not-raspberry / spawn_and_check Goto Github PK

When killing the process (after post-checks fail), first try to terminate it with SIGTERM, wait for termination for some time and then send SIGKILL.

The waiting can be done similarly to checks polling so having #3 resolved first would definitely help.

Running a service with execute and at least one check may fail if the previous service didn't terminate on time:

1. Process X₁ is ran with the executor.
2. SIGTERM is sent to the process X₁ but the service takes some time to tear down
3. Process X₂ is ran with the executor - pre-checks fail because they see the previous service.
4. X₁ terminates after some time but it's too late - X₂'s executor failed.

DRY polling - currently there is a simple wait_until implementation in test.wait and a polling loop inside the execute function. There will be another polling for the pre-checks so the rule of three has to be applied.

This will probably require changing the check functions' signature to subprocess.Popen -> bool.

UDP connecion can fail on ICMP level (The destination host responds with 'Port Unreachable' which gets translated to 'Connection Refused' on UDP level). People are experiencing it (http://stackoverflow.com/questions/16130786/why-am-i-getting-the-error-connection-refused-in-python-sockets) however I was unable to reproduce it locally - which is strange because I have dealt with failing UDP connections in the past.

Provide a code that raises socket.error with the proper errno from a UDP connection.

Longest test durations:

10.02s call     test/unit/test_executor.py::test_execute_pre_checks_fail[5]
10.02s call     test/unit/test_executor.py::test_execute_pre_checks_fail[3]
10.02s call     test/unit/test_executor.py::test_execute_pre_checks_fail[4]
10.02s call     test/unit/test_executor.py::test_execute_pre_checks_fail[2]
10.02s call     test/unit/test_executor.py::test_execute_pre_checks_fail[1]
5.17s call     test/integration/test_spawn_and_check.py::test_execute_check_http[5]
5.15s call     test/integration/test_spawn_and_check.py::test_execute_check_tcp[5]
5.11s call     test/integration/test_spawn_and_check.py::test_execute_check_unix[5]
4.09s call     test/integration/test_spawn_and_check.py::test_execute_failing_checks
3.26s call     test/integration/test_spawn_and_check.py::test_execute_same_service_sequentially
3.16s call     test/integration/test_spawn_and_check.py::test_execute_check_tcp[3]
3.15s call     test/integration/test_spawn_and_check.py::test_execute_check_http[3]
3.11s call     test/integration/test_spawn_and_check.py::test_execute_check_unix[3]
1.14s call     test/integration/test_spawn_and_check.py::test_execute_process_killed

Those tests are mostly sleeping. It would be nice to e.g. reduce timeouts for unit tests.

Another, more complicated way is to mock time.sleep and time.time.