[Core] Use fd instead of handle for windows log redirection by jjyao · Pull Request #53852 · ray-project/ray

jjyao · 2025-06-16T15:56:18Z

Why are these changes needed?

Windows has two sets of APIs for working with files. One is the WinAPI which works with HANDLE. The other is C API which works with fd (just like Unix). Currently in master we are mix using both and seems have caused issues like #52739. In this PR, we use the C API instead which is more consistent with the Unix code path.

ray start previously failed on my windows machine now passes with this PR.

Related issue number

Closes #52739

Checks

I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
I've run scripts/format.sh to lint the changes in this PR.
I've included any doc changes needed for https://docs.ray.io/en/master/.
- I've added any new APIs to the API Reference. For example, if I added a
  method in Tune, I've added it in doc/source/tune/api/ under the
  corresponding .rst file.
I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
Testing Strategy
- Unit tests
- Release tests
- This PR is not tested :(

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

jjyao · 2025-06-17T04:36:24Z

src/ray/util/compat.cc

+  if (handle == INVALID_HANDLE_VALUE) {
+    return Status::IOError("") << "Fails to get file handle for flushing";
+  }
+  if (!FlushFileBuffers(handle)) {


For flush, there is no corresponding C API so we need to covert fd back to HANDLE and use the WinAPI

will mixing the APIs in this way cause issues like the one this PR intends to fix?

From the testing, it didn't cause issues.

I actually didn't fully understand the root cause. I initially started the PR just trying to simplify things by uniformly using fd and then realized that it fixed the issue.

One thing I noticed is that converting from HANDLE to fd transfer the ownership. But getting a HANDLE back from fd (_get_osfhandle) should be safe as long as we don't close the handle directly. All these are based on my very limited Windows knowledge.

The _open_osfhandle call transfers ownership of the Win32 file handle to the file descriptor. To close a file opened by using _open_osfhandle, call [_close](https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/close?view=msvc-170). The underlying OS file handle is also closed by a call to _close. Don't call the Win32 function CloseHandle on the original handle. If the file descriptor is owned by a FILE * stream, then a call to [fclose](https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/fclose-fcloseall?view=msvc-170) closes both the file descriptor and the underlying handle. In this case, don't call _close on the file descriptor or CloseHandle on the original handle.

I don't pretend to understand this and the PR is merged, but could you please leave a comment in compat.h that describes your current understanding and can aid in future debugging? This might break again in the future.

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

edoakes · 2025-06-17T13:27:20Z

Have you verified that this fixes the reported issue? If so, can you detail how you tested it?

czgdp1807 · 2025-06-17T17:52:08Z

I tried it and it works,

(ray-dev) PS C:\Users\czgdp1807\Anyscale\ray\python> ray start --head
Usage stats collection is enabled. To disable this, add `--disable-usage-stats` to the command that starts the cluster, or run the following command: `ray disable-usage-stats` before starting the cluster. See https://docs.ray.io/en/master/cluster/usage-stats.html for more details.

Local node IP: 127.0.0.1

--------------------
Ray runtime started.
--------------------

Next steps

  To connect to this Ray cluster:
    import ray
    ray.init()

  To terminate the Ray runtime, run
    ray stop

  To view the status of the cluster, use
    ray status
(ray-dev) PS C:\Users\czgdp1807\Anyscale\ray\python> ray status
======== Autoscaler status: 2025-06-17 17:50:13.618875 ========
Node status
---------------------------------------------------------------
Active:
 1 node_1f5d161361f94b4077b0d9ee34c485990b85c77bd55bd6db82d8766a
Pending:
 (no pending nodes)
Recent failures:
 (no failures)

Resources
---------------------------------------------------------------
Total Usage:
 0.0/4.0 CPU
 0B/7.21GiB memory
 0B/3.09GiB object_store_memory

Total Constraints:
 (no request_resources() constraints)
Total Demands:
 (no resource demands)

czgdp1807

Seems like fds are bing directly used in this PR and that's why its working on Windows 11.

PhilippWillms · 2025-06-18T16:57:42Z

Could we get a cherry-pick for that PR to be integrated in recently published 2.47 release?

jjyao · 2025-06-18T17:23:01Z

@PhilippWillms we will release it as 2.48 which should be released in 2-3 weeks. Would that work?

israbbani

I don't understand the change fully, but it might be worth documenting this in the code to aid future debugging.

Thanks for debugging and fixing this!

israbbani · 2025-06-19T18:56:23Z

src/ray/util/compat.cc

+  if (handle == INVALID_HANDLE_VALUE) {
+    return Status::IOError("") << "Fails to get file handle for flushing";
+  }
+  if (!FlushFileBuffers(handle)) {


I don't pretend to understand this and the PR is merged, but could you please leave a comment in compat.h that describes your current understanding and can aid in future debugging? This might break again in the future.

israbbani · 2025-06-19T18:57:28Z

src/ray/util/compat.cc

+Status CompleteWrite(int fd, const char *data, size_t len) {
+  const int ret = _write(fd, data, len);
+  if (ret == -1) {
+    return Status::IOError("") << "Fails to write to file because " << strerror(errno);


s/Fails/Failed/

israbbani · 2025-06-19T18:57:36Z

src/ray/util/compat.cc

+Status CompleteWrite(int fd, const char *data, size_t len) {
+  const int ret = _write(fd, data, len);
+  if (ret == -1) {
+    return Status::IOError("") << "Fails to write to file because " << strerror(errno);


s/Fails/Failed/

israbbani · 2025-06-19T18:58:09Z

src/ray/util/compat.cc

  }
  return Status::OK();
 }
-Status Close(MEMFD_TYPE_NON_UNIQUE fd) {


Why was this MEMFD_TYPE_NON_UNIQUE before and isn't that just an alias for int anyway?

israbbani · 2025-06-19T18:58:32Z

src/ray/util/compat.cc

  }
-  if ((DWORD)len != bytes_written) {
+  if (ret != static_cast<int>(len)) {
    return Status::IOError("") << "Fails to write all requested bytes, requests to write "


s/Fails/Failed/
s/write/wrote/

israbbani · 2025-06-19T18:59:56Z

src/ray/util/pipe_logger.cc

-  auto ostream =
-      std::make_shared<boost::iostreams::stream<boost::iostreams::file_descriptor_sink>>(
-          std::move(fd_sink));
+#if defined(__APPLE__) || defined(__linux__)


Probably also deserves a comment that explains the edge case that caused this issue.

PhilippWillms · 2025-06-19T21:08:47Z

@PhilippWillms we will release it as 2.48 which should be released in 2-3 weeks. Would that work?

Ok, fine for me.

…ect#53852) Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com> Signed-off-by: elliot-barn <elliot.barnwell@anyscale.com>

Use fd instead of handle for windows log redirection

f677eec

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

jjyao added the go add ONLY when ready to merge, run all tests label Jun 16, 2025

jjyao added 3 commits June 16, 2025 09:35

up

dff5223

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

up

c380097

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

up

0aab32b

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

jjyao commented Jun 17, 2025

View reviewed changes

jjyao marked this pull request as ready for review June 17, 2025 05:44

jjyao requested a review from a team as a code owner June 17, 2025 05:44

jjyao assigned israbbani Jun 17, 2025

fix

d390abc

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

czgdp1807 approved these changes Jun 17, 2025

View reviewed changes

jjyao requested a review from edoakes June 17, 2025 20:24

edoakes approved these changes Jun 17, 2025

View reviewed changes

jjyao merged commit 906db98 into ray-project:master Jun 18, 2025
5 checks passed

jjyao deleted the jjyao/fdwin branch June 18, 2025 04:47

israbbani reviewed Jun 19, 2025

View reviewed changes

minerharry pushed a commit to minerharry/ray that referenced this pull request Jun 27, 2025

[Core] Use fd instead of handle for windows log redirection (ray-proj…

248c707

…ect#53852) Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>

elliot-barn pushed a commit that referenced this pull request Jul 2, 2025

[Core] Use fd instead of handle for windows log redirection (#53852)

a6e4521

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com> Signed-off-by: elliot-barn <elliot.barnwell@anyscale.com>

Conversation

jjyao commented Jun 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why are these changes needed?

Related issue number

Checks

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

edoakes commented Jun 17, 2025

Uh oh!

czgdp1807 commented Jun 17, 2025

Uh oh!

czgdp1807 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

PhilippWillms commented Jun 18, 2025

Uh oh!

jjyao commented Jun 18, 2025

Uh oh!

israbbani left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PhilippWillms commented Jun 19, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

jjyao commented Jun 16, 2025 •

edited

Loading