Summary:
See task T70610458 description for the TSAN error reported.

The scenario that can trigger this race condition is:
 * Thread A has a ThreadEntryNode N whose `next` points to thread B's ThreadEntry.
 * Thread B exits, which causes it to delete its ThreadEntry and all its nodes; therefore writing `N->next = <the next entry after B>`.
 * At the same time, thread A accesses the thread local corresponding to N; when checking whether the ThreadEntryNode is already added to the linked list, it checks whether the `next` pointer is null; this read races with the above write.

Even though this is a technically a race condition, it is unlikely to actually trigger any negative effect in practice, because at thread exit, the overwrite of `N->next = ...` will never assign a nullptr, so it will never affect the outcome of `if (!next)` in thread A. Only the thread that owns the ThreadEntryNode has the ability to change between `next == nullptr` and `next != nullptr`. (The former is called "zero" in the code). Nonetheless, C++ says "race condition leads to undefined behavior", so it's still good to fix...

It's not clear what the best way to fix this is. Here I've presented one fix, which is to introduce another field in ThreadEntryNode that is private to the thread and simply indicates whether the node is zero; this field is strictly accessed by only the owning thread and therefore will not cause a race condition. I used a bit in `id` to not increase memory consumption, but maybe that's not important - I don't know. I'm also unsure about the performance implications.

Let me know if anyone has different ideas; I'm very open to suggestions here.

Reviewed By: yfeldblum

Differential Revision: D22743031

fbshipit-source-id: 12722ca241a224b546a0f6763b125e922813c844

Summary: Turned it into a thread_local instead. Only affects tests.

Reviewed By: yfeldblum

Differential Revision: D22744469

fbshipit-source-id: cbe8263d560f5abad27c7e61b1963aecf090a94e

Summary:
See RefCountTest.cpp for a newly added test. When one thread is decrementing a local refcount and another thread is decrementing a global refcount, and the global decrement results in global zero, the second thread can reasonably expect to be able to safely destroy the refcount object. However, a memory ordering issue in LocalRefCount::collect() causes TSAN to detect a write-delete race condition on the inUpdate_ variable (though that variable is not the cause of the issue).

This diff changes collect() to use memory_order_acquire so that upon reading false, collect() may exit while assuming that all memory accesses in update() are complete. This way, any memory accesses (such as a read of refCount_.state_ in update()) happen before a delete of the whole TLRefCount like so:

 - A memory access in update() is sequenced before the release store of false into inUpdate_ at the SCOPE_EXIT of the function,
 - which synchronizes with the acquire load at the end of collect()
 - which is sequenced before the deletion that can only happen after collect()

Without this diff, the second relationship above is broken and TSAN complains.

I also updated the comment on top of inUpdate_ to better reflect the purpose. I couldn't understand the original comment (which seems to suggest it's related to refcount *correctness* rather than a write-delete race), but maybe I missed something there.

Reviewed By: yfeldblum

Differential Revision: D22805717

fbshipit-source-id: 34cf72676760526ba457f939e307ed03dc722528

Summary:
we do have some periodic maintentenance tasks... if those are initialized
lazily on a requests handling path, they  will capture current rctx and keep it
it alive forever.

Reviewed By: yfeldblum

Differential Revision: D22859344

fbshipit-source-id: 3655d874dee22aaf74b4d80631d4a58e83539752

Summary: AppBytesWitten was no longer updated correctly for AsyncSSLSocket, fix this.

Reviewed By: bschlinker

Differential Revision: D22847139

fbshipit-source-id: 8cabb9883e5ad64c49a283a0d83b8087def4d5ae

Summary:
Currently when context string is "", we will still append whitespace to log header. This looks like:
```
I0729 17:36:53.249270 438847 ThriftServerEventHandler.cpp:21 ] Sampling servers
```
which is less readable compared to
```
I0729 17:36:53.249270 438847 ThriftServerEventHandler.cpp:21] Sampling servers
```
. Ideally empty context string should mean "there's no context string to report here", so we should not put an extra delimiter in such case.

Reviewed By: simpkins

Differential Revision: D22835634

fbshipit-source-id: c7fc6de4036050eac3b7d16bf607c637d1394c44

Reviewed By: yfeldblum

Differential Revision: D22843001

fbshipit-source-id: 55f48de90f16c53983ea94c8e1102c5b33e25195

Summary:
I'm using AsyncPipe heavily in D22593723. At one iteration I needed to assign with move a holding object and it wouldn't compile because the operator doesn't return any value

Here's the error: `folly/experimental/coro/AsyncPipe.h:54:3: error: control reaches end of non-void function [-Werror,-Wreturn-type]`

Reviewed By: yfeldblum

Differential Revision: D22593724

fbshipit-source-id: 53f27b95c1dff0068a375291d4ee6f445f892dd9

Summary: [Folly] Move an `exception_ptr` in `exceptionStr`, which should be ever-so-slightly cheaper than copying it since that may avoid atomic refcount operations. It's not much cheaper overall because the throw/catch machinery remains expensive.

Reviewed By: ot, luciang

Differential Revision: D22529858

fbshipit-source-id: 7e5d4ec5c1ca601a766c63133ae759651ae7be85

Summary: [Folly] Outline `exceptionStr` function bodies and move them to the `.cpp`.

Reviewed By: ot, luciang

Differential Revision: D22529702

fbshipit-source-id: 165c14009f224e86e63b3a3126c8b837cf3793f5

Summary: [Folly] Skip the safe-assert dep on FileUtil since it really needs only the combinators.

Reviewed By: Orvid

Differential Revision: D22699899

fbshipit-source-id: 73485439e397c10210d09eacbb38116e3e0a50a0

Summary:
F14MapFallback.h and F14SetFallback.h can now be compiled
standalone when F14 intrinsics are disabled.

Reviewed By: andrewjcg, yfeldblum

Differential Revision: D22785153

fbshipit-source-id: 96a1d3b2773a83dee8484fa40f35bbc791aec57c

Summary: [Folly] Drop support for boost below 1.61, which means removing multiple versions of support for `boost-context` embedded in `folly/fibers/`.

Reviewed By: fredemmott

Differential Revision: D22779795

fbshipit-source-id: fafc9b38784dcb206283e11d01d77dd8655354a6

Summary: [Folly] Tweak safe-assert uses of wrapper functions.

Reviewed By: nbronson

Differential Revision: D22699636

fbshipit-source-id: 97592e55f64f2e819680424c0f1ddd97e1f7b96c

Summary: adding CANCELED to AsyncSocketExceptionType

Reviewed By: yfeldblum

Differential Revision: D22345735

fbshipit-source-id: 73c61b409531d71a39d82a46d9c0ebc720801c45

Summary:
Adding a new method `ConcurrentHashMap::erase_key_if`. This is a more general form of `erase_if_equal`, which allows arbitrary predicates to be evaluated against the existing value type, rather than simple equality.

`erase_key_if` was chosen as a name to avoid confusion with `std::erase_if`.

Reviewed By: magedm

Differential Revision: D22775217

fbshipit-source-id: 151975216d5d4288ec4e88b81f6a8d17a091e8f0

Summary:
- I find the ability to make a branch and play with CI on GitHub super handy
- Removing the `- master` limitation gives me this ability

Reviewed By: yi-xian

Differential Revision: D22771835

fbshipit-source-id: 8e8839cb860ab4d1dfa0dda590afaf165127f60d

Summary: Fix 1-byte buffer overrun when parsing the string "-".

Reviewed By: mhlakhani

Differential Revision: D22695995

fbshipit-source-id: 9d1834c068fc3889a514b77d53d626eb6f72b6c6

Summary: In addition to `folly::fibers::Semaphore` which only accepts single token increment/decrement. `folly::fibers::BatchSemaphore` is able to accept batch tokens

Differential Revision: D21874795

fbshipit-source-id: 1d4c62bdb079c058529cd917078d81d09d585084

Summary:
* Implement `SemaphoreBase` which is able to accept batch tokens, most of the logic is copy-pasted from `folly::fibers::Semaphore` with batch token as parameter
* This serves as base object which is not public accesseble
* BatchSemaphore which is able to accept batch tokens and Semaphore will use `SemaphoreBase` as a fundation

Differential Revision: D21874699

fbshipit-source-id: 2536284a2480f2972bd00010833de49bcd5304b2

Summary:
- Replace deprecated method
- Stops spam to my console as I run with show deprecation warnings

Reviewed By: yi-xian

Differential Revision: D22776372

fbshipit-source-id: d29193e4c4c26d7facfabf9038dcb33c1af92434

Summary:
MSVC's maintains an intrusive linked list of iterators for
debugging purposes in some build configurations, which means that
F14MapFallback's hacky conversion from unordered_map::local_iterator to
unordered_map::iterator doesn't work.  Fortunately, on that platform
local_iterator and iterator are the same type, so no hack is needed.

Reviewed By: Orvid

Differential Revision: D22776318

fbshipit-source-id: 80283c4c451afffe522e55e47cd94971126a6fa6

Summary: we should report the name of the executore/threadpool that is saturated. propagate names of the executor/threadpool onto the queue we are monitoring.

Reviewed By: andriigrynenko

Differential Revision: D20773871

fbshipit-source-id: f7f78edd62d515cca19e30a63773180404843c1b

Summary:
as titled

The issue here was that we were clearing out the value of current rctx from TLS, then calling `onUnset`. So calling `RequestContext::get` from within the `onUnset` callback resulted in returning the default global RCTX.

Reviewed By: yfeldblum, zhxchen17

Differential Revision: D22750092

fbshipit-source-id: 509676d047cf6aec05d08ff3bf656d6d1558fbf5

Summary:
- Make OpenR build all deps from source until we remove fbzmq as a dep
- Allow a project to move it's Open Source CI forward to an alternative version of ubuntu via a new `--ubuntu-version` parameter

Reviewed By: wez

Differential Revision: D22768987

fbshipit-source-id: 07205efbd0c87a702638cf30b84a2850d064fa8e

Reviewed By: andrewjcg

Differential Revision: D22743063

fbshipit-source-id: 1da870027c92b11509056769626ed1f442d1b426

Summary: [Folly] Move the ubuntu build script to 18.04 from 14.04 and the boost default dep to 1.65.1.

Reviewed By: fredemmott

Differential Revision: D22779335

fbshipit-source-id: 61be6ef8f3cefa05a8d016a061cbc42605215711

Summary: `SDKROOT` is a requirement if we manually specify the location of the compiler on macOS. Otherwise it wouldn't be able to find the system libraries headers.

Reviewed By: wez

Differential Revision: D22577887

fbshipit-source-id: 98140e6f9e564d665db085d21023986b240b3732

Summary:
Original commit changeset: c6f013565680

Redo D22417488 (https://github.com/facebook/folly/commit/43d80f110b181dd7f72f0f935322a58f9cbb5147), with D22577887 fixing open source CI builds

Reviewed By: wez

Differential Revision: D22577884

fbshipit-source-id: 0c94483e05715493859d29bc2b58f0847a95826b

Summary:
The upstream [liburing](https://git.kernel.dk/cgit/liburing/tree/src/include/liburing.h) has changed to use ```extern "C" ```, we must remove the duplicated `extern "C"`.

Pull Request resolved: https://github.com/facebook/folly/pull/1416

Differential Revision: D22752570

Pulled By: yfeldblum

fbshipit-source-id: fa62d8e462e04676af6d563225455400d7e83c4b

Summary:
- OpenR building has been broken for quite awhile and it seems this is the main cause

Example CI Failure: https://github.com/facebook/openr/runs/631232675?check_suite_focus=true

Reviewed By: wez

Differential Revision: D22737174

fbshipit-source-id: 31d44c36d495295d30fe3ddc81991dafed100703

Summary: We should also use saturation detector on queues fronting CPUThreadPoolExecutor.

Reviewed By: andriigrynenko

Differential Revision: D20873689

fbshipit-source-id: 25ffc4cf62ddb222ede4f983f88a1c6545a0d4a5

Summary: The pthread_create redefinition also overrides the TSAN-specific pthread_create. This diff deletes this test because the test itself is of little value - see discussion in this diff.

Reviewed By: yfeldblum

Differential Revision: D22735273

fbshipit-source-id: a23089654697b85698f226b1110af3ed1a0b7fce

Summary: The test forgot to join the pthread.

Reviewed By: yfeldblum

Differential Revision: D22743458

fbshipit-source-id: 02231ce1e915b9f8679378eaf2ec18dc5dd67280

Summary:
Add some coroutine-native versions of the retrying algorithms available
for retrying folly::Future-based operations.

Adds the following:
* `retryWhen()` - This is the fundamental retry algorithm that allows
  you to specify logic to determine whether to retry and if so when to
  retry.
* `retryN()` - For simple use-cases that immediately retries operations
  up to N times.
* `retryWithExponentialBackoff()` - Retries the operation after an
   exponentially increasing delay with some random jitter applied.

Reviewed By: yfeldblum

Differential Revision: D22535487

fbshipit-source-id: c5594d0edfc2414015c345bee79e34f5cdafb789

Summary:
Building Folly on Fedora 32 fails with a link error, saying
SingletonVault::kDefaultShutdownTimeout_ is undefined. I did not
investigate, but this constant is only used in one place, so inline
it.

Reviewed By: yfeldblum

Differential Revision: D22713828

fbshipit-source-id: 707797eaef4633e642d07b2b02d9f55187afb99f

Summary:
This algorithm takes an awaitable and a duration and
cancels the operation and completes with a `FutureTimeout`
error if the operation does not complete within the specified
duration.

This differs from `timed_wait()` in that it still waits
for the child operation to complete, even once the timeout
has elapsed, rather than just detaching from the
operation and letting it complete in the background.

This relies on the operation responding to cancellation
in a timely manner for the algorithm to have the effect
of returning shortly after the timeout elapses.

The benefit of waiting for the child operation to complete,
however, is that it allows you to safely pass parameters by
reference to the child operation and be safe in knowing that
the child operation will complete before the awaiting
coroutine resumes and potentially destroys those parameters.

Reviewed By: yfeldblum

Differential Revision: D22199932

fbshipit-source-id: 4f6295a23a860ec52b639eb56a02f8029ffb9009

Summary: Basic FIFO queues observation hooks.

Reviewed By: andriigrynenko

Differential Revision: D20773870

fbshipit-source-id: 3f9cbab1b8b10204139ce153d92e08ffb5f6254c

Summary:
The `F14*Fallback.h` headers are meant to be included textually
from `F14*.h` headers and currently fail to compile in modular
builds, as a) they have references to items defined in their intended
enclosing header and b) even if successfully compiled alone would
provide conflicting definitions of the `F14*` structures.

This diff adds the `FOLLY_F14_VECTOR_INTRINSICS_AVAILABLE`
gate in the `F14*Fallback.h` headers so that they won't get compiled
when they're not used.

Reviewed By: yfeldblum, luciang

Differential Revision: D22568097

fbshipit-source-id: a75b3a4a687e3034dd9c4c3f7fd0a41bfd465272

Reviewed By: luciang

Differential Revision: D22555074

fbshipit-source-id: b853730e5d0dd96b6f94d71e2332653a1a44c9a1