QA (public) » openQA Project (public)

Right now, the "signals" are implemented by placing files "in the way" of os-autoinst. Without changing it much, we can have a Mojolicious::Lite script just listening in the same machine that os-autoinst and then converting those proper calls into hacky files. So, for me, "worker" should probably become a Mojolicious::Lite script (just an idea).

ancorgs wrote:

So, for me, "worker" should probably become a Mojolicious::Lite script (just an idea).

IDK, maybe the solution is to avoid the hacky files and create a proper loop in the os-autoinst part that monitor the channel and affect to the test.

Now is something like this:

[ Worker ] -> [ os-autoinst -> Test ]

Worker is a process and os-autoinst + Test is another process. The loop in os-autoinst take care of run the test in the same process calling the run() function inside the test. So when os-autoinst delegate the execution with the test this can;t control it anymore, except when the test goes out in a waitforneedle.

This lost of controls is IMO the problem, and the solutions are basically a variant of two:

1) Cooperative: Intercept every API that delegate again the control of the test to os-autoinst (sendkey[w], waitforneedle, check_screen, take_screenshot...). Is like a yield in a cooperative multithread env.

2) Preemptive: os-autoins forks the test loop and takes control from outside in a different loop. Basically it can kill, remove and restart the process that contains the tests loop from outside

Any ideas?

telnet localhost 20013
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
GET /
HTTP/1.1 400 Bad Request
Content-Length: 12
Connection: close
Content-Type: text/html;charset=UTF-8
Server: Mojolicious (Perl)
Date: Wed, 26 Nov 2014 12:29:37 GMT

Hello World!