Any single-threaded task benchmark will spend a good chunk of time in `kqueue()` on osx and `epoll()` on linux, and the reason for this is that each time a task is terminated it will hit the syscall. When a task terminates, it context switches back to the scheduler thread, and the scheduler thread falls out of `run_sched_once` whenever it figures out that it did some work.
If we know that `epoll()` will return nothing, then we can continue to do work locally (only while there's work to be done). We must fall back to `epoll()` whenever there's active I/O in order to check whether it's ready or not, but without that (which is largely the case in benchmarks), we can prevent the costly syscall and can get a nice speedup.
I've separated the commits into preparation for this change and then the change itself, the last commit message has more details.
The green scheduler can optimize its runtime based on this by deciding to not go
to sleep in epoll() if there is no active I/O and there is a task to be stolen.
This is implemented for librustuv by keeping a count of the number of tasks
which are currently homed. If a task is homed, and then performs a blocking I/O
operation, the count will be nonzero while the task is blocked. The homing count
is intentionally 0 when there are I/O handles, but no handles currently blocked.
The reason for this is that epoll() would only be used to wake up the scheduler
anyway.
The crux of this change was to have a `HomingMissile` contain a mutable borrowed
reference back to the `HomeHandle`. The rest of the change was just dealing with
this fallout. This reference is used to decrement the homed handle count in a
HomingMissile's destructor.
Also note that the count maintained is not atomic because all of its
increments/decrements/reads are all on the same I/O thread.
This is part of the overall strategy I would like to take when approaching
issue #11165. The only two I/O objects that reasonably want to be "split" are
the network stream objects. Everything else can be "split" by just creating
another version.
The initial idea I had was the literally split the object into a reader and a
writer half, but that would just introduce lots of clutter with extra interfaces
that were a little unnnecssary, or it would return a ~Reader and a ~Writer which
means you couldn't access things like the remote peer name or local socket name.
The solution I found to be nicer was to just clone the stream itself. The clone
is just a clone of the handle, nothing fancy going on at the kernel level.
Conceptually I found this very easy to wrap my head around (everything else
supports clone()), and it solved the "split" problem at the same time.
The cloning support is pretty specific per platform/lib combination:
* native/win32 - uses some specific WSA apis to clone the SOCKET handle
* native/unix - uses dup() to get another file descriptor
* green/all - This is where things get interesting. When we support full clones
of a handle, this implies that we're allowing simultaneous writes
and reads to happen. It turns out that libuv doesn't support two
simultaneous reads or writes of the same object. It does support
*one* read and *one* write at the same time, however. Some extra
infrastructure was added to just block concurrent writers/readers
until the previous read/write operation was completed.
I've added tests to the tcp/unix modules to make sure that this functionality is
supported everywhere.
All tests except for the homing tests are now working again with the
librustuv/libgreen refactoring. The homing-related tests are currently commented
out and now placed in the rustuv::homing module.
I plan on refactoring scheduler pool spawning in order to enable more homing
tests in a future commit.
This reimplements librustuv without using the interfaces provided by the
scheduler in libstd. This solely uses the new Runtime trait in order to
interface with the local task and perform the necessary scheduling operations.
The largest snag in this refactoring is reimplementing homing. The new runtime
trait exposes no concept of "homing" a task or forcibly sending a task to a
remote scheduler (there is no concept of a scheduler). In order to reimplement
homing, the transferrence of tasks is now done at the librustuv level instead of
the scheduler level. This means that all I/O loops now have a concurrent queue
which receives homing messages and requests.
This allows the entire implementation of librustuv to be only dependent on the
runtime trait, severing all dependence of librustuv on the scheduler and related
green-thread functions.
This is all in preparation of the introduction of libgreen and libnative.
At the same time, I also took the liberty of removing all glob imports from
librustuv.
Understand 'pkgid' in stage0. As a bonus, the snapshot now contains now metadata
(now that those changes have landed), and the snapshot download is half as large
as it used to be!
This replaces the link meta attributes with a pkgid attribute and uses a hash
of this as the crate hash. This makes the crate hash computable by things
other than the Rust compiler. It also switches the hash function ot SHA1 since
that is much more likely to be available in shell, Python, etc than SipHash.
Fixes#10188, #8523.
This commit implements the support necessary for generating both intermediate
and result static rust libraries. This is an implementation of my thoughts in
https://mail.mozilla.org/pipermail/rust-dev/2013-November/006686.html.
When compiling a library, we still retain the "lib" option, although now there
are "rlib", "staticlib", and "dylib" as options for crate_type (and these are
stackable). The idea of "lib" is to generate the "compiler default" instead of
having too choose (although all are interchangeable). For now I have left the
"complier default" to be a dynamic library for size reasons.
Of the rust libraries, lib{std,extra,rustuv} will bootstrap with an
rlib/dylib pair, but lib{rustc,syntax,rustdoc,rustpkg} will only be built as a
dynamic object. I chose this for size reasons, but also because you're probably
not going to be embedding the rustc compiler anywhere any time soon.
Other than the options outlined above, there are a few defaults/preferences that
are now opinionated in the compiler:
* If both a .dylib and .rlib are found for a rust library, the compiler will
prefer the .rlib variant. This is overridable via the -Z prefer-dynamic option
* If generating a "lib", the compiler will generate a dynamic library. This is
overridable by explicitly saying what flavor you'd like (rlib, staticlib,
dylib).
* If no options are passed to the command line, and no crate_type is found in
the destination crate, then an executable is generated
With this change, you can successfully build a rust program with 0 dynamic
dependencies on rust libraries. There is still a dynamic dependency on
librustrt, but I plan on removing that in a subsequent commit.
This change includes no tests just yet. Our current testing
infrastructure/harnesses aren't very amenable to doing flavorful things with
linking, so I'm planning on adding a new mode of testing which I believe belongs
as a separate commit.
Closes#552
These two attributes are no longer useful now that Rust has decided to leave
segmented stacks behind. It is assumed that the rust task's stack is always
large enough to make an FFI call (due to the stack being very large).
There's always the case of stack overflow, however, to consider. This does not
change the behavior of stack overflow in Rust. This is still normally triggered
by the __morestack function and aborts the whole process.
C stack overflow will continue to corrupt the stack, however (as it did before
this commit as well). The future improvement of a guard page at the end of every
rust stack is still unimplemented and is intended to be the mechanism through
which we attempt to detect C stack overflow.
Closes#8822Closes#10155
In the ideal world, uv I/O could be canceled safely at any time. In reality,
however, we are unable to do this. Right now linked failure is fairly flaky as
implemented in the runtime, making it very difficult to test whether the linked
failure mechanisms inside of the uv bindings are ready for this kind of
interaction.
Right now, all constructors will execute in a task::unkillable block, and all
homing I/O operations will prevent linked failure in the duration of the homing
operation. What this means is that tasks which perform I/O are still susceptible
to linked failure, but the I/O operations themselves will never get interrupted.
Instead, the linked failure will be received at the edge of the I/O operation.
