[Dataflow Streaming] Avoid calling WorkItem::getSerializedSize

[arunpandianp]

Prior to this change WorkItem::getSerializedSize is called on the GetWorkStream thread for every work item, it shows up like in cpu profiles. Replacing it with the byteSize computed from the response bytes makes the work submit path a bit more efficient

#33578

[arunpandianp]

Run Java Precommit

[arunpandianp]

R: @scwhittle

[github-actions]

Stopping reviewer notifications for this pull request: review requested by someone other than the bot, ceding control. If you'd like to restart, comment assign set of reviewers

[scwhittle]

I was thinking that since java protos are immutable that this would be cheap. But it may not cache the size internally so seems ok to change.

Do you have any #s from testing/profiles to add to description as motivation?

[arunpandianp]

Run Java Precommit

[arunpandianp]

I was thinking that since java protos are immutable that this would be cheap. But it may not cache the size internally so seems ok to change.

Do you have any #s from testing/profiles to add to description as motivation?

added a profile screenshot. The results are cached internally https://stackoverflow.com/a/1903792.
But it was getting called in the Getwork loop for the first time and calculating size for all workitems in a single thread seems costly.

[arunpandianp]

Run Java Unit Tests