feat(core): amortize many ready messages into fewer, larger buffers #1423
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
Introduce an EncodedBytes combinator which encodes multiple ready messages - up to a yield threshold - before yielding the next bytes buffer. Or, if the message stream polls to pending, yield the available bytes immediately. These ammortized buffers exhibit far better throughput when streaming a high rate of small messages, because hyper and h2 avoid copying the yielded buffer and dispatch each as a separate, non-vectorized tcp send.
jgraettinger
force-pushed
the
johnny/read-chunks
branch
from
bf55a49
to
b4f6445
Compare
LucioFranco
requested changes
Aug 25, 2023
LucioFranco
approved these changes
Sep 1, 2023
jfoster-twilio
added a commit
to jfoster-twilio/tonic
that referenced
this pull request
Jun 25, 2024
hyperium#1423 introduced logic to buffer multiple ready messages in order to amortize the cost of sends to the underlying transport. This also introduced a change in behavior for tonic in the following scenario: A stream of ready messages less than the yield threshold is trailed by a status. Previously the ready messages would all have been yielded from the stream and sent, followed by the status. After the change was introduced the status is yielded from the stream and sent but the accumulated ready messages in the buffer are never sent out. This change adjusts the logic to restore the previous behavior while still retaining the amoritization benefits. Namely it flushes the accumulated ready messages prior to yielding the status ensuring they are sent out from the stream in the order they are read.
github-merge-queue bot
pushed a commit
that referenced
this pull request
Jun 25, 2024
) #1423 introduced logic to buffer multiple ready messages in order to amortize the cost of sends to the underlying transport. This also introduced a change in behavior for tonic in the following scenario: A stream of ready messages less than the yield threshold is trailed by a status. Previously the ready messages would all have been yielded from the stream and sent, followed by the status. After the change was introduced the status is yielded from the stream and sent but the accumulated ready messages in the buffer are never sent out. This change adjusts the logic to restore the previous behavior while still retaining the amoritization benefits. Namely it flushes the accumulated ready messages prior to yielding the status ensuring they are sent out from the stream in the order they are read.
Merged
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.
Introduce an EncodedBytes combinator which encodes multiple ready messages - up to a yield threshold - before yielding the next bytes buffer.
Or, if the message stream polls to pending, yield the available bytes immediately.
These ammortized buffers exhibit far better throughput when streaming a high rate of small messages, because hyper and h2 avoid copying the yielded buffer and dispatch each as a separate, non-vectorized tcp send.
Motivation
We have observed that tonic exhibits poor CPU performance when sending a high data-rate of small messages. Each message is dispatched as a single
tcp_sendmsg, and each incurs overhead for routing and other kernel activities in the network path. It's incumbent to increase the size of network sends without adding latency.Solution
Introduce a combinator which repeatedly polls its delegate stream for ready messages, extending a current buffer (up to a yield threshold). This approach avoids any new allocation and adds just one "magic" constant: a yield threshold after which the combinator will immediately yield the next chunk of bytes.