_round_timestamp called too often

The way you wrote that report sounds like _round_timestamp() is getting slower and slower, which is more than unlikely.

It's true that 33% of the time is spent in _round_timestamp, which makes sense since it's called 100k time, which also "make sense" since it's used by groupby(). I imagine the time series tested had a few thousand datapoints?

I wonder if groupby() is not called on too many timestamps here. Unless you try to re-aggregate the *whole* time serie – which is *very* unlikely if you appened a few datapoints at the end, it should not be called so often.

it is getting slower but it will speed up again eventually, and then slow down, etc... i'm guessing it's because at some point there is some expiration happening that makes it run _round_timestamp() less.

Fix proposed to branch: master
Review: https://review.openstack.org/299444

yeah, so apparently we are re-aggregating the whole time serie....

Fix proposed to branch: stable/2.0
Review: https://review.openstack.org/300529

Reviewed: https://review.openstack.org/299444
Committed: https://git.openstack.org/cgit/openstack/gnocchi/commit/?id=3bdfaeb67e7041a0ea541143bd3cfc697d8ab5d4
Submitter: Jenkins
Branch: master

commit 3bdfaeb67e7041a0ea541143bd3cfc697d8ab5d4
Author: gordon chung <email address hidden>
Date: Wed Mar 30 09:34:40 2016 -0400

    resample only data affected by new measures

    currently, we are passing in the full unaggregated timeserie
    when we update each of the aggregates for each granularity.
    this is incorrect as the unaggregated timeserie corresponds to the
    largest granularity of the metric. this means that for smaller
    granularities, we are updating and resampling points that are already
    aggregated and not affected by new incoming measures.

    Change-Id: I687c2a18b332494f5c5cb7fdfe6f2b3d1e8de804
    Closes-Bug: #1562820

Reviewed: https://review.openstack.org/300529
Committed: https://git.openstack.org/cgit/openstack/gnocchi/commit/?id=abd58c762d4ea13d24a85f0a6969d31db15710bf
Submitter: Jenkins
Branch: stable/2.0

commit abd58c762d4ea13d24a85f0a6969d31db15710bf
Author: gordon chung <email address hidden>
Date: Wed Mar 30 09:34:40 2016 -0400

    resample only data affected by new measures

    currently, we are passing in the full unaggregated timeserie
    when we update each of the aggregates for each granularity.
    this is incorrect as the unaggregated timeserie corresponds to the
    largest granularity of the metric. this means that for smaller
    granularities, we are updating and resampling points that are already
    aggregated and not affected by new incoming measures.

    Change-Id: I687c2a18b332494f5c5cb7fdfe6f2b3d1e8de804
    Closes-Bug: #1562820
    (cherry picked from commit 3bdfaeb67e7041a0ea541143bd3cfc697d8ab5d4)