Daniel P. Berrangé <berra...@redhat.com> wrote:
> On Thu, May 18, 2023 at 09:13:58AM +0000, Wang, Wei W wrote:
>> On Thursday, May 18, 2023 4:52 PM, Wang, Lei4 wrote:
>> > When destination VM is launched, the "backlog" parameter for listen() is
>> > set
>> > to 1 as default in socket_start_incoming_migration_internal(), which will
>> > lead to socket connection error (the queue of pending connections is full)
>> > when "multifd" and "multifd-channels" are set later on and a high number of
>> > channels are used. Set it to a hard-coded higher default value 512 to fix
>> > this
>> > issue.
>> >
>> > Reported-by: Wei Wang <wei.w.w...@intel.com>
>> > Signed-off-by: Lei Wang <lei4.w...@intel.com>
>> > ---
>> > migration/socket.c | 2 +-
>> > 1 file changed, 1 insertion(+), 1 deletion(-)
>> >
>> > diff --git a/migration/socket.c b/migration/socket.c index
>> > 1b6f5baefb..b43a66ef7e 100644
>> > --- a/migration/socket.c
>> > +++ b/migration/socket.c
>> > @@ -179,7 +179,7 @@
>> > socket_start_incoming_migration_internal(SocketAddress *saddr,
>> > QIONetListener *listener = qio_net_listener_new();
>> > MigrationIncomingState *mis = migration_incoming_get_current();
>> > size_t i;
>> > - int num = 1;
>> > + int num = 512;
>> >
>>
>> Probably we need a macro for it, e.g.
>> #define MIGRATION_CHANNEL_MAX 512
>>
>> Also, I think below lines could be removed, as using a larger value of num
>> (i.e. 512)
>> doesn't seem to consume more resources anywhere:
>> - if (migrate_use_multifd()) {
>> - num = migrate_multifd_channels();
>> - } else if (migrate_postcopy_preempt()) {
>> - num = RAM_CHANNEL_MAX;
>> - }
>
> Given that this code already exists, why is it not already sufficient ?
Ah, I "think" I remember now.
> The commit description is saying we're setting backlog == 1 wit
> multifd, but this later code is setting it to match the multfd
> channels. Why isn't that enough ?
Are you using -incoming defer?
No? right.
With multifd, you should use -incoming defer. It is more, you should
use -incoming defer always. The problem is that the way qemu starts,
when we do the initial listen, the parameters migration_channels and
migration_multifd hasn't yet been parsed.
Can you confirm that if you start with -incoming defer, everything works
as expected?
Later, Juan.
