Convert python values into proto values in bulk

[pyalex]

Signed-off-by: pyalex [email protected]

What this PR does / why we need it:

Packing python native values (or NumPy arrays) into proto objects (generated proto classes) takes roughly 80% of the time during materialization (not counting offline / online storage part). Ironically protobuf serialization itself takes around 10% of the time.

This PR refactors python_values_to_proto_values function to process data in bulk (essentially column by column), which improves performance by running all type checks only once (per batch / column). According to my analysis, this move will decrease packing time roughly by 50% without introducing significant changes (see first comment).

Which issue(s) this PR fixes:

Fixes #

Does this PR introduce a user-facing change?:

none

[codecov-commenter]

Codecov Report

Merging #2172 (a65682c) into master (c9ff695) will increase coverage by 0.00%.
The diff coverage is 85.00%.

@@           Coverage Diff           @@
##           master    #2172   +/-   ##
=======================================
  Coverage   84.58%   84.58%           
=======================================
  Files         102      102           
  Lines        8186     8195    +9     
=======================================
+ Hits         6924     6932    +8     
- Misses       1262     1263    +1     

Flag	Coverage Δ
integrationtests	74.27% <85.00%> (-0.27%)	⬇️
unittests	59.03% <52.50%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
sdk/python/feast/type_map.py	72.65% <77.77%> (-0.52%)	⬇️
sdk/python/feast/feature_store.py	91.39% <100.00%> (+0.03%)	⬆️
sdk/python/feast/infra/provider.py	90.09% <100.00%> (+0.18%)	⬆️
sdk/python/feast/online_response.py	87.71% <100.00%> (ø)
.../integration/online_store/test_universal_online.py	98.13% <0.00%> (+0.46%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c9ff695...a65682c. Read the comment docs.

[pyalex]

My performance analysis was based on case proposed by @judahrand:
Feature view with 3 features of primitive type and 3 features of list type:

fv = FeatureView(
    name="fv",
    entities=["fake_entity"],
    features=[
        Feature(
            name=str(i),
            dtype=ValueType.DOUBLE,
        ) for i in range(3)
    ] + [
        Feature(
            name=str(i),
            dtype=ValueType.FLOAT_LIST,

        ) for i in range(5, 8)
    ],
    ttl=None,
    batch_source=FakeSource,
)

Dataset with 10 thousands rows was generated

ROWS = 10_000
data = {
    'id': np.random.random_integers(low=0, high=int(1e12), size=(ROWS,))
}
data.update(
    {
        str(i): np.random.random(size=(ROWS,)).astype('float64')
        for i in range(3)
    }
)
data.update(
    {
        str(i): [np.random.random(size=(10,)).astype('float32') for _ in range(ROWS)]
        for i in range(5, 8)
    }
)

feast.infra.provider._convert_arrow_to_proto was benchmark target ran on my local machine.

Results:
master: 570 ms (average)
this PR: 320 ms (average).

[judahrand]

This looks great!

[adchia]

Mind making an issue tracking this after the PR goes in?

[adchia]

/lgtm

[feast-ci-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: adchia, pyalex

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [adchia]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment