Issues with pyrtma bench

[tjmadonna]

pyrtma_bench.py needs to be updated to use the latest pyrtma class names (ex: rtmaClient -> Client). There may be more.

[jmw182]

The examples may also need to be updated (e.g. I changed types -> internal_types).

[jmw182]

The examples seem to work now. @tjmadonna can this issue be closed?
EDIT: Nevermind, now I see that pyrtma_bench has not been updated yet.

[jmw182]

I updated pyrtma_bench in 9b1b9fc. It mostly works again, but still fails if configured to run with publishers but no subscribers. I'm not sure why subscribers are required to avoid this error:

Traceback (most recent call last):
File "pyrtma_bench.py", line 225, in
msg = mod.read_message(timeout=0.100)
File "c:\git\pyrtma\pyrtma\client.py", line 233, in read_message
), "Did not send all the header to message manager."
AssertionError: Did not send all the header to message manager.

[jmw182]

The above error does not occur when using the c++ message manager. Running with 1 publisher and 0 subscribers works with the c++ manager, but using > 1 publisher and 0 subscribers or 0 publishers and subscribers causes the test to hang in the main while loop. It sends the Exit signal after the timeout period, but this does not do anything since there are no subscribers to receive it.

[jmw182]

I renamed this issue since the remaining bugs I described on September 29 are unrelated to class names.

[jmw182]

Recap from debugging this week and meeting with both Tylers today:
At this point the issue seems to happen with both message managers, and with either pyrtma_bench.py or rtma_bench.exe.
The issue appears to be related to unsynchronized subprocesses. When one publisher process finishes before the others even begin, the main process does not get all of the publisher done signals and hangs. This is more likely to happen when there are no subscribers (as the publisher thread normally waits for all subscribers to be ready prior to their send loops), or when the number of messages per publisher is small (such that the processes can finish very quickly).

Adding a pause BEFORE sending the publisher done signal can resolve the issue, but a pause AFTER the signal and before disconnecting does NOT resolve the issue.