Refine file_impl creation by reactor #3116
Draft
Conversation
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
Contributor
Author
|
@scylladb/seastar-maint , please review |
This is to keep only the fs_info evaluation in the lambda. This will help detaching this code into its own helper function in the next patch. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Place it in the internal namespace, next to s_fstype map itself. And also move the associated xfs_concurrency_from_kernel_version() helper. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Next patches will pass fs_info reference around preemption points, so the object itself should be non-moveable. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Now this configuration is applied directoy to the fs_info object. Next patches will move those objects creation to thread-pool thread where no reactor and its config are available. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Code that opens file or directory does it via thread-pool thread. In it not only the open() itself is called, but also stat() is. Next patches will make those lambdas call fstatfs() too and return back both -- the stat struct and fs_info pointer. This patch prepares this code to carry fs_info pointer out of thread-pool-invoked lambda. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Right now the fs_info map is populated in the reactor thread. This is done it make_file_impl -- it calls reactor::fstatfs(), that switches to thread-poll, then it instantiates an fs_info object in global map. After this patch, calling fstatfs and populating the map of fs_info-s is done in the thread pool. The caller in this case is reactor::open_file_dma and reactor::open_directory that do opening in thread-pool anyway. The returned pointer to fs_info object is then passed (in reactor thread) to make_file_impl() which no longer needs to access the map of fs_infos. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Now it always returns ready futures with file_impl pointer, no need to mess with futures any longer. Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
This method is private and was only used to facilitate the make_file_impl Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
xemul
force-pushed
the
br-file-move-fsinfo-get-into-open-thread
branch
from
2c83dbb to
ac83108
Compare
Contributor
Author
|
upd:
|
avikivity
reviewed
Dec 2, 2025
| int r = ::fcntl(fd, F_SETFL, open_flags | o_direct_flag); | ||
| if (r == -1 && strict_o_direct) { | ||
| auto maybe_ret = wrap_syscall(r, st); // capture errno (should be EINVAL) | ||
| auto maybe_ret = wrap_syscall(r, std::make_pair(st, fsi)); // capture errno (should be EINVAL) |
Member
There was a problem hiding this comment.
I don't think we should be moving fs_info here. First, I'd like to keep the syscall thread minimal since it operates without a full environment (we don't want to allocate here), and second, this code is specific to non-io_uring paths. io_uring is able to execute open() via io_uring and we'd have to duplicate this code.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Creation of file is done in two reactor methods -- open_file_dma() and open_directory(). Both follow similar scheme: first they switch to thread-poll to call open(name) and stat(fd), then call make_file_impl() which returns a future (!) with file_impl pointer, and finally the file is created. The make_file_impl, in turn, is sleeping function because it sometimes (usually once in a lifetime) wants to call thread-poll again to call fstatfs(fd) and populate a collection of fs_info objects.
This PR moves the call to fstats() from make_file_impl() to open_...()'s thread-pool lambda. This change has several interesting consequences.