-
Notifications
You must be signed in to change notification settings - Fork 71
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
Simplify cancellation logic #396
Merged
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
talex5
force-pushed
the
simplify-cancel
branch
2 times, most recently
from
December 21, 2022 12:06
f3faa58
to
1c1fff4
Compare
talex5
force-pushed
the
simplify-cancel
branch
8 times, most recently
from
December 21, 2022 14:14
1fcacbd
to
e2ac0c8
Compare
Before, the fiber context held an atomic recording whether the operation had succeeded or been cancelled, to ensure that only one of these things happened. However, this doesn't need to be atomic for single-threaded operations, and even for parallel operations, we might perfer to keep the atomic state elsewhere. Now, operations just set a regular function and it's up to the operation to ensure it's atomic if needed. This is needed to implement lock-free waiters. Otherwise we have this problem: 1. Domain 1 starts waiting for something from another domain. 2. Domain 2 prepares to provide a value to it. 3. Domain 1 cancels the operation and continues. 4. Domain 1 starts a new operation with a new cancel function. 5. Domain 2 sees the operation isn't cancelled and sends the value. This isn't a problem when using a lock because both domains take the lock. Currently, the waiters module still uses a lock (but now provides its own mutex), and so works the same as before. Some other related improvements: - In eio_linux, we no longer wait for the cancel response before continuing. Before, cancelling an io_uring request would wait for the result of the cancellation before the cancel operation returned. But this isn't very useful, makes cancelling lots of things very slow since you can only do one at a time, and doesn't work with the new cancellation system where you can't switch fibers while cancelling. - The "Cancelling multiple jobs" test has been removed. It was testing what happens when a cancel function switches fibers, but this is no longer allowed. - In eio_luv, we only report cancellation if the operation didn't succeed. This changes the signature of `Low_level.await_with_cancel` slightly, since it now needs to know whether the operation was successful or not. - Moved the cancellation logic into the domain manager implementation. This was the only place where a fiber could have a cancel function while not suspended.
talex5
force-pushed
the
simplify-cancel
branch
from
December 21, 2022 14:17
e2ac0c8
to
036d83d
Compare
talex5
added a commit
to talex5/opam-repository
that referenced
this pull request
Feb 1, 2023
CHANGES: New features: - Add `Eio.Net.run_server` (@bikallem @talex5 ocaml-multicore/eio#408). Runs an accept loop in one or more domains, with cancellation and graceful shutdown, and an optional maximum number of concurrent connections. - Add `Buf_read.BE` and `LE` parsers (@Cjen1 ocaml-multicore/eio#399). Parse numbers in various binary formats. - Add `Eio.Buf_read.uint8` (@talex5 ocaml-multicore/eio#418). Performance: - Make `Eio.Condition` lock-free (@talex5 ocaml-multicore/eio#397 ocaml-multicore/eio#381). In addition to being faster, this allows using conditions in signal handlers. - Make `Eio.Semaphore` lock-free (@talex5 @polytypic ocaml-multicore/eio#398). - Make `Eio.Stream` lock-free when the capacity is zero (@talex5 ocaml-multicore/eio#413 ocaml-multicore/eio#411). - Make `Eio.Promise` lock-free (@talex5 ocaml-multicore/eio#401). Bug fixes: - eio_linux: call `Uring.submit` as needed (@talex5 @bikallem ocaml-multicore/eio#428). Previously, we could fail to submit a job promptly because the SQE queue was full. - Fix luv signals (@haesbaert ocaml-multicore/eio#412). `libuv` automatically retries polling if it gets `EINTR`, without giving OCaml signal handlers a chance to run. - eio_luv: fix some resource leaks (@talex5 @patricoferris ocaml-multicore/eio#421). - eio_luv: fix "unavailable signal" error on Windows (@talex5 ocaml-multicore/eio#420, reported by @nojb). - Fix `Buf_write.BE.uint48` and `LE.uint48` (@adatario ocaml-multicore/eio#418). Documentation: - Add example programs (@talex5 ocaml-multicore/eio#389). - Update network examples to use `run_server` (@talex5 ocaml-multicore/eio#417). - Add a warning to the tutorial about `Fiber.first` (@talex5 ocaml-multicore/eio#394). - Clarify the epoch used for `Eio.Time.now` (@bikallem ocaml-multicore/eio#395). - Describe `secure_random` as an infinite source (@patricoferris ocaml-multicore/eio#426). - Update README for OCaml 5 release (@talex5 ocaml-multicore/eio#384 ocaml-multicore/eio#391 ocaml-multicore/eio#393). Other changes: - Delay setting `SIGPIPE` handler until the `run` function is called (@talex5 ocaml-multicore/eio#420). - Remove debug-level logging (@talex5 ocaml-multicore/eio#403). - eio-luv: improve `process.md` test (@smondet ocaml-multicore/eio#414). - Update to Dune 3 (@talex5 ocaml-multicore/eio#410). - Remove test dependency on Astring (@talex5 ocaml-multicore/eio#402 ocaml-multicore/eio#404). - Simplify cancellation logic (@talex5 ocaml-multicore/eio#396). - time: `Mtime.Spand.to_s` has been deprecated in mtime 2.0.0 (@bikallem ocaml-multicore/eio#385).
talex5
added a commit
to talex5/opam-repository
that referenced
this pull request
Feb 1, 2023
CHANGES: New features: - Add `Eio.Net.run_server` (@bikallem @talex5 ocaml-multicore/eio#408). Runs an accept loop in one or more domains, with cancellation and graceful shutdown, and an optional maximum number of concurrent connections. - Add `Buf_read.BE` and `LE` parsers (@Cjen1 ocaml-multicore/eio#399). Parse numbers in various binary formats. - Add `Eio.Buf_read.uint8` (@talex5 ocaml-multicore/eio#418). Performance: - Make `Eio.Condition` lock-free (@talex5 ocaml-multicore/eio#397 ocaml-multicore/eio#381). In addition to being faster, this allows using conditions in signal handlers. - Make `Eio.Semaphore` lock-free (@talex5 @polytypic ocaml-multicore/eio#398). - Make `Eio.Stream` lock-free when the capacity is zero (@talex5 ocaml-multicore/eio#413 ocaml-multicore/eio#411). - Make `Eio.Promise` lock-free (@talex5 ocaml-multicore/eio#401). Bug fixes: - eio_linux: call `Uring.submit` as needed (@talex5 @bikallem ocaml-multicore/eio#428). Previously, we could fail to submit a job promptly because the SQE queue was full. - Fix luv signals (@haesbaert ocaml-multicore/eio#412). `libuv` automatically retries polling if it gets `EINTR`, without giving OCaml signal handlers a chance to run. - eio_luv: fix some resource leaks (@talex5 @patricoferris ocaml-multicore/eio#421). - eio_luv: fix "unavailable signal" error on Windows (@talex5 ocaml-multicore/eio#420, reported by @nojb). - Fix `Buf_write.BE.uint48` and `LE.uint48` (@adatario ocaml-multicore/eio#418). Documentation: - Add example programs (@talex5 ocaml-multicore/eio#389). - Update network examples to use `run_server` (@talex5 ocaml-multicore/eio#417). - Add a warning to the tutorial about `Fiber.first` (@talex5 ocaml-multicore/eio#394). - Clarify the epoch used for `Eio.Time.now` (@bikallem ocaml-multicore/eio#395). - Describe `secure_random` as an infinite source (@patricoferris ocaml-multicore/eio#426). - Update README for OCaml 5 release (@talex5 ocaml-multicore/eio#384 ocaml-multicore/eio#391 ocaml-multicore/eio#393). Other changes: - Delay setting `SIGPIPE` handler until the `run` function is called (@talex5 ocaml-multicore/eio#420). - Remove debug-level logging (@talex5 ocaml-multicore/eio#403). - eio-luv: improve `process.md` test (@smondet ocaml-multicore/eio#414). - Update to Dune 3 (@talex5 ocaml-multicore/eio#410). - Remove test dependency on Astring (@talex5 ocaml-multicore/eio#402 ocaml-multicore/eio#404). - Simplify cancellation logic (@talex5 ocaml-multicore/eio#396). - time: `Mtime.Spand.to_s` has been deprecated in mtime 2.0.0 (@bikallem ocaml-multicore/eio#385).
talex5
added a commit
to talex5/opam-repository
that referenced
this pull request
Feb 1, 2023
CHANGES: New features: - Add `Eio.Net.run_server` (@bikallem @talex5 ocaml-multicore/eio#408). Runs an accept loop in one or more domains, with cancellation and graceful shutdown, and an optional maximum number of concurrent connections. - Add `Buf_read.BE` and `LE` parsers (@Cjen1 ocaml-multicore/eio#399). Parse numbers in various binary formats. - Add `Eio.Buf_read.uint8` (@talex5 ocaml-multicore/eio#418). Performance: - Make `Eio.Condition` lock-free (@talex5 ocaml-multicore/eio#397 ocaml-multicore/eio#381). In addition to being faster, this allows using conditions in signal handlers. - Make `Eio.Semaphore` lock-free (@talex5 @polytypic ocaml-multicore/eio#398). - Make `Eio.Stream` lock-free when the capacity is zero (@talex5 ocaml-multicore/eio#413 ocaml-multicore/eio#411). - Make `Eio.Promise` lock-free (@talex5 ocaml-multicore/eio#401). Bug fixes: - eio_linux: call `Uring.submit` as needed (@talex5 @bikallem ocaml-multicore/eio#428). Previously, we could fail to submit a job promptly because the SQE queue was full. - Fix luv signals (@haesbaert ocaml-multicore/eio#412). `libuv` automatically retries polling if it gets `EINTR`, without giving OCaml signal handlers a chance to run. - eio_luv: fix some resource leaks (@talex5 @patricoferris ocaml-multicore/eio#421). - eio_luv: fix "unavailable signal" error on Windows (@talex5 ocaml-multicore/eio#420, reported by @nojb). - Fix `Buf_write.BE.uint48` and `LE.uint48` (@adatario ocaml-multicore/eio#418). Documentation: - Add example programs (@talex5 ocaml-multicore/eio#389). - Update network examples to use `run_server` (@talex5 ocaml-multicore/eio#417). - Add a warning to the tutorial about `Fiber.first` (@talex5 ocaml-multicore/eio#394). - Clarify the epoch used for `Eio.Time.now` (@bikallem ocaml-multicore/eio#395). - Describe `secure_random` as an infinite source (@patricoferris ocaml-multicore/eio#426). - Update README for OCaml 5 release (@talex5 ocaml-multicore/eio#384 ocaml-multicore/eio#391 ocaml-multicore/eio#393). Other changes: - Delay setting `SIGPIPE` handler until the `run` function is called (@talex5 ocaml-multicore/eio#420). - Remove debug-level logging (@talex5 ocaml-multicore/eio#403). - eio-luv: improve `process.md` test (@smondet ocaml-multicore/eio#414). - Update to Dune 3 (@talex5 ocaml-multicore/eio#410). - Remove test dependency on Astring (@talex5 ocaml-multicore/eio#402 ocaml-multicore/eio#404). - Simplify cancellation logic (@talex5 ocaml-multicore/eio#396). - time: `Mtime.Spand.to_s` has been deprecated in mtime 2.0.0 (@bikallem ocaml-multicore/eio#385).
talex5
added a commit
to talex5/opam-repository
that referenced
this pull request
Feb 1, 2023
CHANGES: New features: - Add `Eio.Net.run_server` (@bikallem @talex5 ocaml-multicore/eio#408). Runs an accept loop in one or more domains, with cancellation and graceful shutdown, and an optional maximum number of concurrent connections. - Add `Buf_read.BE` and `LE` parsers (@Cjen1 ocaml-multicore/eio#399). Parse numbers in various binary formats. - Add `Eio.Buf_read.uint8` (@talex5 ocaml-multicore/eio#418). Performance: - Make `Eio.Condition` lock-free (@talex5 ocaml-multicore/eio#397 ocaml-multicore/eio#381). In addition to being faster, this allows using conditions in signal handlers. - Make `Eio.Semaphore` lock-free (@talex5 @polytypic ocaml-multicore/eio#398). - Make `Eio.Stream` lock-free when the capacity is zero (@talex5 ocaml-multicore/eio#413 ocaml-multicore/eio#411). - Make `Eio.Promise` lock-free (@talex5 ocaml-multicore/eio#401). Bug fixes: - eio_linux: call `Uring.submit` as needed (@talex5 @bikallem ocaml-multicore/eio#428). Previously, we could fail to submit a job promptly because the SQE queue was full. - Fix luv signals (@haesbaert ocaml-multicore/eio#412). `libuv` automatically retries polling if it gets `EINTR`, without giving OCaml signal handlers a chance to run. - eio_luv: fix some resource leaks (@talex5 @patricoferris ocaml-multicore/eio#421). - eio_luv: fix "unavailable signal" error on Windows (@talex5 ocaml-multicore/eio#420, reported by @nojb). - Fix `Buf_write.BE.uint48` and `LE.uint48` (@adatario ocaml-multicore/eio#418). Documentation: - Add example programs (@talex5 ocaml-multicore/eio#389). - Update network examples to use `run_server` (@talex5 ocaml-multicore/eio#417). - Add a warning to the tutorial about `Fiber.first` (@talex5 ocaml-multicore/eio#394). - Clarify the epoch used for `Eio.Time.now` (@bikallem ocaml-multicore/eio#395). - Describe `secure_random` as an infinite source (@patricoferris ocaml-multicore/eio#426). - Update README for OCaml 5 release (@talex5 ocaml-multicore/eio#384 ocaml-multicore/eio#391 ocaml-multicore/eio#393). Other changes: - Delay setting `SIGPIPE` handler until the `run` function is called (@talex5 ocaml-multicore/eio#420). - Remove debug-level logging (@talex5 ocaml-multicore/eio#403). - eio-luv: improve `process.md` test (@smondet ocaml-multicore/eio#414). - Update to Dune 3 (@talex5 ocaml-multicore/eio#410). - Remove test dependency on Astring (@talex5 ocaml-multicore/eio#402 ocaml-multicore/eio#404). - Simplify cancellation logic (@talex5 ocaml-multicore/eio#396). - time: `Mtime.Spand.to_s` has been deprecated in mtime 2.0.0 (@bikallem ocaml-multicore/eio#385).
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.
Before, the fiber context held an atomic recording whether the operation had succeeded or been cancelled, to ensure that only one of these things happened. However, this doesn't need to be atomic for single-threaded operations, and even for parallel operations, we might perfer to keep the atomic state elsewhere.
Now, operations just set a regular function and it's up to the operation to ensure it's atomic if needed.
This is needed to implement lock-free waiters. Otherwise we have this problem:
This isn't a problem when using a lock because both domains take the lock. Currently, the waiters module still uses a lock and it now provides its own mutex that works the same as before.
Some other related improvements:
In eio_linux, we no longer wait for the cancel response before continuing.
Before, cancelling an io_uring request would wait for the result of the cancellation before the cancel operation returned. But this isn't very useful, makes cancelling lots of things very slow since you can only do one at a time, and doesn't work with the new cancellation system where you can't switch fibers while cancelling.
The "Cancelling multiple jobs" test has been removed. It was testing what happens when a cancel function switches fibers, but this is no longer allowed.
In eio_luv, we only report cancellation if the operation didn't succeed.
This changes the signature of
Low_level.await_with_cancel
slightly, since it now needs to know whether the operation was successful or not.Fixed a bug in the mock domain manager used for the README examples. I think this was wrong before too, but it was only detected with the new system.
TODO: