Timestamp generator

[mambocab]

Addresses PYTHON-676. The history isn't great, but I plan to squash before/on merge if there are no objections.

Unit and integration tests pass locally on py2.7. I have not tested this over actual timestamp discontinuities.

[aboudreault]

LGTM Jim. Can you give a try to our benchmark scripts to see how the lock in that generator affects the performance?

[beltran]

I think this line should be:
diff = self.last - now
as this value may be logged below

[mambocab]

Yep, good catch!

[beltran]

This comparison would be:
self.last + 1 - now >= self.warning_threshold
as the comparison should be done with the value that is going to be returned right?

[mambocab]

I think it should be since_last_warn >= self.warning_threshold -- we're trying to avoid flooding logs, so we care about how long it's been since the last warning was issued. Sound right?

[beltran]

I think that is checked in the next line right?
since_last_warn >= self.warning_interval

This lines checks the returned timestamp drifts more than warning_threshold seconds into the future.

[mambocab]

Yep -- sorry, I should've spent more time reading for context.

[mambocab]

Sorry again for my confusion. To address your original concern: I think either behavior (comparing to self.last - now or to self.last + 1 - now) is reasonable. I had a look at at the Java implementation:

https://github.com/datastax/java-driver/blob/da72105c2fcef8f0865047815df146cfbb2354e6/driver-core/src/main/java/com/datastax/driver/core/AbstractMonotonicTimestampGenerator.java#L60

https://github.com/datastax/java-driver/blob/da72105c2fcef8f0865047815df146cfbb2354e6/driver-core/src/main/java/com/datastax/driver/core/LoggingMonotonicTimestampGenerator.java#L65

It looks like the behavior there is the same as that in the PR -- so self.last - now. Personally, I don't see a good reason to make the implementation work differently than the Java driver's. Do you object to it strongly enough?

[beltran]

Yeah, self.last - now should be fine

[beltran]

I've changed some things about the seconds / microseconds inside the class MonotonicTimestampGenerator. Maybe a review would be worth it.

[mambocab]

@bjmb I've added a small fix that tears down the patched time.time. I'm +1 to your changes, they look great. Unless you object, I'll squash my initial commits together, squash the last 3 together, and merge.

[beltran]

@mambocab great! Sure, go ahead with the squash and merge.