NDT-DEV email list created

[Aaron Brown <>]

Hey Peter,

I'm curious about this:

  * web100clt currently (as of 2015-12-21) requires that its output queue be
  * drained at the end of the c2s test and crashes if it is not. We don't want
  * to break it, but that extra second or so should not be part of our measured
  * transmitted data, because the spec says that it's a 10 second upload from
  * client to server, not a "10 seconds or a little bit more depending on when
  * the client starts counting" upload.  Therefore, here we read from the
  * sockets for just a little longer, but ignore the resulting data, so that old
  * clients don't crash, but newer clients are not penalized for quitting after
  * 10 seconds, just like the spec says they can.

Where was this crashing?

Cheers,

Aaron

On Tue, Jan 26, 2016 at 2:12 PM, Peter Boothe <> wrote:

We've fixed a lot of NDT bugs on MLab's platform in the past month. We think we have reached a stable point and it is worth incorporation back into the upstream system.

Please check out the pull request at https://github.com/ndt-project/ndt/pull/202 - it significantly improves the log messages, fixes a lot of small bugs, fixes a lot of race conditions that only happen at scale, fixes a lot of bugs having to do with signal handling and doing the right thing when errno==SIGINT, and it also fixes some bugs that only occur when you turn on debug logging (!).  All-in-all, we think we've done some good to the codebase and would like to contribute these changes back upstream.

  -Peter

--

ᴹ̶LAB | Measure the Internet, save the data, and make it universally accessible and useful.