aoe: flush aoe device when finished serving

[nak3]

fixes #389

[lpabon]

Please add a comment to why this is necessary for others to understand at some future date.

[lpabon]

Why use defer?

[nak3]

Sure, I updated and added the comment.

[xiang90]

@nak3 Did you verify this that this patch fixes #389?

[nak3]

Yes, I did verify it.

(test record)

1. Start and stop torusblk with aoe block device

[vagrant@torus1 torus]$ sudo ./bin/torusblk aoe test1 --debug lo 1 1
...
Attached to AoE device (lo, Major: 1, Minor: 1). Server loop begins ...
...
2016-12-13 12:28:08.869225 D | block: Syncing block volume: test1
^C
Received an interrupt, stopping services...
2016-12-13 12:28:11.547470 E | aoe: failed to handle frame: bad file descriptor
2016-12-13 12:28:11.547612 E | aoe: ReadFrom failed: bad file descriptor
2016-12-13 12:28:11.556207 D | aoe: canceling background goroutines
2016-12-13 12:28:11.556368 D | aoe: waiting for background goroutines to stop
2016-12-13 12:28:11.564921 D | aoe: stopping device sync goroutine
2016-12-13 12:28:11.565137 D | block: not syncing

2. Confirm /dev/etherd/e1.1 has been removed

[vagrant@torus1 torus]$ ls /dev/etherd/p
discover err flush interfaces revalidate

3. fdisk works

[vagrant@torus1 torus]$ sudo fdisk -l > /dev/null
[vagrant@torus1 torus]$ echo $?
0

[xiang90]

lgtm. defer to @lpabon

[mischief]

would this be better as a defer, or possibly in the aoe server's Close()?

[nak3]

I got a review #412 (comment) before. And I agreed with reviewer that there are no reason to use defer.

For the idea of aoe server's Close(), as.Close() here should be called after NewServer() but before as.Serve(). So, since flush() should be surely called after the device was created, it should not be called with defer as.Close().