[pve-devel] [PATCH qemu-server] fix #4549: capture and bubble up qemu errors correctly

[Thomas Lamprecht]

Am 12/06/2023 um 13:22 schrieb Fiona Ebner:
> Am 09.05.23 um 10:46 schrieb Christoph Heiss:
>> @@ -5922,13 +5916,19 @@ sub vm_start_nolock {
>>  		$tpmpid = start_swtpm($storecfg, $vmid, $tpm, $migratedfrom);
>>  	    }
>>  
>> +	    # '\r\n' needs to be used as a line separator here, as that is what run_command() splits
>> +	    # lines on. If just a newline is used, all error message lines end up on a single line
>> +	    # in the migration tasklog.
> 
> So this is essentially a hack to work around the fact that we don't call
> print once for each line anymore ;)
> 
>> +	    my $err = '';
>> +	    $run_params{logfunc} = sub { $err .= "QEMU: $1\r\n"; };
> 
> Style nit: please use $run_params->{logfunc}

does not work here as this is an actual hash value, not the (for our perl code more common)
hash ref.

my %foo = (a => 1);
$foo{b} = 2;

vs.

my $foo = {a => 1};
$foo->{b} = 2;

> 
> So you're eating up all the log lines here...
> 
>> +
>>  	    my $exitcode = run_command($cmd, %run_params);
>>  	    if ($exitcode) {
>>  		if ($tpmpid) {
>>  		    warn "stopping swtpm instance (pid $tpmpid) due to QEMU startup error\n";
>>  		    kill 'TERM', $tpmpid;
>>  		}
>> -		die "QEMU exited with code $exitcode\n";
>> +		die "QEMU exited with code $exitcode\r\n$err";
> 
> ...and only print them in the failure case. This means all non-critical
> errors/warnings/info will never see the light of day anymore? To get a
> warning for testing you can use e.g.
> qm set 100 --args '-no-hpet'
> with QEMU >= 8.0.
> 
> The current approach also produces a leading space in the task logs for
> the follow-up lines:
> 
>> TASK ERROR: start failed: QEMU exited with code 1
>>  QEMU: kvm: -no-hpet: warning: -no-hpet is deprecated, use '-machine hpet=off' instead
>>  QEMU: bridge 'vmbr1' does not exist
>>  QEMU: kvm: -netdev type=tap,id=net1,ifname=tap100i1,script=/var/lib/qemu-server/pve-bridge,downscript=/var/lib/qemu-server/pve-bridgedown,vhost=on: network script /var/lib/qemu-server/pve-bridge failed with status 512
> 
> Instead of getting rid of the "print once per line" handling, can't we
> rather adapt the invocation for backup to bubble up the other log lines too?

would it make sense to differ between stderr/stdour and do log_warn and print,
respectively?

> 
> Always prefixing the messages with "QEMU: " seems fine to me (even
> things like the "bridge 'vmbr1' does not exist" message technically come
> from the pve-bridge script, it's still part of the QEMU invocation ;)).

not sure if we really need a prefix at all here..