support CentOS / RHEL / SLES which all have older kernels & older util-linux-ng aside from patches/bugfixes).
Olivier/Anders - I agree, merging the code sooner rather than later makes sense.
- Jimmy
Dnia 2014-03-25, wto o godzinie 19:50 +0100, Anders Ingemann pisze:
[ cut ]
> Alright. What gets me confused though, is that I thought kpartx -a was
> blocking rather than asynchronous (which it would be if a sleep call
> is really necessary). Aren't calls to the kernel that do things like
> mapping devices synchronous?
I'm wondering whether it is problem with kernel or with udev
which has some delay when creating device files.
Everything was OK when I was manually repeating steps from
scripts (kpartx, mount, etc.) - but the same commands
were failing when executed from the scripts.
I hope to play with it on additional machines this week - and
we'll see how it goes.
--
Tomasz Rybak GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860
http://member.acm.org/~tomaszrybak