[igt-dev] [PATCH i-g-t] lib/igt_kms: only check chamelium's presence in chamelium tests

[Martin Peres]

On 22/03/2021 10:39, Kunal Joshi wrote:
>> Yeah, but then that would not protect against tests that would forget to
>> re-plug the screens and thus would lead to unstable IGT results when the
>> test list is randomized (shard runs).
> 
> Correct, but from the recents runs i have observed that chamelium
> crashes in between (chameleond stops), There is no reply coming back
> and the test gives SIGQUIT.
> 
> I will try to come up with a non blocking plug all display,
> 
>> I think the idea was that the overhead of checking was measured in ms on
>> devices that have a chamelium, and no impact at all on machines without a
>> chamelium, so the convenience was deemed greater than the cost.
>>
>> Do you have specific measurements that would contradict this?
> 
> For this when chamelium is reachable its neglible in ms
> When chamelium is not reachable or crashed it takes 1m15.654s to ~4m causing inactivity timeout
> 
> The problem is chamelium crashes in between the runs not when the admin
> or someone starts it so time wasted is multiplied by all the tests we
> run.
> 
> So probably we have to find some other way of dealing with this,
> Either decreasing xmlrpc timeout below inactivity timeout or non
> blocking xmlrpc request.

I think the answer is pretty clear here: fix the problem at the source, 
don't work around it. You know the code is open source, right?

Martin

> 
> Thanks for the feedback
> 
> Regards
> Kunal Joshi
>>
>>>
>>> Currently when chamelium is not reachable we get SIGQUIT on all the
>>> tests
>>>
>>> Starting subtest: basic-fence-flip
>>> Received signal SIGQUIT.
>>> Stack trace:
>>> #0 [fatal_sig_handler+0xd6]
>>> #1 [killpg+0x40]
>>> #2 [pselect+0x59]
>>> #3 [xmlrpc_server_info_disallow_auth_ntlm+0x389]
>>> #4 [xmlrpc_server_info_disallow_auth_ntlm+0x786]
>>> #5 [xmlrpc_server_info_disallow_auth_ntlm+0xd1f]
>>> #6 [xmlrpc_client_call2+0xa8]
>>> #7 [xmlrpc_client_call2f_va+0x9e]
>>> #8 [__chamelium_rpc_va+0x87]
>>> #9 [__chamelium_rpc.constprop.8+0x8c]
>>> #10 [chamelium_wait_reachable+0x65]
>>> #11 [igt_display_require+0xbe]
>>> #12 [test_flip+0x4b]
>>> #13 [__real_main1093+0x1c38]
>>> #14 [main+0x27]
>>> #15 [__libc_start_main+0xe7]
>>> #16 [_start+0x2a]
>>>
>>> When chamelium is not reachable chamelium_wait_reachable() never returns
>>> back and we have inactivity timeout which kills the test.
>>
>> Isn't that exactly what we want? This seems to be a setup error, and we get
>> a clear message: we can't check the state of the chamelium, so better not
>> run KMS tests which depend on a state we can't check.
>>
>>>
>>> Regards
>>> Kunal Joshi
>>>