msgq dies on socket send error

[shane]

Jeremy reported an error to the bind10-dev list:

​https://lists.isc.org/pipermail/bind10-dev/2011-August/002558.html

The included trace ends like this:

"/b/work/BIND10-systest/20110812084501-MacOS/build/src/bin/msgq/b10-msgq", 
line 344, in __send_data
    return sock.send(data)
socket.error: [Errno 32] Broken pipe

We need to wrap this, so that msgq does not exit.

[jelte]

Ready for review. As discussed on the planning call, I only caught the exception, and only SIGPIPE, and did not add any tests for this. If someone sees an easy way to do so, we can, but i'd prefer spending time looking at a replacement :)

Minor other notes; this module also doesn't use our logging framework, so as the error response (apart from dropping the send action and closing the socket), I just added a print(), as the rest of this code uses. Also, my editor found some whitespace it automatically removed.

[vorner]

Hello

Maybe the message should be „Dropping the connection and message“ instead of just message.

Should it have a changelog that the bug should be gone now?

I also noticed that the import pprint is not needed, there are only two commented-out calls to it. Should it be cleaned out?

And, finally, the branch does not compile for me. It fails in some dhcp-related C++ tests, even after merging with master. Should we wait for it to clean up? Do you have an idea if Tomek knows?

Thanks

[jelte]

Replying to vorner:

Hello

Maybe the message should be „Dropping the connection and message“ instead of just message.

made it "dropping message and closing connection"

Should it have a changelog that the bug should be gone now?

ohyeah, how about:

1. [defect] jelte

msgq no longer crashes if the remote end is closed while msgq tries to send data. It will now simply drop the message and close the connection itself.

I also noticed that the import pprint is not needed, there are only two commented-out calls to it. Should it be cleaned out?

removed

And, finally, the branch does not compile for me. It fails in some dhcp-related C++ tests, even after merging with master. Should we wait for it to clean up? Do you have an idea if Tomek knows?

luckily i missed this :) I'm assuming this has been fixed now, but either way this change should be orthogonal (except that perhaps now you haven't verified whether msgq now even runs at all)

[vorner]

Thanks, this looks OK to merge.

[jelte]

Thanks, merged, closing ticket.