Summary:
IOBufs now support larger than 4GiB chunks. One of the checks
was not removed. Tracing through the capacity, its use if all 64-bit
compatible.

Test Plan: moveToFbString a 4GiB+ IOBuf element.

Reviewed By: maxim@fb.com

Subscribers: net-systems@, sdwilsh, njormrod, folly-diffs@

FB internal diff: D1705316

Tasks: 5648445

Signature: t1:1705316:1416965241:d9d064bbde53e033d32bfe872b3445ee59005f5e

Summary:
If a single IOBuf has size exceeding 2^32, then our zlib
compression algorithm fails. Specifically, zlib z_stream.avail_in is
only 32 bytes (I think it's a long?
http://www.gzip.org/zlib/zlib_faq.html#faq32), and so a too-big IOBuf
will overflow the z_stream and cause data loss.

This diff breaks up large IOBufs into smaller chunks.

Test Plan:
fbconfig -r folly && fbmake runtests

Also compressed biggrep's configerator blob, which is how this bug was
caught. It now works. See the associated task.

Reviewed By: robbert@fb.com

Subscribers: trunkagent, sdwilsh, njormrod, folly-diffs@

FB internal diff: D1702925

Tasks: 5648445

Signature: t1:1702925:1416958232:459d498ff1db13e1a20766855e6f2f97da8cde8c

Summary:
Some platforms that SSLContext run on call exit() despite best practices.
This cause the static structures to be destructed and cause race conditions
and crashes.

This new's the static structures so that they won't be destructed on exit()

Test Plan: Unit tests

Reviewed By: seanc@fb.com

Subscribers: trunkagent, ssl-diffs@, seanc, njormrod, folly-diffs@

FB internal diff: D1702186

Tasks: 5572637

Signature: t1:1702186:1416941649:c5bdfd8cc604fc3618f51bdb8b910b5b2cf350ad

Summary: also kill duplicated test file

Test Plan: run all exp/wangle tests

Reviewed By: davejwatson@fb.com

Subscribers: fugalh, njormrod, folly-diffs@

FB internal diff: D1698503

Signature: t1:1698503:1416612139:7139c6aa2ca79eabdaec6c33192194df984af348

Summary: bad merge with D1661169 caused a unittest to fail.  Update unittest

Test Plan: fbconfig -r folly/experimental/wangle; fbmake runtests

Reviewed By: jsedgwick@fb.com

Subscribers: doug, fugalh, njormrod, folly-diffs@

FB internal diff: D1698435

Signature: t1:1698435:1416611458:555b6b50baa1bc5850b10e19c9c94978f93f2a5a

Summary: As above. This paid off with a couple bugfixes.

Test Plan: run em all

Reviewed By: hans@fb.com

Subscribers: fugalh, njormrod, folly-diffs@

FB internal diff: D1695106

Signature: t1:1695106:1416522038:0c3345aadf954bf346d35b99877e7f8dfcf3ceff

Summary:
mostly copypasta from TAsyncTransportHandler

Test Plan: compiles

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@

FB internal diff: D1690973

Signature: t1:1690973:1416529528:4feb187a68ad5405662b9b0efb160edd253a2977

Summary:
MLE: my first custom matcher

revealed some include issues, also fixed

Test Plan: guess

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@

FB internal diff: D1686223

Signature: t1:1686223:1416334120:41bce88d56dd9ca107f3ddc3927c6be9419370cf

Summary:
Basically the same interface as before, but you must specify the read and write types for the ends of pipeline. Implementation is cleaner as well; there's fewer levels of indirection

This dynamic casts shit all over the place and is less typesafe then the previous iteration, but I think with some carefully placed static_asserts, could be just as safe (in the case where you don't do any modification, anyway)

Right now you can only add to the front or back of the pipeline but the way it's set up you could add any number of mutations, including ones that are triggered by handlers. But this should (might?) be enough for Tunnel, which was the motivation.

Test Plan: basic test compiles, thrift2 diff still works with a one line change

Reviewed By: hans@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@, bmatheny

FB internal diff: D1661169

Tasks: 5002299

Signature: t1:1661169:1416521727:1f126279796c0b09d1905b9f7dbc48a9e5540271

Summary: Now a subprocess can be reliably made a group leader -- good for job control.

Test Plan:
unit test, checked that the pgid test worked in bash, too (only OKAY_XXX is printed)

==> XXX.sh <==
#!/bin/sh
test $(cut -d ' ' -f 5 /proc/$$/stat) == $$ && echo OKAY_XXX
./YYY.sh

==> YYY.sh <==
#!/bin/sh
test $(cut -d ' ' -f 5 /proc/$$/stat) == $$ && echo OKAY_YYY
./ZZZ.sh

==> ZZZ.sh <==
#!/bin/sh
test $(cut -d ' ' -f 5 /proc/$$/stat) == $$ && echo OKAY_ZZZ

Reviewed By: simpkins@fb.com

Subscribers: net-systems@, njormrod, folly-diffs@, simpkins

FB internal diff: D1492526

Signature: t1:1492526:1416528620:3cf98b1c1e334a7d551b2c2f3e76b1c70f07ad7c

Summary: Diff that had @override-unit-failures broke the build

Test Plan: fbconfig -r folly && fbmake dbg && fbmake runtests

Reviewed By: robbert@fb.com

Subscribers: sdwilsh, njormrod, folly-diffs@

FB internal diff: D1694388

Tasks: 5662947

Signature: t1:1694388:1416510278:d50f4896814a34d73d55be292961f5ecd74b1bb4

Blame Revision: D1680735

Summary: someone made another event base test while the big set still lived in thrift. merge it now that the old tests have been moved.

Test Plan: fbconfig folly/io/async/test && fbmake runtests --extended-tests

Reviewed By: davejwatson@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1690764

Signature: t1:1690764:1416413748:f3ae64d509432844bb2e1b689a75d66825269d15

Summary: Documentation brain-dump.  The sources already have pretty good doxygen-style comments, but could use a good high-level overview.

Test Plan: Reviewers get to edit

Reviewed By: jsedgwick@fb.com, afrind@fb.com, dcsommer@fb.com, alandau@fb.com

Subscribers: fugalh, subodh, kmdent, seanc, vloh, sjenkins, kparton, doug, net-systems@, njormrod, folly-diffs@, sgolemon, alikhtarov, andrii, tudorb

FB internal diff: D1690976

Tasks: 5648428

Signature: t1:1690976:1416419706:3894976c32904f040d518b9a4ce26ccb40d7ed80

Summary:
This one is a little tricky, but it explains what happened in t5388156. Basically what we observed in the stack traces is that the HHWheelTimer count_ is 0, but there is still a timeout scheduled on the AsyncTimeout.

How I can see this happening is the following:
- SASL times out
- We fall back to insecure mode in the callback and schedule a send of a real message to the server. Here we schedule another timeout on the same HHWheelTimer object. Since count_ is 0 at this point, we schedule a timeout in AsyncTimeout.
- Now the initial SASL timeout was late, so the HHWheelTimer is in a catchup state. This means that it's possible for the callback on the timeout that was just scheduled to fire on the same "tick".
- This callback invokes SR's callbacks which try to detach the event base from the channel. But it can't detach because we have something scheduled on the AsyncTimeout.
- We crash.

Test Plan:
- will wait for contbuild
- Something like this repros the problem: https://phabricator.fb.com/P17220063. Note I hacked the catchup part to get it to work. Now running loadgen with security = permitted causes the crash.

// hphp tests are stalled, have to do this...

Reviewed By: mshneer@fb.com

Subscribers: bmatheny, mshneer, sandeepkk, trunkagent, njormrod, folly-diffs@, alandau, mhorowitz, srenfro, hitesh, wstefancik

FB internal diff: D1680735

Tasks: 5388156

Signature: t1:1680735:1416252123:e76668860ccda9380a87996b4fa3de957e129404

Summary:
All pretty trivial:

https://github.com/facebook/folly/issues/99
https://github.com/facebook/folly/issues/98
https://github.com/facebook/folly/issues/45

Test Plan: Will watch jenkins fbthrift build

Reviewed By: dcsommer@fb.com

Subscribers: doug, njormrod, folly-diffs@

FB internal diff: D1669952

Signature: t1:1669952:1415643677:906234f0a89f38645b0072d3c88762d8fa2729dc

Summary:
ServerBootstrap a la netty.

This should be enough for some refactoring of thrift server and proxygen servers - but there are still lots of TODOs left to do

Test Plan:
Unittests included

Depends on D1638358

Reviewed By: jsedgwick@fb.com

Subscribers: trunkagent, doug, fugalh, alandau, bmatheny, mshneer, jsedgwick, afrind, dcsommer

FB internal diff: D1649521

Tasks: 5488516

Signature: t1:1649521:1416256073:fc003fd471bdfd137160dd6d7befd933ee8addd2

Summary: Fix the TARGETS from the refactor

Test Plan: ran it locally

Reviewed By: davejwatson@fb.com

Subscribers: doug, njormrod, folly-diffs@

FB internal diff: D1689151

Signature: t1:1689151:1416353673:3d3575b10f963e9f558954b8f3eba8bf48cdaa74

Summary:
getLoad() is called in servicerouter loadbalancing when a load_counter is not specified.  I added a custom getLoad() for leaf and made it so that different metrics can be tested without code pushes.

Changes:
1) Overloaded PriorityThreadManager::getCodel() so you can specify a PRIORITY.  aggr->leaf thrift calls are marked as HIGH priority.
2) The value of FLAGS_load_metric determines what metric the custom getLoad() will use (codel, max requests, or fallback to ThriftServer::getLoad()).
3) Smc listener for FLAGS_load_metric.
4) Removed server->setIsUnevenLoad(false) since there isn't reason this need to be set, and activeRequests_ and getPendingCount() are not tracked when isUnevenLoad_ is false.

Test Plan: Run leaf and check that smc listener works.  Custom getLoad() won't be called until t5625774 is resolved.

Reviewed By: hitesh@fb.com

Subscribers: njormrod, folly-diffs@, trunkagent, varunk, fugalh, alandau, nkgupta, bmatheny, jaredg, xie, mshneer, yguo

FB internal diff: D1685215

Signature: t1:1685215:1416339297:0d1f8cb0cec27242d95f0067337d80149b0a720e

Summary: Clean up API per stepan's comments in D1641057: runBeforeLoop callbacks should be run before the if block, so runInLoop calls will work.

Test Plan: added unittest

Reviewed By: hans@fb.com

Subscribers: trunkagent, doug, alandau, bmatheny, njormrod, mshneer, folly-diffs@

FB internal diff: D1679773

Signature: t1:1679773:1415916488:f20dd98bf40dd4cb19a4b11875dc0c9fa3e7870e

Blame Revision: D1641057

Summary:
These lived in fbthrift previously, but they should move with the
main code too.

Test Plan: unit tests

Reviewed By: davejwatson@fb.com

Subscribers: doug, alandau, bmatheny, njormrod, mshneer, folly-diffs@

FB internal diff: D1683229

Signature: t1:1683229:1416010730:36fb7e4c9916ae7a9b5972cd476f82014c5f4c78

Summary: as discussed

Test Plan: compiles :)

Reviewed By: steveo@fb.com

Subscribers: fugalh, njormrod, folly-diffs@

FB internal diff: D1687921

Signature: t1:1687921:1416331366:4e58a69daf39797e0d456ba86a463a9da2fd0369

Summary:
SSLContext takes it upon itself to
perform initialization of openssl along with
cleaning it up.

However in applications where there are multiple
libraries using openssl, cleanup is not really
desirable. Thus the application should own init
and cleanup of openssl, rather than SSLContext.

This is bad because openssl's init apis are not thread
safe.

This diff modifies SSLContext with 3 changes to
make this possible in the context of apps:

1. Add macro SSLCONTEXT_NO_REFCOUNT, for apps to be
able to disable refcounting
2. Move all steps of init into initializeOpenSSL.
As a result, apps can choose when to call
SSLContext::intitializeOpenSSL and if it has
been called before, openssl will not be reinitialized.
3. Move randomize as a static method to make it a part
of the initialization process.

Test Plan: Unit tests.

Reviewed By: ranjeeth@fb.com

Subscribers: ranjeeth, folly-diffs@, ssl-diffs@, njormrod, seanc, trunkagent, dihde

FB internal diff: D1686397

Signature: t1:1686397:1416270565:de805794d65af1c164ed334ff7ba93fe60b2b78a

Summary:
Change counting index to unsigned to avoid unsigned-signed
comparison errors.

@build-break

Test Plan:
fbconfig -r folly && fbmake dbg
fbmake runtests wangle

Reviewed By: robbert@fb.com, davejwatson@fb.com

Subscribers: sdwilsh, fugalh, njormrod, folly-diffs@

FB internal diff: D1686315

Signature: t1:1686315:1416267974:d8f6075550b9d611790263338dd4f49e7ef451cc

Blame Revision: D1595200

Summary:
provides new APIs observe(ObserverPtr<T>) and observe(Observer<T>*) that
are not associated with a subscription. The latter method avoids allocations at the cost of disowning responsiblity for object lifetimes.

It could be used for batch-allocating filter chains, for instance. I hacked up some recursive templates that would make that easy to do, but
haven't included them. I'd rather see the first use case and then decide whether to let users set them up themselves or to provide an
abstraction that makes it easier.

Test Plan: unit

Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, fugalh, njormrod

FB internal diff: D1595200

Tasks: 5002278

Signature: t1:1595200:1416266578:95c0532af42443fd83e2f94d30790c3c0831ce65

Summary:
see D1618704 and D1604575
Really need to make a decision about making a push for EventBaseManager everywhere. Getting/setting EBs when there's only ever one seems so silly, but in (e.g.) tests where one-off EBs are used without using the EBM, stuff like this will break.

Test Plan: definitely compiles and sorta works, will write unit tests if the dust settles

Reviewed By: davejwatson@fb.com

Subscribers: folly-diffs@, trunkagent, fugalh, njormrod

FB internal diff: D1618727

Summary:
* Unroll readVarint() - no loops in the fast path
* Split Cursor::skip() into fast and slow paths

Test Plan:
```
------ old ------   ------ new ------
Benchmark                               time/iter  iters/s  time/iter  iters/s        win
=========================================================================================
CompactProtocol_read_Empty                13.21ns   75.69M    13.21ns   75.71M      0.03%
CompactProtocol_read_SmallInt             29.95ns   33.39M    28.96ns   34.53M      3.41%
CompactProtocol_read_BigInt               67.83ns   14.74M    36.77ns   27.19M     84.46%
CompactProtocol_read_SmallString          55.72ns   17.95M    54.04ns   18.50M      3.06%
CompactProtocol_read_BigString           342.93ns    2.92M   332.18ns    3.01M      3.08%
CompactProtocol_read_BigBinary           186.18ns    5.37M   181.45ns    5.51M      2.61%
CompactProtocol_read_LargeBinary         190.65ns    5.25M   184.85ns    5.41M      3.05%
CompactProtocol_read_Mixed               101.97ns    9.81M    97.06ns   10.30M      4.99%
CompactProtocol_read_SmallListInt        148.09ns    6.75M   139.52ns    7.17M      6.22%
CompactProtocol_read_BigListInt           71.69us   13.95K    63.34us   15.79K     13.19%
CompactProtocol_read_BigListMixed          1.26ms   790.67     1.18ms   845.66      6.95%
CompactProtocol_read_LargeListMixed      139.80ms     7.15   134.68ms     7.42      3.78%
```

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, alandau, bmatheny, njormrod, mshneer, folly-diffs@

FB internal diff: D1681244

Signature: t1:1681244:1416265579:f630c53f9c31788f847d0af72198bbd9e5ea43f1

Summary: this should address some of the wordiness issues with exception_wrapper, i.e. always having to use make_exception_wrapper

Test Plan: wrote a little test to make sure everything compiles (or doesn't) as expected, not worth committing

Reviewed By: davejwatson@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1679460

Signature: t1:1679460:1416265170:5cd72d95dd855cd4e594dbbc49d0c53d012fbc99

Summary:
In unusual but possible circumstances, the EventBase and thus pending tasks will outlive the pool, so we shouldn't keep references of any kind to the pool in the task.
The only reference we were keeping was used to access the task stats rx subject. Store the subject as a shared ptr and give a copy of the ptr to the Thread object, which is itself
owned by a shared ptr and captured by every task. I thought this had to do with the thread local leak in mentioned in the test plan of D1682860 but this patch doesn't actually fix that :(
Thankfully, while task surfing I saw @phillip's awesome D1682698. Patching that in fixes the leak! Woo. Either way, this is more correct.

Test Plan: unit under clang/asan

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@, philipp

FB internal diff: D1683221

Tasks: 5336655

Signature: t1:1683221:1416264933:946d29b5a3eb22ed08812f2adefb7284b1899e4e

Summary:
just extend CPUThreadPoolExecutor to use a queue that is itself composed of N mpmc queues, one per priority

the verbosity is starting to kill me, i had thought before of truncating Executor of all these pool types and now I'm definitely going to do that unless someone fights me.

Test Plan: added unit; maybe i'm not being clever enough as i couldn't think of many ways to test this reliably so there's just a basic preemption test

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fugalh, njormrod, folly-diffs@, bmatheny

FB internal diff: D1676452

Tasks: 5002392

Signature: t1:1676452:1416263990:cdf5d44e4a50a6180ba547a3ed4c0c24d4ffdd8f

Summary: Fix addition bug instead of string concat

Test Plan: ?? It compiles?  The only call paths are when syscalls fail, it would be a lot of work to stub them out

Reviewed By: mshneer@fb.com

Subscribers: trunkagent, doug, njormrod, folly-diffs@

FB internal diff: D1679713

Tasks: 5604354

Signature: t1:1679713:1415925209:ed05d3d1ce3e7e8db32c4afc86c5f24748d9c65b

Summary:
Initial pass at moving acceptor to wangle.  Involves moving most of the config stuff from proxygen/lib/services, and *all* of the ssl stuff from proxygen/lib/ssl.

Only minor changes:
* Acceptor can be overriden to use thrift socket types, so I don't have to change TTransportException everywhere just yet
* proxygen::Exception to std::runtime_exception in a few spots - looks like it is entirely bad config exceptions, so it should be okay
* Just used std::chrono directly instead of stuff in Time.h (which is just typedefs and simple helpers)

Test Plan:
used in D1539327

fbconfig -r proxygen/httpserver; fbmake runtests

Probably other projects are broken, will iterate to fix

None of the failling tests look related

Reviewed By: dcsommer@fb.com

Subscribers: oleksandr, netego-diffs@, hphp-diffs@, ps, trunkagent, doug, fugalh, alandau, bmatheny, njormrod, mshneer, folly-diffs@

FB internal diff: D1638358

Tasks: 5002353

Signature: t1:1638358:1414526683:87a405e3c24711078707c00b62a50b0e960bf126

Summary:
See this LSan abort: https://phabricator.fb.com/P17233565.

Destructor of the object stored in `folly::ThreadLocal` itself may be using
`folly::ThreadLocal` with the same tag, with the current implementation these
objects may escape cleanup happening on thread exit.

Test Plan:
_build/opt/folly/test/thread_local_test --gtest_filter=ThreadLocalPtr.CreateOnThreadExit

Reviewed By: lucian@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1682698

Tasks: 5596043

Signature: t1:1682698:1416006810:100aaa5c17cecceeea568165d552d9d7907f38d0

Summary:
Add a specialized Cursor method that reads exactly 1 byte, in
addition to the existing one that reads into a buffer. Use it when
reading varints and standalone bytes in a CompactProtocol struct.

Test Plan:
```
------ old ------   ------ new ------
Benchmark                               time/iter  iters/s  time/iter  iters/s        win
=========================================================================================
CompactProtocol_read_Empty                18.68ns   53.54M    13.21ns   75.69M     41.37%
CompactProtocol_read_SmallInt             42.60ns   23.47M    29.95ns   33.39M     42.27%
CompactProtocol_read_BigInt               83.62ns   11.96M    68.40ns   14.62M     22.24%
CompactProtocol_read_SmallString          67.33ns   14.85M    55.62ns   17.98M     21.08%
CompactProtocol_read_BigString           353.83ns    2.83M   330.19ns    3.03M      7.07%
CompactProtocol_read_BigBinary           190.82ns    5.24M   182.90ns    5.47M      4.39%
CompactProtocol_read_LargeBinary         200.95ns    4.98M   187.00ns    5.35M      7.43%
CompactProtocol_read_Mixed               137.42ns    7.28M   102.98ns    9.71M     33.38%
CompactProtocol_read_SmallListInt        203.98ns    4.90M   146.68ns    6.82M     39.18%
CompactProtocol_read_BigListInt          120.50us    8.30K    71.56us   13.97K     68.31%
CompactProtocol_read_BigListMixed          1.62ms   617.07     1.26ms   795.60     28.93%
CompactProtocol_read_LargeListMixed      177.50ms     5.63   140.73ms     7.11     26.29%
```

Reviewed By: haijunz@fb.com

Subscribers: trunkagent, alandau, bmatheny, njormrod, mshneer, folly-diffs@

FB internal diff: D1678077

Signature: t1:1678077:1415923409:22accee6b62b6e2bf471f3758a290f71978a8c4e

Summary: Introduces a Singleton which keeps a map persistence_id => mcrouter_t*. Makes mcrouter instance not know if it's managed by McrouterManager.

Test Plan: unit tests

Reviewed By: pavlo@fb.com

Subscribers: trunkagent, alikhtarov, njormrod, folly-diffs@

FB internal diff: D1673274

Signature: t1:1673274:1415735863:c990a6a526f9525c68cc23892d690a9b3cb94ace

Summary:
It already does the work (`runInEventBaseThread`) but it will now be convenient to pass an `EventBase` where wangle wants an `Executor`.

Had to rip off the `boost::noncopyable` from `wangle::Executor` which is an interface and does not require non-copyability so that didn't really belong there in the first place I think. (Without this change, you get an obscure compiler error because of the double-inheritance from `boost::noncopyable`).

Test Plan: Things build, tests pass

Reviewed By: davejwatson@fb.com

Subscribers: jsedgwick, trunkagent, fugalh, exa, njormrod, folly-diffs@, andrii

FB internal diff: D1671500

Signature: t1:1671500:1415727572:a7dba33c669ca122aecaee3c700f9e53e54838d1

Summary:
Patch D1585087 exposes two flaws in EventBase().  It introduces IO
worker threads to the ThriftServer which are constructed/destructed in
parallel.

Within the construction phase, a new EventBase() is instantiated for
each thread and unwound in destruction.

When using the BaseControllerTask (in Python), the following sequence
is observed:

a = event_init() [ThriftServer]
b = event_init() [IO worker 1]
c = event_init() [IO worker 2]
...
event_base_free(c)
event_base_free(b)
event_base_free(a)  -> segfault

1. event_init() should only ever be called once.  It internally
modifies a global variable in libevent, current_base to match the
return value.  event_base_free() will set current_base back to NULL if
the passed in arg matches current_base.   Therefore subsequent calls
must use event_base_new().

2. Since current_base is a global and EventBase() is called by multiple
threads, it is important to guard with a mutex.  The guard itself also
exposed the bug because:

a = event_init()  [current_base = a]
b = event_init()  [current_base = b]
...
event_base_free(b) [b == current_base -> current_base = NULL]

So current_base ends up prematurely set to NULL.

Test Plan:
Run dba/core/daemons/dbstatus/dbstatus_tests.lpar, which no longer
segfaults

Reviewed By: jsedgwick@fb.com, davejwatson@fb.com

Subscribers: dihde, evanelias, trunkagent, njormrod, ncoffield, lachlan, folly-diffs@

FB internal diff: D1663654

Tasks: 5545819

Signature: t1:1663654:1415732265:d51c4c4cae99c1ac371460bf18d26d4f917a3c52

Blame Revision: D1585087

Summary: find a non-loopback ipv4 or ipv6 address and feed it to folly::IPAddress::hash

Test Plan:
1)
fbconfig -r mcrouter
fbmake runtests

2)
compare hostids on different hosts

Reviewed By: pavlo@fb.com

Subscribers: njormrod, folly-diffs@, trunkagent, ps, bmatheny, alikhtarov

FB internal diff: D1668944

Tasks: 5557721

Signature: t1:1668944:1415736928:fb4b042a575c0b00f52780f3abf54bf7630b3a97

Summary: Using a ThreadLocal cleans up the code quite a bit.  Also reuse the shared_ptr in create instead of creating a new one saves an allocation

Test Plan:
fbconfig thrift/lib/cpp/test:RequestContextTest; fbmake runtests

fbconfig common/services/cpp/test:trace_test; fbmake runtests

Reviewed By: hans@fb.com

Subscribers: trunkagent, doug, alandau, bmatheny, njormrod, mshneer, folly-diffs@, vloh

FB internal diff: D1663960

Signature: t1:1663960:1415390250:36d9b772016d2a12d804e98edbc1725af882e507