Simplify client rpc execution #1993
Merged
397
−437
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Motivation: The existing code for executing RPCs on the client is layered, the one-shot, retry and hedging executors serialize the messages of the requests they're processing into bytes. These 'raw' requests are processed by the stream executor which is structured in such a way as requiring a task group of events it needs to handled. These events are handled in a run method (which is in turn another task for the caller to run). This requires quite a bit of machinery (and allocations) for it to work. Modifications: - Push the serialization/deserialization of messages down into the stream executor. This lets a typed request be transformed directly to `RPCRequestPart`s and for `RPCResponsePart`s to be transformed directly to a typed response (rather than via an intermediary request/response typed to `[UInt8]`). - Remove the sendability requirement from the `next` function for client interceptors. This makes sense: interceptors should be straight-line code so shouldn't be shuffled off into a subtask. This unlocks the ability to remove the task group and event stream previously used by the stream executor as it can add child tasks to a provided task group (rather than needing a separate task group with an event stream). - Apply this change to the one-shot, hedging, and retry executors. Result: - ~35% reduction in allocations for unary RPCs on the client
FranzBusch
reviewed
Jul 25, 2024
| next: @Sendable ( | ||
| next: ( |
There was a problem hiding this comment.
you might want to adopt sending here instead. So that you can call an interceptor on a child task if one wants to
There was a problem hiding this comment.
Yeah I was thinking about that. I was going to go through this module separately and adopt sending since we're still using v5 language mode and I think we need it in a few places.
gjcairo
reviewed
Jul 26, 2024
Comment on lines
28
to
29
| @inlinable // would be private | ||
| static func execute<Input: Sendable, Output: Sendable>( |
There was a problem hiding this comment.
Nit: do we want to rename to _execute?
There was a problem hiding this comment.
Ah thanks. We actually don't want it to be _execute, it should be internal, it's the comment that's wrong.
gjcairo
approved these changes
Jul 29, 2024
clintonpi
pushed a commit
to clintonpi/grpc-swift
that referenced
this pull request
Jul 29, 2024
* Simplify client rpc execution Motivation: The existing code for executing RPCs on the client is layered, the one-shot, retry and hedging executors serialize the messages of the requests they're processing into bytes. These 'raw' requests are processed by the stream executor which is structured in such a way as requiring a task group of events it needs to handled. These events are handled in a run method (which is in turn another task for the caller to run). This requires quite a bit of machinery (and allocations) for it to work. Modifications: - Push the serialization/deserialization of messages down into the stream executor. This lets a typed request be transformed directly to `RPCRequestPart`s and for `RPCResponsePart`s to be transformed directly to a typed response (rather than via an intermediary request/response typed to `[UInt8]`). - Remove the sendability requirement from the `next` function for client interceptors. This makes sense: interceptors should be straight-line code so shouldn't be shuffled off into a subtask. This unlocks the ability to remove the task group and event stream previously used by the stream executor as it can add child tasks to a provided task group (rather than needing a separate task group with an event stream). - Apply this change to the one-shot, hedging, and retry executors. Result: - ~35% reduction in allocations for unary RPCs on the client * use task group void * fix comment
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation:
The existing code for executing RPCs on the client is layered, the one-shot, retry and hedging executors serialize the messages of the requests they're processing into bytes. These 'raw' requests are processed by the stream executor which is structured in such a way as requiring a task group of events it needs to handled. These events are handled in a run method (which is in turn another task for the caller to run). This requires quite a bit of machinery (and allocations) for it to work.
Modifications:
RPCRequestParts and forRPCResponseParts to be transformed directly to a typed response (rather than via an intermediary request/response typed to[UInt8]).nextfunction for client interceptors. This makes sense: interceptors should be straight-line code so shouldn't be shuffled off into a subtask. This unlocks the ability to remove the task group and event stream previously used by the stream executor as it can add child tasks to a provided task group (rather than needing a separate task group with an event stream).Result: