Skip to content

Conversation

@ehigham
Copy link
Member

@ehigham ehigham commented Jun 5, 2024

Fixes: #14559
hl.nd.arrays constructed from stream pipelines can cause out of memory exceptions owing to a limitation in the python CSE algorithm that does not eliminate partially redundant expressions in if-expressions. Explicitly let-binding the input collection prevents it from being evaluated twice: once for the flattened data stream and once for the original shape.

Fixes: hail-is#14559
hl.nd.arrays constructed from stream pipelines can cause out of memory
exceptions owing to a limitation in the python CSE algorithm that does
not eliminate partially redundant expressions if if-expressions.

Explicitly let-binding the input collection prevents it from being evaluated
twice: once for the flattened data stream and once for the original shape.
@ehigham ehigham force-pushed the ehigham/fixes/14559-ndarray-oom branch from 0056b39 to 697b700 Compare June 5, 2024 00:52
@ehigham ehigham marked this pull request as ready for review June 5, 2024 17:11
@ehigham ehigham requested a review from patrick-schultz June 5, 2024 17:50
@hail-ci-robot hail-ci-robot merged commit 545fa5b into hail-is:main Jun 6, 2024
@ehigham ehigham deleted the ehigham/fixes/14559-ndarray-oom branch June 6, 2024 15:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

MakeNDArray OOM on stream data

3 participants