-
Notifications
You must be signed in to change notification settings - Fork 667
Moving PooledRecvBufferAllocator
from NIOPosix to NIOCore
#3110
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@@ -12,10 +12,8 @@ | |||
// | |||
//===----------------------------------------------------------------------===// | |||
|
|||
import NIOCore | |||
|
|||
/// A receive buffer allocator which cycles through a pool of buffers. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should extend this to express when users might want to use this type.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is this addition to the comment accurate and sufficient: This type is useful when allocating new buffers for every IO read is expensive or undesirable.
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Allocating is always expensive, so I'm not sure this is particularly insightful. Maybe something like:
Channels can read multiple times per cycle (based on
ChannelOptions.maxMessagesPerRead
), and they reuse the inbound buffer for each read. If aChannelHandler
holds onto this buffer, then CoWing will be needed. APooledRecvBufferAllocator
cycles through preallocated buffers to avoid CoWs during the same read cycle.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added your suggestion in the documentation.
mutating func record(actualReadBytes: Int) { | ||
/// - Parameters: | ||
/// - actualReadBytes: Number of bytes being recorded | ||
/// - Returns: whether the next buffer will be larger than the last. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This isn't actually returning anything so I'd get rid of this line
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Good catch. Removed from the code.
@@ -12,10 +12,8 @@ | |||
// | |||
//===----------------------------------------------------------------------===// | |||
|
|||
import NIOCore | |||
|
|||
/// A receive buffer allocator which cycles through a pool of buffers. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Allocating is always expensive, so I'm not sure this is particularly insightful. Maybe something like:
Channels can read multiple times per cycle (based on
ChannelOptions.maxMessagesPerRead
), and they reuse the inbound buffer for each read. If aChannelHandler
holds onto this buffer, then CoWing will be needed. APooledRecvBufferAllocator
cycles through preallocated buffers to avoid CoWs during the same read cycle.
internal struct PooledRecvBufferAllocator { | ||
/// | ||
/// This type is useful when allocating new buffers for every IO read is expensive or undesirable. | ||
public struct PooledRecvBufferAllocator { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This'll need a "NIO" prefix now that it's coming into a public namespace, see https://github.com/apple/swift-nio/blob/main/docs/public-api.md#1-no-global-namespace-additions-that-arent-prefixed
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've added the prefix. Great catch.
/// - allocator: `ByteBufferAllocator` used to construct a new buffer if needed | ||
/// - body: Closure where the caller can use the new or existing buffer | ||
/// - Returns: A tuple containing the `ByteBuffer` used and the `Result` yielded by the closure provided. | ||
public mutating func buffer<Result>( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This should be @inlinable
. In turn it will mean a bunch of other functions and properties will need to be as well. For anything that's private
that needs to be @inlinable
/@usableFromInline
they can become internal
and be prefixed with an _
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've changed this and related functions/properties to be @inlinable/@usableFromInline
as you suggested.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One nit but LGTM otherwise
/// Channels can read multiple times per cycle (based on `ChannelOptions.maxMessagesPerRead`), and they reuse | ||
/// the inbound buffer for each read. If a `ChannelHandler` holds onto this buffer, then CoWing will be needed. | ||
/// A `NIOPooledRecvBufferAllocator` cycles through preallocated buffers to avoid CoWs during the same read cycle. | ||
public struct NIOPooledRecvBufferAllocator { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
As a public type we should annotate whether this is Sendable
or not. It looks like it is.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've introduced the change requested.
Moving
PooledRecvBufferAllocator
from NIOPosix to NIOCore, making it part of its public interface.Motivation:
We want to introduce the capability of reusing receiving buffers from a pool into swift-nio-transport-services project. To prevent duplicating this functionality there, this change is making the existing functionality from NIO part of its public API.
Modifications:
Moved the type
PooledRecvBufferAllocator
from NIOPosix module to NIOCore and turned it into a public type.Result:
PooledRecvBufferAllocator
is part of NIOCore public API.