server_timer_proc - better handle server test time expiration

[davidBar-On]

• Version of iperf3 (or development branch, such as master or
  3.1-STABLE) to which this pull request applies:
  make

• Issues fixed (if any): none

• Brief description of code changes (suitable for use as a commit message):
  Enhancements to server_timer_proc() for better handling of server test time expiration

[bmah888]

Thanks for the PR! Will review.

[bmah888]

I tried fixing up a couple merge conflicts so that we can test this a bit with master-branch based code. Let's see how this goes.

[bmah888]

Thanks again for the PR and for your patience! I was about to approve and merge this PR but I thought of a (possibly silly) question. The new code you added to server_timer_proc() closes the socket for each test stream and also removes the stream object from the list of streams. That's all great. But I'm trying to figure out how the thread associated with the stream gets cancelled / ended. From what I can tell, this does happen but I can't tell how or where. Can you help me understand this part?

(Yes I should know because I wrote most of the original thread-handling code.)

[davidBar-On]

I'm trying to figure out how the thread associated with the stream gets cancelled / ended. From what I can tell, this does happen but I can't tell how or where. Can you help me understand this part?

Regarding the question, because the sockets are closed and test->done is set, the worker_run() will terminate either per the test->done or because read/write will fail. That terminates the threads.

Regarding the reason for the PR, I don't remember what was the exact problem, and for some reason I didn't described it, but I believe it was:

1. server_timer_proc() close the control socket.
2. select() fail and cleanup_server(test) or client's goto cleanup_and_fail are executed.
3. The threads are terminated using pthread_cancel().
4. If a thread is reading/writing when it is canceled than that may cause core dump. The purpose of the PR is to prevent this core dump.

If I remember correctly, I added closing the sockets in server_timer_proc() and did not add threads canceling, since I was not sure if canceling will not cause problems in the code executed after select() fails.

[bmah888]

Look good after conflict resolution. Approved, and thanks for answering my questions, will merge this shortly.