Github user geomacy commented on the issue:
https://github.com/apache/brooklyn-server/pull/791
You're right @ahgittin, the `sync` should be done before the `close`; I
have added another commit to do this.
I agree I'm not sure that the problem was caused by a failure to commit
data to disk when @aledsage's laptop crashed, but it's certainly plausible.
That's an interesting link you shared above. I've done a bit more digging and
come across a very interesting email here:
https://lists.apple.com/archives/darwin-dev/2005/Feb/msg00072.html
It's definitely worth reading; in short, it looks like just adding the
`sync()` in Brooklyn, while it may help, isn't in itself enough to _ensure_ the
data is committed to disk on pretty much any system, not just OSX. :-(
In fact the JNI call that you referred to on OSX seems to be the only way
to guarantee the data is committed, and of course it's only available on the
Mac.
I think this PR may still be something worth doing (it will eliminate one
failure point, by ensuring the data is flushed from host memory to the disk
device). I don't know what else we might want to do to take this further,
would be interested to hear what you think.
---
