Make BTrDB connections serializable in Ray

[gtfierro]

This PR adds a new shareable flag to the connect() method. Connections created in this way can be serialized and shared among Ray workers in a Ray cluster. However, this is potentially insecure: it is possible (POC in development) to pull an API key out of the serialized object as it rests in the Plasma store. However, this requires access to the same Ray cluster.

Uses the semver package to choose the right Ray serialization method based on which version of Ray is installed.

Usage looks as follows:

import ray
import btrdb
ray.init(ignore_reinit_error=True)
conn  =  btrdb.connect(profile='collab', shareable=True)
# /opt/conda/lib/python3.7/site-packages/btrdb/__init__.py:81: UserWarning: a shareable connection is potentially insecure; other users of the same cluster may be able to access your API key
conn_copy = ray.get(ray.put(conn))  #ok

import ray
import btrdb
ray.init(ignore_reinit_error=True)
conn  =  btrdb.connect(profile='collab', shareable=False)
conn_copy = ray.get(ray.put(conn))  # raises Serialization Exception

Clubhouse story [ch9329]

[shortcut-integration]

This pull request has been linked to Clubhouse Story #9329: Update btrdb-python for Ray 1.0 compatibility.

[mchestnut91]

This all looks good to me. I verified that it works with this dummy function:

import ray
import btrdb

@ray.remote
def get_info(i, conn):
    return f"{i} {conn.info()}"

ray.init(ignore_reinit_error=True)
conn = btrdb.connect(profile="collab", shareable=True)

infos = [get_info.remote(i, conn) for i in range(3)]
print(ray.get(infos)) # prints connection info 3 times

[mchestnut91]

lgtm. Thanks for figuring this out!

[RamiroCope]

Just took it for a spin and was able to get data out. Setting shareable=False as default also makes it not break Data Science's code base, which is nice.