Use SerializableLock in hdf5 test

[mrocklin]

I believe that this was the cause of an intermittent error in #661 . This forces dependence on dask master.

[mrocklin]

@shoyer is it sufficient to have a per-process lock for HDF5 reading or do we need a per-machine lock?

[shoyer]

I'm pretty sure per process is enough.
On Mon, Nov 14, 2016 at 6:01 AM Matthew Rocklin notifications@github.com
wrote:

@shoyer https://github.com/shoyer is it sufficient to have a
per-process lock for HDF5 reading or do we need a per-machine lock?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#662 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/ABKS1jbnmb9p2mPP-hLe0Zb_n2eP5qR-ks5q-Gm4gaJpZM4KxSDJ
.

[pitrou]

I've submitted a fix upstream at h5py/h5py#776