[PoC] io: add iouring support#22392

Closed

soulxu wants to merge 34 commits intoenvoyproxy:mainfrom

soulxu:new_iouring_version_2

Member

soulxu commented Jul 26, 2022 •

edited

Loading

Commit Message: io: add iouring support
Additional Description:
Risk Level: low, it should be enabled by default.
Testing: unit tests and integration tests are expected
Docs Changes: TODO
Release Notes: TODO
Platform Specific Features: Only linux platform support iouring
Runtime guard: (TODO) A runtime flag should be added for this feature.

Signed-off-by: He Jie Xu hejie.xu@intel.com
Signed-off-by: Zhi Hao Xie zhihao.xie@intel.com
Signed-off-by: Dmitry Rozhkov dmitry.rozhkov@intel.com

repokitteh-read-only bot commented Jul 26, 2022

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #22392 was opened by soulxu.

see: more, trace.

Member Author

soulxu commented Jul 26, 2022 •

edited

Loading

This continues the work of #19082

It is still in the early stage, and a lot of things need to be figured out.

zhxie reviewed

View reviewed changes

Contributor

zhxie left a comment

Leave some comments (undecided or dangerous implementation) in advance.

source/common/network/io_uring_socket_handle_impl.cc Outdated

+                      delete[] reinterpret_cast<const uint8_t*>(data);
+                      delete this_fragment;
+                    });
+                buffer.addBufferFragment(*fragment);

Contributor

zhxie Jul 28, 2022

We may not able to do this in readv and sends and will end up in memcpy.

source/common/network/io_uring_socket_handle_impl.cc

+              Api::IoCallUint64Result IoUringSocketHandleImpl::recv(void* /*buffer*/, size_t /*length*/,
+                                                                    int /*flags*/) {
+                PANIC("not implemented");

Contributor

zhxie Jul 28, 2022

I do not see and equivalent operations of flags in io_uring like MSG_PEEK. We may implement on our own.

source/common/network/io_uring_socket_handle_impl.cc Outdated

+                }
+                if (isLeader()) {
+                  io_uring_factory_.getOrCreate().unregisterEventfd();
+                  file_event_adapter_.reset();

Contributor

zhxie Jul 28, 2022

If the adapter deconstructs since the leading handle closes, what will happen to the remaining existing handles?

github-actions bot commented Aug 27, 2022

This pull request has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in 7 days if no further activity occurs. Please feel free to give a status update now, ping for review, or re-open when it's ready. Thank you for your contributions!

github-actions bot added the stale label

Member Author

soulxu commented Aug 27, 2022

still working on it

github-actions bot removed the stale label

soulxu added 5 commits

September 19, 2022 05:58


          Add dimitry's implement and move to the core

5b57402

Signed-off-by: Dmitry Rozhkov <dmitry.rozhkov@intel.com>

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          pass the io uring factory

13760ab

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          not set non-blocking flag for iouring

302a7f0

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          test socket interface

466f9fe

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          Not using IoUringSocketHandle when the current thread isn't registered

435d9ef

Signed-off-by: He Jie Xu <hejie.xu@intel.com>

soulxu force-pushed the new_iouring_version_2 branch from 779bcf6 to 435d9ef Compare

September 19, 2022 09:08

zhxie and others added 17 commits

September 20, 2022 12:52


          network: handle io_uring factory safely (#4)

85effee

network: handle io_uring factory safely

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          iouring: change the iouring:getOrCreate to just get method

a5edb42

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          SocketInterface: only create IoUringHandle when the iouring factory

c7aa1f0

created

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          Only register Read event for eventfd

2ccb461

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          network: write sequentially (#7)

6c94104

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          initialize bootstrap ext before add listener

38e970b

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          skip eventfd register when it alreayd registered

0eb8777

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          network: force starting read after write (#9)

a4fcd42

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: implement readv and writev (#8)

7e08174

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          Ensure the write request submitted

0eb7119

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          SKip invoking callback if the fd already closed when uring ready event

2df8192

back

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          network: fix incorrect ptr assignment (#11)

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          Using return zero instead of raise close event for halfclose

43b6048

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          Return error when result is negative

4e84781

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          hacky fix for invalid fd for a write completion

a37f079

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          add simple shutdown implement

4d32b11

Signed-off-by: He Jie Xu <hejie.xu@intel.com>


          network: cancel submitted read before close (#13)

79cc8fb

* network: cancel before close

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

zhxie added 7 commits

October 27, 2022 14:00


          network: delay write result (#14)

9ceda00

* network: sync write

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: fix (#15)

75bdfd6

* network: fix writev

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: escape cancel first

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: add read request when read or connect (#16)

395fa4b

* network: delay read request

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: add read request after connect

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: fix write result handling (#17)

b39c113

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: support max read length (#18)

eda48cd

* network: escape 0 byte to read in readv

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: error should be constructed with errno

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: escape errors in readv

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: unify read and readv

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: support max read length

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* nit: change member names

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: fix incorrect merge

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: fix wrong write error

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: minor improvement

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: activate read event after enabling (#19)

cc9fef1

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: ensure connect request will get submitted (#20)

d919c92

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

soulxu mentioned this pull request

[PoC] io: support iouring - version 2 #24305

Closed

zhxie added 5 commits

December 2, 2022 09:59


          [backport] network: stop adding read request on read (#23)

2b5939e

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          [backport] network: drain after writing (#24)

1cc8331

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          [backport] network: limit write slice size (#25)

d2d3959

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          [backport] network: move instead of copy (#26)

2f925e8

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>


          network: separate adapter from IO handle (#22)

e677b6b

* io: add event for io_uring impl

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: migrate

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: fix early accept

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* io: fix empty callback handling

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: fix accept handling

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: escape completion on closed fds

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* deps: no need to upgrade liburing

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: fix accept preparation

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: stop adding read request on read

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* io: stop erasing callback on cancellation

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: delete canceled request

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: avoid migrating between io_uring

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: drain after writing

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: limit write slice size

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* network: minor fix

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: fix connect result handling

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] network: move instead of copy

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* io: delay submit

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* nit

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

* [backport] delay read callback

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Signed-off-by: Xie Zhihao <zhihao.xie@intel.com>

Member Author

soulxu commented Dec 13, 2022

Please reference the second iteration #24305

soulxu closed this

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet