Introduce a backpressure-propagating buffer #451
Merged
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
|
@carllerche's feedback is that the delay-based probing may be unnecessary if inner services are responsible for notifying the task when they need to change (even when they return Ready). |
olix0r
changed the title
The tower buffer implementation uses the underlying channel for backpressure. This admits requests before the inner service becomes ready, and doesn't fully propagate error conditions. This replacement buffer implementation exposes readiness via a `Watch`. This prevents callers from sending requests when the inner service is not ready and allows service errors to be propagated reliably. The buffer's background task drops the inner service as soon as it fails. The effect of this is that the buffer's channel is only used for requests that race to submit requests while the inner service is ready, and so it seems likely that they only need enough slots to satisfy the proxy's expected concurrency (i.e. in terms of cores).
|
I've removed the probing logic from the buffer and moved it into a dedicated middleware -- there's no reason it needs to be coupled to the buffer. |
hawkw
approved these changes
Mar 6, 2020
hawkw
approved these changes
Mar 6, 2020
| .as_mut() | ||
| .expect("Service must not be dropped") | ||
| .call(request); | ||
| let _ = tx.send(Ok(fut)); |
There was a problem hiding this comment.
how often do we think requesters will hang up after requests are dispatched? is this something that we would want to record a TRACE/DEBUG event for?
There was a problem hiding this comment.
This will happen whenever a request is canceled (i.e. the client resets the stream or drops the connection). Could go either way
kleimkuhler
reviewed
Mar 6, 2020
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
Mar 10, 2020
This release builds on changes in the prior release to ensure that balancers process updates eagerly. Cache capacity limitations have been removed; and services now fail eagerly, rather than making all requests wait for the timeout to expire. Also, a bug was fixed in the way the `LINKERD2_PROXY_LOG` env variable is parsed. --- * Introduce a backpressure-propagating buffer (linkerd/linkerd2-proxy#451) * trace: update tracing-subscriber to 0.2.3 (linkerd/linkerd2-proxy#455) * timeout: Introduce FailFast, Idle, and Probe middlewares (linkerd/linkerd2-proxy#452) * cache: Let services self-evict (linkerd/linkerd2-proxy#456)
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
Mar 10, 2020
This release builds on changes in the prior release to ensure that balancers process updates eagerly. Cache capacity limitations have been removed; and services now fail eagerly, rather than making all requests wait for the timeout to expire. Also, a bug was fixed in the way the `LINKERD2_PROXY_LOG` env variable is parsed. --- * Introduce a backpressure-propagating buffer (linkerd/linkerd2-proxy#451) * trace: update tracing-subscriber to 0.2.3 (linkerd/linkerd2-proxy#455) * timeout: Introduce FailFast, Idle, and Probe middlewares (linkerd/linkerd2-proxy#452) * cache: Let services self-evict (linkerd/linkerd2-proxy#456)
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.
The tower buffer implementation uses the underlying channel for
backpressure. This admits requests before the inner service becomes
ready, and doesn't fully propagate error conditions.
This replacement buffer implementation exposes readiness via a
Watch.This prevents callers from sending requests when the inner service is
not ready and allows service errors to be propagated reliably. The
buffer's background task drops the inner service as soon as it fails.
The effect of this is that the buffer's channel is only used for
requests that race to submit requests while the inner service is ready,
and so it seems likely that they only need enough slots to satisfy the
proxy's expected concurrency (i.e. in terms of cores).