On Tue, Feb 26, 2019 at 09:34:21 +0100, Diego Michelotto wrote:
HI Peter
Il giorno 26 feb 2019, alle ore 09:08, Peter Krempa <pkrempa@redhat.com> ha scritto:
On Mon, Feb 25, 2019 at 19:19:03 +0100, Diego Michelotto wrote:
From: dmichelotto <diego.michelotto@cnaf.infn.it <mailto:diego.michelotto@cnaf.infn.it>>
With your permission I'll remove the above line when applying the patch so that your authorship of the patch shows up as:
Author: Diego Michelotto <diego.michelotto@cnaf.infn.it <mailto:diego.michelotto@cnaf.infn.it>>
instead of
Author: dmichelotto <diego.michelotto@cnaf.infn.it <mailto:diego.michelotto@cnaf.infn.it>>
Sure, you can change it.
Added GPFS as shared file system recognized during live migration security checks.
GPFS is 'IBM General Parallel File System' also called 'IBM Spectrum Scale'
BUG: https://bugzilla.redhat.com/show_bug.cgi?id=1679528
Signed-off-by: Diego Michelotto <diego.michelotto@cnaf.infn.it> --- src/util/virfile.c | 9 ++++++++- src/util/virfile.h | 1 + tests/virfiledata/mounts3.txt | 1 + tests/virfilemock.c | 5 +++++ tests/virfiletest.c | 1 + 5 files changed, 16 insertions(+), 1 deletion(-)
One thing which might be worth exploring is whether GPFS is also migratable with cache != none and thus should be also added to virStorageFileIsClusterFS.
GPFS stalls badly when a process issues a lot of fsyncs in a short timeframe. It is better if the VM I/O is not synchronous, with previous version of libvirt we are using writeback caching.
If you prefer I’ll try to make a new patch, but let me know if it’s easier for you to add yourself GPFS to the virStorageFileIsClusterFS function.
It will be better to send this as a separate patch. The safety of migration with caching is a complex problem which is harder to prove that it's correct. That one will require a justification of it's own. I'll push this meanwhile.