Optimize memory usage during materialization #2073
Conversation
judahrand
requested review from
achals,
adchia,
felixwang9817,
tsotnet,
woop and
a team
as code owners
|
Hi @judahrand. Thanks for your PR. I'm waiting for a feast-dev member to verify that this patch is reasonable to test. If it is, they should reply with Once the patch is verified, the new status will be reflected by the I understand the commands that are listed here. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
|
/ok-to-test |
|
@judahrand: Cannot trigger testing until a trusted user reviews the PR and leaves an In response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
Signed-off-by: Judah Rand <[email protected]>
|
/ok-to-test |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: adchia, judahrand 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:
Approvers can indicate their approval by writing |
|
Failures don't look related to PR |
|
FYI @tsotnet looks like an issue with feature server docker images |
Codecov Report
@@ Coverage Diff @@
## master #2073 +/- ##
===========================================
+ Coverage 58.19% 83.31% +25.12%
===========================================
Files 100 100
Lines 8082 8084 +2
===========================================
+ Hits 4703 6735 +2032
+ Misses 3379 1349 -2030
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
What this PR does / why we need it:
I have run memory usage into issues on initial materialization of a large FeatureView with approaching 10 million rows and 30 columns. Acquiring the Arrow table succeeds but the conversion to Protobuf seems to require duplicating the whole dataset in memory which is not practical.
This initial improvement batches the table returned by
to_arrowinto RecordBatches which improves the situation. Looking forward I think that there should be the option (or requirement?) for RetrievalJobs to yield these RecordBatches directly to further reduce memory requirements.This PR improves the situation from #2071 but I do not believe is enough to close it.
Which issue(s) this PR fixes:
Fixes #
Does this PR introduce a user-facing change?: