concurrency-limit: use tokio-util's PollSemaphore#968
Merged
Conversation
Currently, our concurrency limit middleware uses `tokio::sync::Semaphore` to implement the concurrency limit. Because acquiring permits from the `Semaphore` type is only possible via the `acquire` and `acquire_owned` methods, which are `async fn`s returning unameable futures that are `!Unpin`, we cannot simply poll the semaphore in `poll_ready`. Instead, to allow polling the concurrency limit service to acquire a permit, we box the `acquire_owned` future in `poll_ready`. This means that every time we try to acquire a new concurrency limit permit, we must allocate a new box. The `tokio-util` crate has a [`PollSemaphore`][1] type which allows wrapping a `tokio::sync::Semaphore` to add a `poll_acquire` method to poll the semaphore to acquire a permit. Unlike our implementation, `PollSemaphore` uses the [`tokio_util::sync::ReusableBoxFuture`][2] type for boxing the `acquire_owned` future returned by the `Semaphore`. `ReusableBoxFuture` is a safe abstraction around unsafe code that allows *one* allocation to be reused for storing multiple type-erased `Box::pin`ned futures. This means creating a single allocation for each clone of a `ConcurrencyLimit` service, rather than having each clone of the concurrency limit service allocate once *every* time a new permit is acquired and drop that allocation once the permit is acquired. This should reduce the overhead of the concurrency limit layer a bit. It also makes the code much simpler! :) [1]: https://docs.rs/tokio-util/0.6.5/tokio_util/sync/struct.PollSemaphore.html [2]: https://docs.rs/tokio-util/0.6.5/tokio_util/sync/struct.ReusableBoxFuture.html Signed-off-by: Eliza Weisman <[email protected]>
hawkw
commented
Apr 7, 2021
| fn poll_ready(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> { | ||
| trace!(available = %self.semaphore.available_permits(), "acquiring permit"); | ||
| trace!( | ||
| // available = %self.semaphore.available_permits(), |
Contributor
Author
There was a problem hiding this comment.
logging the number of available permits here depends on upstream PR tokio-rs/tokio#3683
kleimkuhler
approved these changes
Apr 7, 2021
Contributor
kleimkuhler
left a comment
kleimkuhler
left a comment
There was a problem hiding this comment.
Nice! Thanks for the helpful description.
olix0r
approved these changes
Apr 9, 2021
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
Apr 21, 2021
This release primarily improves protocol detection error messages in the admin server so that logs clearly indicate when the client expected a different TLS server identity than that of the running proxy. A number of internal improvements have been made, especially eliminating some potential runtime panics detected by oss-fuzz. It is not expected that these panics could be triggered in typical cluster configurations. --- * discover: replace `linkerd-channel` with `tokio-util` `PollSender` (linkerd/linkerd2-proxy#969) * replace `linkerd-channel` with `tokio-stream` (linkerd/linkerd2-proxy#970) * concurrency-limit: use `tokio-util`'s `PollSemaphore` (linkerd/linkerd2-proxy#968) * http: Do not fail fuzz tests when all IO is not read (linkerd/linkerd2-proxy#973) * transport: Fix orig-dst compilation on non-Linux targets (linkerd/linkerd2-proxy#974) * Update trust-dns to fix possible panic (linkerd/linkerd2-proxy#975) * addr: fix `to_http_authority` panic with IPv6 (linkerd/linkerd2-proxy#976) * outbound: skip logical stacks when no profile is discovered (linkerd/linkerd2-proxy#963) * split: change traffic splits to require a profile (linkerd/linkerd2-proxy#964) * inbound: only build profile route stacks when a profile is resolved (linkerd/linkerd2-proxy#966) * profiles: make receiver param in `route_request` non-optional (linkerd/linkerd2-proxy#967) * outbound: move target types into stack modules (linkerd/linkerd2-proxy#971) * outbound: only build logical stacks for profiles with logical addrs (linkerd/linkerd2-proxy#972) * app: inbound: add fuzzer (linkerd/linkerd2-proxy#977) * admin: Fail connections when HTTP detection fails (linkerd/linkerd2-proxy#979) * reduce error boilerplate with `thiserror` (linkerd/linkerd2-proxy#980) * app: inbound: fuzzer: generalise fuzzers and solve coverage build (linkerd/linkerd2-proxy#978) * admin: Assume meshed connections are HTTP/2 (linkerd/linkerd2-proxy#982) * chore: Fix deprecations on nightly (linkerd/linkerd2-proxy#983)
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
May 12, 2021
This release simplifies internals so that endpoint-forwarding logic is completely distinct from handling of load balanced services. The ingress-mode outbound proxy has been simplified to *require* the `l5d-dst-override` header and to fail non-HTTP communication. This ensures that the ingress-mode proxy does not unexpectedly revert to insecure communication. Finally, a regression was recently introduced that caused all proxy logs to be output with ANSI control characters. Logs are now output in plaintext by default --- * discover: replace `linkerd-channel` with `tokio-util` `PollSender` (linkerd/linkerd2-proxy#969) * replace `linkerd-channel` with `tokio-stream` (linkerd/linkerd2-proxy#970) * concurrency-limit: use `tokio-util`'s `PollSemaphore` (linkerd/linkerd2-proxy#968) * http: Do not fail fuzz tests when all IO is not read (linkerd/linkerd2-proxy#973) * transport: Fix orig-dst compilation on non-Linux targets (linkerd/linkerd2-proxy#974) * Update trust-dns to fix possible panic (linkerd/linkerd2-proxy#975) * addr: fix `to_http_authority` panic with IPv6 (linkerd/linkerd2-proxy#976) * outbound: skip logical stacks when no profile is discovered (linkerd/linkerd2-proxy#963) * split: change traffic splits to require a profile (linkerd/linkerd2-proxy#964) * inbound: only build profile route stacks when a profile is resolved (linkerd/linkerd2-proxy#966) * profiles: make receiver param in `route_request` non-optional (linkerd/linkerd2-proxy#967) * outbound: move target types into stack modules (linkerd/linkerd2-proxy#971) * outbound: only build logical stacks for profiles with logical addrs (linkerd/linkerd2-proxy#972) * app: inbound: add fuzzer (linkerd/linkerd2-proxy#977) * admin: Fail connections when HTTP detection fails (linkerd/linkerd2-proxy#979) * reduce error boilerplate with `thiserror` (linkerd/linkerd2-proxy#980) * app: inbound: fuzzer: generalise fuzzers and solve coverage build (linkerd/linkerd2-proxy#978) * admin: Assume meshed connections are HTTP/2 (linkerd/linkerd2-proxy#982) * chore: Fix deprecations on nightly (linkerd/linkerd2-proxy#983) * fuzz: Add logging to fuzz targets (linkerd/linkerd2-proxy#985) * fuzz: don't panic when the proxy closes the conn (linkerd/linkerd2-proxy#986) * Commit lock files for fuzzers (linkerd/linkerd2-proxy#984) * fuzz: allow client requests to fail (linkerd/linkerd2-proxy#989) * tower: update dependency to 0.4.7 (linkerd/linkerd2-proxy#990) * outbound: Make the Endpoint::logical_addr field optional (linkerd/linkerd2-proxy#991) * trace: explicitly disable ANSI terminal colors (linkerd/linkerd2-proxy#994) * ingress: Require the l5d-dst-override header (linkerd/linkerd2-proxy#992) * outbound: Do not support TCP-forwarding in ingress-mode (linkerd/linkerd2-proxy#995) * Decouple tcp forward stack from Endpoint target (linkerd/linkerd2-proxy#996) * Pickup linkerd-await wrapper in docker build (linkerd/linkerd2-proxy#999) * docs: Add fuzzing report (linkerd/linkerd2-proxy#1000) * tests: Use io::Error in mocked connector (linkerd/linkerd2-proxy#1001) * outbound: Decouple endpoint & logical stacks (linkerd/linkerd2-proxy#1002)
olix0r
added a commit
to linkerd/linkerd2
that referenced
this pull request
May 18, 2021
This release simplifies internals so that endpoint-forwarding logic is completely distinct from handling of load balanced services. The ingress-mode outbound proxy has been simplified to *require* the `l5d-dst-override` header and to fail non-HTTP communication. This ensures that the ingress-mode proxy does not unexpectedly revert to insecure communication. Finally, a regression was recently introduced that caused all proxy logs to be output with ANSI control characters. Logs are now output in plaintext by default --- * discover: replace `linkerd-channel` with `tokio-util` `PollSender` (linkerd/linkerd2-proxy#969) * replace `linkerd-channel` with `tokio-stream` (linkerd/linkerd2-proxy#970) * concurrency-limit: use `tokio-util`'s `PollSemaphore` (linkerd/linkerd2-proxy#968) * http: Do not fail fuzz tests when all IO is not read (linkerd/linkerd2-proxy#973) * transport: Fix orig-dst compilation on non-Linux targets (linkerd/linkerd2-proxy#974) * Update trust-dns to fix possible panic (linkerd/linkerd2-proxy#975) * addr: fix `to_http_authority` panic with IPv6 (linkerd/linkerd2-proxy#976) * outbound: skip logical stacks when no profile is discovered (linkerd/linkerd2-proxy#963) * split: change traffic splits to require a profile (linkerd/linkerd2-proxy#964) * inbound: only build profile route stacks when a profile is resolved (linkerd/linkerd2-proxy#966) * profiles: make receiver param in `route_request` non-optional (linkerd/linkerd2-proxy#967) * outbound: move target types into stack modules (linkerd/linkerd2-proxy#971) * outbound: only build logical stacks for profiles with logical addrs (linkerd/linkerd2-proxy#972) * app: inbound: add fuzzer (linkerd/linkerd2-proxy#977) * admin: Fail connections when HTTP detection fails (linkerd/linkerd2-proxy#979) * reduce error boilerplate with `thiserror` (linkerd/linkerd2-proxy#980) * app: inbound: fuzzer: generalise fuzzers and solve coverage build (linkerd/linkerd2-proxy#978) * admin: Assume meshed connections are HTTP/2 (linkerd/linkerd2-proxy#982) * chore: Fix deprecations on nightly (linkerd/linkerd2-proxy#983) * fuzz: Add logging to fuzz targets (linkerd/linkerd2-proxy#985) * fuzz: don't panic when the proxy closes the conn (linkerd/linkerd2-proxy#986) * Commit lock files for fuzzers (linkerd/linkerd2-proxy#984) * fuzz: allow client requests to fail (linkerd/linkerd2-proxy#989) * tower: update dependency to 0.4.7 (linkerd/linkerd2-proxy#990) * outbound: Make the Endpoint::logical_addr field optional (linkerd/linkerd2-proxy#991) * trace: explicitly disable ANSI terminal colors (linkerd/linkerd2-proxy#994) * ingress: Require the l5d-dst-override header (linkerd/linkerd2-proxy#992) * outbound: Do not support TCP-forwarding in ingress-mode (linkerd/linkerd2-proxy#995) * Decouple tcp forward stack from Endpoint target (linkerd/linkerd2-proxy#996) * Pickup linkerd-await wrapper in docker build (linkerd/linkerd2-proxy#999) * docs: Add fuzzing report (linkerd/linkerd2-proxy#1000) * tests: Use io::Error in mocked connector (linkerd/linkerd2-proxy#1001) * outbound: Decouple endpoint & logical stacks (linkerd/linkerd2-proxy#1002)
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Currently, our concurrency limit middleware uses
tokio::sync::Semaphoreto implement the concurrency limit. Becauseacquiring permits from the
Semaphoretype is only possible via theacquireandacquire_ownedmethods, which areasync fns returningunameable futures that are
!Unpin, we cannot simply poll the semaphorein
poll_ready. Instead, to allow polling the concurrency limit serviceto acquire a permit, we box the
acquire_ownedfuture inpoll_ready.This means that every time we try to acquire a new concurrency limit
permit, we must allocate a new box.
The
tokio-utilcrate has aPollSemaphoretype which allowswrapping a
tokio::sync::Semaphoreto add apoll_acquiremethod topoll the semaphore to acquire a permit. Unlike our implementation,
PollSemaphoreuses thetokio_util::sync::ReusableBoxFuturetypefor boxing the
acquire_ownedfuture returned by theSemaphore.ReusableBoxFutureis a safe abstraction around unsafe code that allowsone allocation to be reused for storing multiple type-erased
Box::pinned futures. This means creating a single allocation for eachclone of a
ConcurrencyLimitservice, rather than having each clone ofthe concurrency limit service allocate once every time a new permit is
acquired and drop that allocation once the permit is acquired.
This should reduce the overhead of the concurrency limit layer a bit. It
also makes the code much simpler! :)
Signed-off-by: Eliza Weisman [email protected]