Avoid copying continuation byte arrays by using ByteString internally #1923
Assignees
Labels
performance
Performance issues
Comments
alecgrieser
added a commit
to alecgrieser/fdb-record-layer
that referenced
this issue
Nov 18, 2022
… using ByteString internally This updates the logic in the `RecordCursorContinuation` objects to avoid serializing a continuation all the way to a `byte[]` if it is not necessary. For example, when constructing protobuf objects that back continuations, child continuations can be constructed by calling `.toByteString()` on the continuation and passing that to the protobuf builder instead of serializing all the to a `byte[]` and then wrapping it back in a `ByteString`. This also makes sure that (as much as possible) the cursor implementations are more efficient when constructing a `ByteString`. For example, for the proto-based continuations, it should be able to return `.toByteString()` on the protobuf object instead of serializing the _protobuf_ all the way to a `byte[]` and then wrapping it back up, which some implementations did previously. Adopters who need a `ByteString` rather than a `byte[]` (for example, to pack the continuation into a protobuf) are also encouraged to switch to the newer method. This resolves FoundationDB#1923.
alecgrieser
added a commit
to alecgrieser/fdb-record-layer
that referenced
this issue
Nov 21, 2022
… using ByteString internally This updates the logic in the `RecordCursorContinuation` objects to avoid serializing a continuation all the way to a `byte[]` if it is not necessary. For example, when constructing protobuf objects that back continuations, child continuations can be constructed by calling `.toByteString()` on the continuation and passing that to the protobuf builder instead of serializing all the to a `byte[]` and then wrapping it back in a `ByteString`. This also makes sure that (as much as possible) the cursor implementations are more efficient when constructing a `ByteString`. For example, for the proto-based continuations, it should be able to return `.toByteString()` on the protobuf object instead of serializing the _protobuf_ all the way to a `byte[]` and then wrapping it back up, which some implementations did previously. Adopters who need a `ByteString` rather than a `byte[]` (for example, to pack the continuation into a protobuf) are also encouraged to switch to the newer method. This resolves FoundationDB#1923.
MMcM
added a commit
that referenced
this issue
Nov 21, 2022
Resolves #1923: Avoid copying continuation byte arrays by using ByteString internally
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The
RecordCursorContinuationclass has atoBytes()method which returns abyte[]and atoByteString()method which returns aByteString. Internally, there are quite a few cases where we construct the byte representation by constructing a protobuf, often with child continuations that serialize their continuation to abyte[], which we then wrap in aByteStringand hand to the protobuf serializer.We should be able to skip that intermediate state by using
toByteStringon the continuation object. Some care does need to be taken, though, as some implementations oftoByteString()are (more-or-less)ByteString.copy(toBytes()), which could induce additional copying. But assuming we can tidy that up, this should remove some byte copies and object creation.The text was updated successfully, but these errors were encountered: