I was not aware of the pause command. I will try using that command
instead of savevm-start/end.
The reason for a possible requirement to freeze the vm is cause by the
fact that you could have continues writes to the disk which is migrated
in which case, since writes locally obviously must be faster than writes
remote, the temporary backend will never be completely flushed. I have
tested it doing the following while migrating:
ping some_ip > /disk/under/migrate/ping.dump
if the host is not freezed for a couple of seconds the migration will
never finish since temporary storage will never be empty. Consider the
temporary storage like a queue. Then the algorithm is as follows:
while queue not empty:
write old storage;
stream writes to new storage;
send commit ok to file system
flush cache where remote storage resides;
shift storage pointer from old to new;
remove queue;
On 01-17-2013 09:24, Alexandre DERUMIER wrote:
I'm looking to your drive-mirror code,
about
# if writes to disk occurs the disk needs to be freezed
# to be able to complete the migration
Are you sure that drive-mirror cannot handle new writes during the
mirroring ?
I think they must be a bitmap a new writes somewhere, maybe it's
exist some kind of delta write command ?
also, If you really need to freeze the vm, I think you can simply
pause the vm, you don't need to use savevm-start/end.(because this is
used to save vm memory)
my $freeze = sub {
print "Freezing to complete migration...\n";
PVE::QemuServer::vm_mon_cmd($vmid, "savevm-start");
};
my $activate = sub {
print "Reactivate after freeze...\n";
PVE::QemuServer::vm_mon_cmd($vmid, "savevm-end");
};
eval {
while (1) {
my $stats = PVE::QemuServer::vm_mon_cmd($vmid,
"query-block-jobs");
my $stat = @$stats[0];
die "error job is not mirroring" if $stat->{type} ne
"mirror";
print $stat->{offset}."/".$stat->{len}."\n";
last if ($stat->{len} == $stat->{offset});
if ($old_len == $stat->{offset}) {
if ($count > 5) {
# if writes to disk occurs the disk
needs to be freezed
# to be able to complete the migration
&$freeze;
$count = 0;
$frozen = 1;
} else {
$count++ unless $frozen;
}
}
elsif ($frozen) {
&$activate;
$count = 0;
}
$old_len = $stat->{offset};
sleep 1;
}
};
----- Mail original -----
De: "Michael Rasmussen" <m...@datanom.net>
À: "Alexandre DERUMIER" <aderum...@odiso.com>
Envoyé: Mardi 15 Janvier 2013 07:49:35
Objet: Re: [pve-devel] Storage migration: No on-line migration
support
On Tue, 15 Jan 2013 07:12:26 +0100 (CET)
Alexandre DERUMIER <aderum...@odiso.com> wrote:
Are you sure you completly loose the socket ? Or does it hang
temporarly ?
It is completely lost. What triggers it is if some command is timed
out then the next call will generate a lost socket error. This
handles
every time a command times out after 30 sec.
--
Hilsen/Regards
Michael Rasmussen
Get my public GnuPG keys:
michael <at> rasmussen <dot> cc
http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xD3C9A00E
mir <at> datanom <dot> net
http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xE501F51C
mir <at> miras <dot> org
http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xE3E80917
--------------------------------------------------------------
Use free-form input when possible.
- The Elements of Programming Style (Kernighan & Plaugher)
!DSPAM:50f7b55f133081298090781!
--
Hilsen/regards
Michael Rasmussen
_______________________________________________
pve-devel mailing list
pve-devel@pve.proxmox.com
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
