Background

There was a severe performance regression in string deserialization introduced in bson v5 (https://docs.google.com/spreadsheets/d/1Wogm3APTMcZJK4BjXVx5xTaIGQkrtr38GzP89Nd3FV8/edit?usp=sharing) which is related to the implementation of ByteUtils.toUTF8 which is called primarily by getValidatedString in src/parser/deserializer.ts (see flamegraph below.) This issue affects all currently released versions of the driver starting from v5.0.0 which is where the dependency on bson v5 was added.

Unknowns

• What would the impact of trying to reduce/eliminate calls to ByteUtils.toLocalBufferType?
• How difficult will it be to get the Typescript to work as we'd like it to if we remove ByteUtils.toLocalBufferType calls?

Acceptance Criteria

• Update ByteUtils.toUTF8 in src/utils/node_byte_utils.ts to take in start and end and pass them through to the call to toString rather than calling Buffer.subarray directly inside getValidatedString.
  • Address any typescript issues this causes
• Investigate how to reduce/eliminate calls to ByteUtils.toLocalBufferType in ByteUtils.toUTF8
• Audit the remainder of ByteUtils to investigate if there is anywhere else that we are taking an unnecessary performance hit
• Investigate best approach to adding regression tests specifically for the performance of getValidatedString and possibly of ByteUtils.toUTF8 and add these tests.