[pve-devel] [PATCH v4 qemu-server 15/16] memory: virtio-mem : implement redispatch retry.

[Fiona Ebner]

Am 13.02.23 um 13:00 schrieb Alexandre Derumier:
> If some memory can be removed on a specific node,
> we try to rebalance again on other nodes
> 
> Signed-off-by: Alexandre Derumier <aderumier at odiso.com>
> ---
>  PVE/QemuServer/Memory.pm | 51 +++++++++++++++++++++++++++-------------
>  1 file changed, 35 insertions(+), 16 deletions(-)
> 
> diff --git a/PVE/QemuServer/Memory.pm b/PVE/QemuServer/Memory.pm
> index bf4e92a..f02b4e0 100644
> --- a/PVE/QemuServer/Memory.pm
> +++ b/PVE/QemuServer/Memory.pm
> @@ -201,13 +201,28 @@ my sub get_virtiomem_total_current_size {
>      return $size;
>  }
>  
> +my sub get_virtiomem_total_errors_size {
> +    my ($mems) = @_;
> +
> +    my $size = 0;
> +    for my $mem (values %$mems) {
> +	next if !$mem->{error};
> +	$size += $mem->{current};
> +    }
> +    return $size;
> +}
> +
>  my sub balance_virtiomem {
>      my ($vmid, $virtiomems, $blocksize, $target_total) = @_;
>  
> -    my $nb_virtiomem = scalar(keys %$virtiomems);
> +    my $nb_virtiomem = scalar(grep { !$_->{error} } values $virtiomems->%*);
>  
>      print"try to balance memory on $nb_virtiomem virtiomems\n";
>  
> +    die "No more available blocks in virtiomem to balance all requested memory\n"
> +	if $target_total < 0;

I fee like this message is a bit confusing. This can only happen on
unplug, right? And reading that "no more blocks are available" sounds
like a paradox then. It's rather that no more blocks can be unplugged.

If we really want to, if the $target_total is negative, we could set it
to 0 (best to do it at the call-side already) and try to unplug
everything else? We won't reach the goal anymore, but we could still get
closer to it in some cases. Would need a bit more adaptation to avoid an
endless loop: we also need to stop if all devices reached their current
goal this round (and no new errors appeared), e.g. balance_virtiomem()
could just have that info as its return value.

Example:
> update VM 101: -memory 4100,max=65536,virtio=1
> try to balance memory on 2 virtiomems
> virtiomem0: set-requested-size : 0
> virtiomem1: set-requested-size : 4
> virtiomem1: last: 4 current: 4 target: 4
> virtiomem1: completed
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: increase retry: 0
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: increase retry: 1
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: increase retry: 2
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: increase retry: 3
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: increase retry: 4
> virtiomem0: last: 16 current: 16 target: 0
> virtiomem0: too many retry. set error
> virtiomem0: increase retry: 5

Currently it stops here, but with setting $target_total = 0 it continues...

> try to balance memory on 1 virtiomems
> virtiomem1: set-requested-size : 0
> virtiomem1: last: 4 current: 0 target: 0
> virtiomem1: completed

...and gets closer to the goal...

> try to balance memory on 1 virtiomems
> virtiomem1: set-requested-size : 0
> virtiomem1: last: 4 current: 0 target: 0
> virtiomem1: completed
> try to balance memory on 1 virtiomems
> virtiomem1: set-requested-size : 0
> virtiomem1: last: 4 current: 0 target: 0
> virtiomem1: completed

...but then it loops, because I didn't add the other stop condition yet
;). But not sure, likely too much magic.

> +    die "No more available virtiomem to balance the remaining memory\n" if $nb_virtiomem == 0;

"No more virtiomem devices left to try to ..." might be a bit clearer.
Technically, they are still available, we just ignore them because they
don't reach the target in time.

> +
>      #if we can't share exactly the same amount, we add the remainder on last node
>      my $target_aligned = int( $target_total / $nb_virtiomem / $blocksize) * $blocksize;
>      my $target_remaining = $target_total - ($target_aligned * ($nb_virtiomem-1));