Termination max_time criterion

[JoelPasvolsky]

From the current defintion of `max_time as:
"max_time (float/None, optional, default=None):
Wall clock runtime termination criterion. Unlimited by default."

I wouldn't expect this:

import timeit
start_time = timeit.default_timer()
result = sampler.sample(bqm, num_reads=5, max_iter=2, **max_time=6**)
print(timeit.default_timer() - start_time)

print("Found solutions with energy", result.record.energy)

results in

**27.63140569999996**
Found solutions with energy [-139. -137. -141. -139. -139.]

I'd have expected that the sampler operation would terminate in ~6 six of wall time.

Also, be nice to add the units to the definition: "... termination criterion, in seconds. Unlimited"

[randomir]

What sampler are you talking about here?

[JoelPasvolsky]

sampler = KerberosSampler()

[randomir]

If it's Kerberos, then there's no way to guarantee the run time (currently). max_time parameter is on the outer Loop, and if loop body (racing branches with QPU in one branch) takes longer, the Loop will not forcefully interrupt the child workflow (source here). Maybe we could do that, but it probably makes more sense just to document the current behavior instead.

[JoelPasvolsky]

I think what surprised me most was not that it doesn't have the granularity of interrupting the loop body, but that it applies to the loop body rather than the sampler as a whole, that is I expected that when the first RacingBranch completed with more than max_time having elapsed since iteration 0 began, it would terminate.
Yes, I think clearly documenting what we have is enough.

[randomir]

Although, limiting branches' execution time to max_time also makes a lot of sense. We should do that. The only problem is that QPU sampler does not yet support timeout (#121). So, we can implement this once we complete #121.