folly.git
10 years agoget_fast/get_weak_fast API for folly::Singleton
Andrii Grynenko [Tue, 16 Dec 2014 04:36:43 +0000 (20:36 -0800)]
get_fast/get_weak_fast API for folly::Singleton

Summary: This adds API which makes folly::Singleton as performant as Meyers/static-object singletons.

Test Plan:
unit test + benchmark

============================================================================
folly/experimental/test/SingletonTest.cpp       relative  time/iter  iters/s
============================================================================
NormalSingleton                                            333.32ps    3.00G
MeyersSingleton                                  100.00%   333.33ps    3.00G
FollySingletonSlow                                 0.35%    94.36ns   10.60M
FollySingletonFast                                99.43%   335.24ps    2.98G
FollySingletonFastWeak                             0.62%    53.74ns   18.61M
============================================================================

Reviewed By: alikhtarov@fb.com

Subscribers: trunkagent, folly-diffs@

FB internal diff: D1741961

Signature: t1:1741961:1418765462:d9806f1bf5275bfbe2c4c53a41b735bda93753fe

10 years agoAdd specific multi-bind to AsyncServerSocket
Cameron Pickett [Tue, 16 Dec 2014 21:02:26 +0000 (13:02 -0800)]
Add specific multi-bind to AsyncServerSocket

Summary:
Add a bind api similar to binding by port across all interfaces.
However, this bind will only attempt to bind sockets to the supplied
IP addresses. I'd like to add this to aid in moving TURN to IPv6.

TURN will use two specified addresses, one for IPv4, and one for IPv6, to
handle incoming connections from clients. In order to avoid duplicating
workers, we'd like to take advantage of the multi-socket implementation
of AsyncServerSocket. However, we don't want to bind to all interfaces,
especially for port 443.

Test Plan:
Plan to test by adapting TurnTcpListener to use this new API. See that
1. using one address still works as expected,
2. using zero addesses will cause exception
3. using multiple addresses works as expected

Will write unit tests if needed

Reviewed By: davejwatson@fb.com

Subscribers: hannesr, trunkagent, net-systems@, folly-diffs@, naizhi

FB internal diff: D1729442

Tasks: 3633642

Signature: t1:1729442:1418752467:22a60da4ec9009ea0b7fe28a8a436a179e0449aa

10 years agofix Futex when steady_clock and system_clock precisions differ
Nathan Bronson [Tue, 16 Dec 2014 18:34:47 +0000 (10:34 -0800)]
fix Futex when steady_clock and system_clock precisions differ

Summary:
To handle the strange relationship between steady_clock and
system_clock (on some platforms these represent only one type, on some
platforms they are separate) Futex::futexWaitUntil converts the deadline
to a duration and back.  On Xcode 6 system_clock is measured in
microseconds and steady_clock in nanoseconds, resulting in a compilation
failure.  This diff fixes it.

Test Plan:
1. compile snippet manually using Xcode
2. new unit test that causes the same implicit conversion failure in a slightly different way

Reviewed By: mssarang@fb.com

Subscribers: trunkagent, folly-diffs@

FB internal diff: D1740903

Tasks: 5831196

Signature: t1:1740903:1418754770:32c999abf1dc87415ffebf45083a903abbded9f2

10 years agocodemod: folly/wangle/ -> folly/wangle/futures
James Sedgwick [Tue, 16 Dec 2014 18:26:24 +0000 (10:26 -0800)]
codemod: folly/wangle/ -> folly/wangle/futures

Summary: Last thing before moving experimental/wangle here. Once everything is in the same directory I'm probably going to consolidate experimental/wangle/concurrent with the executors in this directory into wangle/executors/. And probably rename some of these targets. For now, a dumb move is enough.

Test Plan: waiting for contbuild

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, chaoyc, search-fbcode-diffs@, lars, ruibalp, hero-diffs@, zeus-diffs@, vikas, danielg, mcduff, cold-storage-diffs@, unicorn-diffs@, ldbrandy, targeting-diff-backend@, netego-diffs@, fugalh, adamsyta, atlas2-eng@, alandau, bmatheny, adityab, everstore-dev@, zhuohuang, sweeney, mwa, jgehring, smarlow, akr, bnitka, jcoens, luk, zhguo, jying, apodsiadlo, alikhtarov, fuegen, dzhulgakov, mshneer, folly-diffs@, wch, tingy, hannesr

FB internal diff: D1740327

Tasks: 5802833

Signature: t1:1740327:1418752541:82d7486293b0a12938730ae66d480c120aded4dc

10 years agoRemove locking when getting ptr to Singleton
Andrii Grynenko [Fri, 5 Dec 2014 22:17:50 +0000 (14:17 -0800)]
Remove locking when getting ptr to Singleton

Summary: This removes one layer on locking on the fast path, when ptr to singleton object is read from SingletonEntry.

Test Plan:
unit test

Before:

============================================================================
folly/experimental/test/SingletonTest.cpp       relative  time/iter  iters/s
============================================================================
NormalSingleton                                            335.26ps    2.98G
MeyersSingleton                                   99.50%   336.96ps    2.97G
FollySingleton                                     0.28%   120.64ns    8.29M
============================================================================

After:

============================================================================
folly/experimental/test/SingletonTest.cpp       relative  time/iter  iters/s
============================================================================
NormalSingleton                                            336.76ps    2.97G
MeyersSingleton                                   99.91%   337.07ps    2.97G
FollySingleton                                     0.36%    92.69ns   10.79M
============================================================================

Reviewed By: alikhtarov@fb.com

Subscribers: trunkagent, folly-diffs@

FB internal diff: D1727604

Signature: t1:1727604:1418701171:1728b516191a8ec4439e981d78634370b4bcf7a1

10 years agoglobal io executor
James Sedgwick [Mon, 15 Dec 2014 21:09:47 +0000 (13:09 -0800)]
global io executor

Summary:
This is something we've talked about for a while. It's also an alternative to the mechanism in D1714645.
If we like it, I'll do something similar for a global cpu executor. That functionality should probably just be
baked into Executor itself instead of a separate subclass, which is why the IOExecutor stuff is in Executor.h/.cpp,
because it'll be pretty similar. The main exception is that for getCPUExecutor() you could return a default global
InlineExecutor instead of exploding as in getIOExecutor()

Test Plan: wangle unit, will start plumbing this into the services in #5003045 if we like it

Reviewed By: davejwatson@fb.com

Subscribers: hannesr, trunkagent, fugalh, alandau, mshneer, folly-diffs@, bmatheny

FB internal diff: D1727894

Tasks: 5002442

Signature: t1:1727894:1418344077:9e54088a6acb3f78e53011a32fd1dfe8b3214c1d

10 years agomove wangle/Executor.h to folly/ root
James Sedgwick [Mon, 15 Dec 2014 19:43:32 +0000 (11:43 -0800)]
move wangle/Executor.h to folly/ root

Summary:
this removes the dep EventBase has on wangle as we prepare to split off wangle
also changes namespace from folly::wangle to folly

Test Plan: just a couple of codemods so waiting for contbuild

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, chaoyc, search-fbcode-diffs@, zeus-diffs@, nli, cold-storage-diffs@, unicorn-diffs@, targeting-diff-backend@, hannesr, vighnesh, fugalh, alandau, bmatheny, adityab, zhuohuang, luk, darshan, gunjan, hdoshi, dzhulgakov, alihussains, panin, ves, mshneer, folly-diffs@, lins, nimishshah

FB internal diff: D1737376

Tasks: 5802833

Signature: t1:1737376:1418423430:82d219c34fcc50218c380a17435e5880e53db6bd

10 years agoImprove performance of stringPrintf and related functions
Chip Turner [Thu, 11 Dec 2014 06:14:09 +0000 (22:14 -0800)]
Improve performance of stringPrintf and related functions

Summary:
It turned out at least one optimization we were doing for
stringPrintf (using the tail of the input buffer) was causing a
performance degradation in some cases -- if the string was already
pre-sized, our resize() call would end up memset'ing the tail.  In some
instances, this could cause significant performance penalties, such as
when multiple stringAppendf calls were made.

So, this diff removes the "optimization" around using the tail of the input
buffer and instead uses a standalone stack buffer.  If vsnprintf deems
that buffer too small, a heap buffer is instead used.

As there is no std::string method that resizes the string to fill the
underlying buffer without setting the values to a default, and as it is
not legal to write past the end of the data buffer (even if capacity()
says there is enough), let's just abandon that optimization.  It turns
out this doesn't have a major performance loss for most cases since,
with this diff, most small strings will fit on-stack and then hopefully
in the string's tail anyway.

Test Plan: runtests

Reviewed By: simpkins@fb.com

Subscribers: trunkagent, net-systems@, lins, anca, folly-diffs@

FB internal diff: D1733774

Tasks: 5735468

Signature: t1:1733774:1418323943:ec47007c9756aca6cf0466bce92722ac4c7e89b2

10 years agoAdd convenience method for folly::join
Andrew Chalfant [Fri, 12 Dec 2014 02:18:26 +0000 (18:18 -0800)]
Add convenience method for folly::join

Summary: While folly has join methods that return a string (and thus don't need a string referenced passed), it does not support this idiom with iterators. This diff adds support for calls like folly::join("/", itr.begin(), itr.end() - 1).

Test Plan: Unit test

Reviewed By: njormrod@fb.com

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

FB internal diff: D1702892

Tasks: 5691439

Signature: t1:1702892:1418344206:9c1736f5d8e41be1df71a415e3803fe846b387b7

10 years agoAsyncUDPSocket
Dave Watson [Tue, 25 Nov 2014 16:13:37 +0000 (08:13 -0800)]
AsyncUDPSocket

Summary:
Move AsyncUDPSocket to folly.

There is also one under realtime/voip/async that looks functionaly equivalent?  I think this one is only used in gangplank currently.

Test Plan: contbuild

Reviewed By: hans@fb.com

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

FB internal diff: D1710675

Tasks: 5788116

Signature: t1:1710675:1417477000:9aebb466757554a5fa49d7c36cb504b4d8711b68

10 years agoImprove benchmarking around stringPrintf functions
Chip Turner [Thu, 11 Dec 2014 05:00:02 +0000 (21:00 -0800)]
Improve benchmarking around stringPrintf functions

Summary:
Before optimizing this code, it is better to measure different
aspects of it to ensure improvements are worth it and don'tintroduce
other regressions.

This adds a new benchmark as well as parameterizes an old one.

Test Plan: run it

Reviewed By: lovro@fb.com

Subscribers: lins, anca, folly-diffs@

FB internal diff: D1733760

Tasks: 5735468

Signature: t1:1733760:1418313479:9f572d5a4cf014bd266d91d0f7a75407d1514f65

10 years agoBump version to 18:0
Dave Watson [Thu, 11 Dec 2014 16:02:41 +0000 (08:02 -0800)]
Bump version to 18:0

10 years agoformatting nits in Executor.h
James Sedgwick [Thu, 11 Dec 2014 15:41:44 +0000 (07:41 -0800)]
formatting nits in Executor.h

Summary:
sorry this was bothering me
also remove extra include

Test Plan: OK

Reviewed By: davejwatson@fb.com

Subscribers: fugalh, folly-diffs@

FB internal diff: D1733723

Signature: t1:1733723:1418312030:088eb084f69e2bd7703724464b2f3e71abc2607a

10 years agofuture_* callbacks in ThreadManager thread
James Sedgwick [Thu, 11 Dec 2014 03:23:24 +0000 (19:23 -0800)]
future_* callbacks in ThreadManager thread

Summary:
This is a squashed diff consisting of the following approved diffs

D1714007
D1714645
D1715686
D1719963
D1720725
D1721838
D1721856
D1721920
D1721956
D1724910
D1728289

- execute future_* handlers in TM thread
- call async thrift clients from any thread, not just EB thread
- store a request's EventBase in the RequestContext for easier client management
* this last one will change in favor of using a global IO executor but this should still land as an intermediate step. See D1727894

Test Plan: see component diff test plans. otherwise, contbuild.

Reviewed By: davejwatson@fb.com

Subscribers: trunkagent, ruibalp, targeting-diff-backend@, hannesr, alandau, prometheus-diffs@, jeremyfein, mshneer, folly-diffs@, bmatheny, tingy

FB internal diff: D1732289

Tasks: 50030455645785

Signature: t1:1732289:1418253508:30b724a91717cf7fe21029e0a1eaf239db1650be

10 years agoConcurrentSkipList Bug
Nicholas Ormrod [Thu, 11 Dec 2014 01:41:17 +0000 (17:41 -0800)]
ConcurrentSkipList Bug

Summary:
Bug reported by Yan Lin (not a facebook employee) through @robbert.

@philipp and @jdelong: you are the only two remaining facebookers to have
made non-trivial changes to this code.

Description of bug: layer 0 is 1->4, and we're looking for 3. We pass
over 4, and see that 1 is less than 3 in the for loop. Now a race
condition: another thread inserts 2, so layer 0 is now 1->2->3. Then,
since ht==0, we return pred->skip(0), which is now 2 instead of 4.

Why this is bad: it really breaks the lower_bound function (lower_bound
calls findNode calls findNodeDownRight), since it returns an element
that is lesser.

This patch doesn't change the behavior of the code in the normal case;
it just recycles previously-computed values so that this race condition
doesn't crash and burn.

Patch based off of Yan Lin's in-email suggestion.

Test Plan:
fbconfig -r folly && fbmake runtests

Reviewed By: philipp@fb.com

Subscribers: sdwilsh, folly-diffs@, jdelong, robbert, philipp

FB internal diff: D1732434

Signature: t1:1732434:1418260198:8c707435825cfa2a1093b681e066f320193e98f2

10 years agoFix clang compilation
Brian Watling [Wed, 10 Dec 2014 22:39:53 +0000 (14:39 -0800)]
Fix clang compilation

Summary: Fix shadowed variable name

Test Plan: compile

Reviewed By: mmandal@fb.com

Subscribers: fugalh, mathieubaudet, folly-diffs@

FB internal diff: D1732033

Signature: t1:1732033:1418250864:d460fea376ac93c05b86977629b95dac60b8b916

10 years agoOnly test multi accept if reuse port supported
Dave Watson [Wed, 10 Dec 2014 22:34:00 +0000 (14:34 -0800)]
Only test multi accept if reuse port supported

Summary: Some contbuild machines aren't on 3.10 yet.

Test Plan:
fbconfig folly/experimental/wangle/bootstrap; fbmake runtests
tested on 3.2 and 3.10

Reviewed By: jsedgwick@fb.com

Subscribers: doug, fugalh, folly-diffs@

FB internal diff: D1732040

Tasks: 5800250

Signature: t1:1732040:1418250452:a3203ab5769494594f2bf0b50f1c62052ddeb24c

10 years agoSupport -Wsign-compare compilation
Sean Cannella [Wed, 10 Dec 2014 21:15:55 +0000 (13:15 -0800)]
Support -Wsign-compare compilation

Summary:
Due to how Conv.h uses the less_than template inside
not-actually-static static ifs clang still ends up validating generated
code inside a FOLLY_RANGE_CHECK which is never executed due to the
templated types. This code however still generates -Wsign-compare issues
so suppress that in order to allow includers to use this flag.

A simple example will illustrate this:

uint64_t foo = 1;
int bar = folly::to<int>(foo);

Test Plan: fbmake runtests

Reviewed By: meyering@fb.com, njormrod@fb.com

Subscribers: trunkagent, folly-diffs@, bmatheny, ranjeeth, subodh, kmdent, fma, shikong, pgriess, jdelong

FB internal diff: D1731411

Signature: t1:1731411:1418243200:ed1f34a1485669c9cb18f9f6029aca70e498953c

10 years agoUse an IO thread pool executor by default
Brian Watling [Wed, 10 Dec 2014 19:04:28 +0000 (11:04 -0800)]
Use an IO thread pool executor by default

Summary:
Remove member variable name shadowing

Facebook: This diff adds a default executor for async requests which do not specify one. This ensures we don't run request callbacks on the event threads (which in turn ensures we don't make blocking tcc calls from the event threads)

Test Plan: fbconfig tao/client; fbmake runtests

Reviewed By: hannesr@fb.com

Subscribers: hannesr, fugalh, zhuohuang, folly-diffs@, tao-eng@

FB internal diff: D1731078

Tasks: 5752037

Signature: t1:1731078:1418237801:5b9bbe74c288292d5e7c7e6a38e67be864627d89

10 years agoAdd mock singleton injection support to folly/experimental Singleton (t5653148).
Adrian Hamza [Wed, 10 Dec 2014 17:43:25 +0000 (09:43 -0800)]
Add mock singleton injection support to folly/experimental Singleton (t5653148).

Summary: Add mock singleton injection support to folly/experimental Singleton (t5653148).

Test Plan:
Added unit tests for MockSingleton, all unit tests passed.
Validated this works for my scenario.

Reviewed By: andrii@fb.com

Subscribers: trunkagent, njormrod, folly-diffs@

FB internal diff: D1690946

Tasks: 5653148

Signature: t1:1690946:1418080331:948d7051a5e5a2653dc393c123f188c56072c6db

10 years agoMulti accept
davejwatson [Wed, 10 Dec 2014 15:58:27 +0000 (07:58 -0800)]
Multi accept

Summary: Use acceptor pool instead of current thread for accepts.  Use reuse_port option to support multiple threads

Test Plan: Updated unittests

Reviewed By: hans@fb.com

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

FB internal diff: D1710619

Tasks: 5788110

Signature: t1:1710619:1417477350:eee5063186e582ef74c4802b8149563af029b3de

10 years agofix Cursor::skipAtMostSlow bug introduced in D1724111
Philip Pronin [Tue, 9 Dec 2014 21:43:33 +0000 (13:43 -0800)]
fix Cursor::skipAtMostSlow bug introduced in D1724111

Test Plan: fbconfig -r folly/io/test && fbmake runtests_opt -j32

Reviewed By: viswanath@fb.com

Subscribers: folly-diffs@

FB internal diff: D1728927

Tasks: 5788015

Signature: t1:1728927:1418161580:9ed66412e35c62486a56e19aa9bf119863b5f1a1

Blame Revision: D1724111

10 years agoAdd REUSEPORT option to AsyncServerSocket
Dave Watson [Mon, 1 Dec 2014 20:52:35 +0000 (12:52 -0800)]
Add REUSEPORT option to AsyncServerSocket

Summary:
Adds a reuse port option to AsyncServerSocket, so multiple sockets can bind to the same accept port.  Allows for multiple accept threads, so accepts can be greater, since there is no longer a single accept lock.

reuse port option is ifdefd, since this wouldn't build with some of the older kernels otherwise.

Postponed overnight

Test Plan: Builds.  Used in an upcoming diff.

Reviewed By: jsedgwick@fb.com

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

FB internal diff: D1710600

Tasks: 54885165788110

Signature: t1:1710600:1418066966:627e03857f9b5ff831f2922add08e90cc525c95c

10 years agoTransportInfo size mismatches
Sean Cannella [Tue, 9 Dec 2014 00:25:48 +0000 (16:25 -0800)]
TransportInfo size mismatches

Summary:
The code that uses TransportInfo assumes these byte counts are
compatible with size_t so make them so.

Test Plan: compiled

Reviewed By: davejwatson@fb.com

Subscribers: fugalh, bmatheny, njormrod, folly-diffs@, benyluo, ranjeeth, fma, kmdent

FB internal diff: D1725026

Tasks: 5538843

Signature: t1:1725026:1418061933:b4072ca926e792716776ddc00d79db9376510859

10 years agoset internal thread factory
Dave Watson [Mon, 24 Nov 2014 20:36:30 +0000 (12:36 -0800)]
set internal thread factory

Summary: Fix up a TODO

Test Plan: ??  Should I bother with a test?

Reviewed By: jsedgwick@fb.com

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

FB internal diff: D1710633

Signature: t1:1710633:1417469890:9e8513140bae79da384c9d5418d4ea0840efbe6c

10 years agoSet a better default for maxReadsPerEvent
Dave Watson [Mon, 8 Dec 2014 18:25:14 +0000 (10:25 -0800)]
Set a better default for maxReadsPerEvent

Summary:
0 (infinity) seems like a bad default for this.  Just found a 2~5% latency win in thrift by setting this to something more reasonable.  Any reason not to set a better default globally?

Since this is already on for mcrouter/proxygen/thrift, the biggest change would probably be service router

tests postponed longer than 20 minutes

Test Plan:
thrift perf test, multifeed canary

Canary URL: https://our.intern.facebook.com/intern/feedtools/mycanary?id=davejwatson_20141208-102635_local.1418063194_d201eceec04e5e097a6dffb645a424d502db8742_

Reviewed By: afrind@fb.com

Subscribers: doug, net-systems@, njormrod, folly-diffs@, jsedgwick, haijunz, andrewcox, alandau, fugalh

FB internal diff: D1725142

Signature: t1:1725142:1418066284:c2e1cbf27b0c7a98cf2395f369683bbcce4cdce4

10 years agoBump up log level
Dave Watson [Mon, 1 Dec 2014 20:55:49 +0000 (12:55 -0800)]
Bump up log level

Summary: This log was getting annoying in tests, bump up the logging level

Test Plan: contbuild

Reviewed By: afrind@fb.com

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

FB internal diff: D1710608

Signature: t1:1710608:1417468546:770dfb7201ff6477b349c01e638f704a2ef1b7ce

10 years agofix local variable shadow
Shijin Kong [Mon, 8 Dec 2014 18:03:30 +0000 (10:03 -0800)]
fix local variable shadow

Summary:
D1720014 line 183 in HHWheelTimer.h broke proxygen/facebook/httpclient clang build:

./folly/io/async/HHWheelTimer.h:183:17: error: declaration shadows a local variable [-Werror,-Wshadow]
Wrapper(F fn) : fn_(std::move(fn)) {}
^
./folly/io/async/HHWheelTimer.h:181:28: note: previous declaration is here
void scheduleTimeoutFn(F fn, std::chrono::milliseconds timeout) {

I renamed the `fn` in `Wrapper` to `f`.

Test Plan: pass

Reviewed By: seanc@fb.com

Subscribers: njormrod, folly-diffs@, subodh, seanc, kmdent, fma

FB internal diff: D1724835

Signature: t1:1724835:1418052896:d9a72230d4fd1a4a734a33142966fe796f4fd362

10 years agoimprove io::Cursor read() performance for small sizeof(T)
Philip Pronin [Sun, 7 Dec 2014 00:49:11 +0000 (16:49 -0800)]
improve io::Cursor read() performance for small sizeof(T)

Summary:
I just found that gcc (4.8.2) failed to unroll the loop in
`pullAtMost()`, so it didn't replace `memcpy` with a simple load
for small `len`.

Test Plan:
fbconfig -r folly/io/test thrift/lib/cpp2/test && fbmake runtests_opt -j32

Ran unicorn-specific thrift deserialization benchmark from
D1724070, verified 50% improvement in `SearchRequest` deserialization
performance.

`thrift/lib/cpp2/test/ProtocolBench` results:

```
|---- before -----| |---- after  -----|
================================================================================================
thrift/lib/cpp2/test/ProtocolBench.cpp          relative  time/iter  iters/s  time/iter  iters/s
================================================================================================
BinaryProtocol_read_Empty                                   21.72ns   46.04M    17.58ns   56.89M
BinaryProtocol_read_SmallInt                                43.03ns   23.24M    23.64ns   42.30M
BinaryProtocol_read_BigInt                                  43.72ns   22.87M    22.03ns   45.38M
BinaryProtocol_read_SmallString                             88.57ns   11.29M    47.01ns   21.27M
BinaryProtocol_read_BigString                              365.76ns    2.73M   323.58ns    3.09M
BinaryProtocol_read_BigBinary                              207.78ns    4.81M   169.09ns    5.91M
BinaryProtocol_read_LargeBinary                            187.81ns    5.32M   172.09ns    5.81M
BinaryProtocol_read_Mixed                                  161.18ns    6.20M    68.41ns   14.62M
BinaryProtocol_read_SmallListInt                           177.32ns    5.64M    96.91ns   10.32M
BinaryProtocol_read_BigListInt                              77.03us   12.98K    15.88us   62.97K
BinaryProtocol_read_BigListMixed                             1.79ms   557.79   923.99us    1.08K
BinaryProtocol_read_LargeListMixed                         195.01ms     5.13   103.78ms     9.64
================================================================================================
```

Reviewed By: soren@fb.com

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

FB internal diff: D1724111

Tasks: 5770136

Signature: t1:1724111:1417977810:b7d643d0c819a0bbac77fa0048206153929e50a8

10 years agoMake ThreadLocal identifiers consistent
Sean Cannella [Sat, 6 Dec 2014 01:14:52 +0000 (17:14 -0800)]
Make ThreadLocal identifiers consistent

Summary:
Implicitly bouncing between size_t and int for identifiers
causes problems when building on 64-bit iOS. Fix this.

Test Plan: fbmake runtests

Reviewed By: meyering@fb.com, njormrod@fb.com

Subscribers: trunkagent, njormrod, folly-diffs@, fma, kmdent, shikong, pgriess

FB internal diff: D1722061

Signature: t1:1722061:1417819039:8e3938cf8d4d241551ed3dd3978c1b11f57398c5

10 years agoRevert "Revert "Make folly::Singleton's destruction happen earlier""
Andrii Grynenko [Fri, 5 Dec 2014 22:28:30 +0000 (14:28 -0800)]
Revert "Revert "Make folly::Singleton's destruction happen earlier""

Summary:
This reverts commit 7871df125fc1fbe5d8030a41e21b8241b25d4e62.

As explained in t5763478, I don't see good reason for a revert.
Also the errors under question shouldn't show up after D1720841.

Test Plan: revert-hammer

Reviewed By: stepan@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1722928

Tasks: 5763478

Signature: t1:1722928:1417818663:1fe402d14c0cf389fb3611355db270c5f2c33738

10 years agoMore logging in HHWheelTimer::scheduleTimeoutFn
Hans Fugal [Fri, 5 Dec 2014 17:46:56 +0000 (09:46 -0800)]
More logging in HHWheelTimer::scheduleTimeoutFn

Summary:
This was supposed to be an update to D1720014 but I screwed up with arc.

Test Plan: run the test by hand and see the output

Reviewed By: jsedgwick@fb.com

Subscribers: exa, njormrod, folly-diffs@

FB internal diff: D1721951

Signature: t1:1721951:1417801009:72a27ec2ca473a996785ff9cae48bd51a43d9045

10 years agos/makeFuture().via/via/
Hans Fugal [Fri, 5 Dec 2014 17:33:35 +0000 (09:33 -0800)]
s/makeFuture().via/via/

Summary:
codemod to replace all `makeFuture().via` with just `via`.

Test Plan: contbuild

Reviewed By: jsedgwick@fb.com

Subscribers: trunkagent, zeus-diffs@, targeting-diff-backend@, abirchall, fugalh, msk, exa, benj, njormrod, folly-diffs@

FB internal diff: D1715876

Signature: t1:1715876:1417712302:b882916d394f90caa23bd73fa68b8f786af649e3

10 years agovia and activate/deactivate chaining
Hans Fugal [Fri, 5 Dec 2014 17:33:30 +0000 (09:33 -0800)]
via and activate/deactivate chaining

Summary:
Better support and test chaining of `via` and `activate`/`deactivate`.

The real problem is that without the ref qualifier a multiple chain of lvalue references can end up with a destructed object being referenced after.

https://akrzemi1.wordpress.com/2014/06/02/ref-qualifiers/

Test Plan:
This is mostly new tests that would fail and now pass.
I think maybe the tests are a bit weak, it would be good to find a way to ensure we aren't going to see the access-after-free bugs in these tests.

Reviewed By: jsedgwick@fb.com

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

FB internal diff: D1714873

Tasks: 5489801

Signature: t1:1714873:1417628538:9e610c5ba5e0a22c19a11d53aa956be45d585058

10 years agoRemove Later
Hans Fugal [Fri, 5 Dec 2014 17:33:25 +0000 (09:33 -0800)]
Remove Later

Summary: dalek-exterminate

Test Plan:
Moved `Later` tests to `via` tests.
fbgs
contbuild

Reviewed By: jsedgwick@fb.com

Subscribers: fbcode-common-diffs@, adityab, lins, trunkagent, fugalh, exa, njormrod, folly-diffs@, hannesr

FB internal diff: D1714862

Tasks: 5409538

Signature: t1:1714862:1417621949:f63f49e1093a021170d2346e8e673db042d2bc56

10 years agoHHWheelTimer::scheduleTimeoutFn
Hans Fugal [Fri, 5 Dec 2014 17:29:04 +0000 (09:29 -0800)]
HHWheelTimer::scheduleTimeoutFn

Summary:
cpp11ify

I'm gonna use this for Wangle timeouts.

Test Plan:
new unit test
_bin/folly/io/async/test/HHWheelTimerTest

Reviewed By: jsedgwick@fb.com

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

FB internal diff: D1720014

Tasks: 5002969

Signature: t1:1720014:1417732753:ddcad431b6c305c9ba27f6b9e3454ab158c13f4b

10 years agoRevert "Make folly::Singleton's destruction happen earlier"
Bartosz Nitka [Fri, 5 Dec 2014 13:30:32 +0000 (05:30 -0800)]
Revert "Make folly::Singleton's destruction happen earlier"

Summary: This reverts commit 6584412293cc5b9aad2004e99d550e3478df1e5d.

Test Plan: thisisrevert

Reviewed By: smarlow@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1721739

Signature: t1:1721739:1417786147:01aaaf50b679da7942b2a30ad039eb0ac1031430

10 years agoinclude gflags header in SSLSessionCacheManager
Nicholas Ormrod [Thu, 4 Dec 2014 22:08:43 +0000 (14:08 -0800)]
include gflags header in SSLSessionCacheManager

Summary:
SSLSessionCacheManager uses glfags but doesn't include it. This
breaks on some systems.

Test Plan: fbconfig -r folly && fbmake runtests

Reviewed By: robbert@fb.com

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

FB internal diff: D1716997

Signature: t1:1716997:1417720032:548c2ad8f54e41d42585c331324f2065733595c1

10 years agoMake folly::Singleton's destruction happen earlier
Andrii Grynenko [Thu, 4 Dec 2014 01:26:45 +0000 (17:26 -0800)]
Make folly::Singleton's destruction happen earlier

Summary: Schedule destroyInstances to be executed via atexit in registrationComplete. registrationComplete is called from main(), so this makes sure folly::Singleton's will be destoyed before all singletons constructed before main() (which should cover all third-party libraries' singletons).

Test Plan: unit test

Reviewed By: chip@fb.com

Subscribers: trunkagent, njormrod, folly-diffs@

FB internal diff: D1717963

Tasks: 5666654

Signature: t1:1717963:1417659478:c50c271d1786be75499565f6ab9c0a1a6f6f347d

10 years agoCorrect getsockopt call made by AsyncSocket
Andre Pinto [Thu, 4 Dec 2014 19:52:53 +0000 (11:52 -0800)]
Correct getsockopt call made by AsyncSocket

Summary:
The getsockopt's last parameter (optlen) is a value-result parameter
and AsyncSocket::getSockOpt was passing a value as argument.

Test Plan: Unit tests

Reviewed By: alikhtarov@fb.com

Subscribers: trunkagent, njormrod, folly-diffs@

FB internal diff: D1717463

Tasks: 4867290

Signature: t1:1717463:1417664828:6c7a74ff31725121f892ce1adba2653e70728192

10 years agoadd consumeUntilDrained API to NotificationQueue::Consumer
James Sedgwick [Thu, 4 Dec 2014 18:29:24 +0000 (10:29 -0800)]
add consumeUntilDrained API to NotificationQueue::Consumer

Summary:
... and employ it in EventBase's destructor to stop leaking memory from unexecuted NotificationQueue-variety runInEventBaseThread() callbacks

in addition to the attached task, this should also fix the root cause of the already-worked-around #5564342

Test Plan:
test no longer leaks.* on the other hand, IOThreadPoolExecutor::stop() no longer actually stops, it joins. I added comments to stop() in headers indicating that it should be treated as best-effort.

* hilariously, this actually isn't true - there's a new leak of a thread local inside the io pool because callbacks can outlive the pool now. i'll leave a full explanation for the upcoming patch.

Reviewed By: davejwatson@fb.com

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

FB internal diff: D1682860

Tasks: 5336655

Signature: t1:1682860:1416347774:ac5f31fb72373992f425c93ac284a0cd27608db4

10 years agoReplace Later in tcc
Hans Fugal [Wed, 3 Dec 2014 17:38:47 +0000 (09:38 -0800)]
Replace Later in tcc

Summary:
s/`sendAsyncLater`/`sendAsyncVia`/ and change it to return a cold future with equivalent semantics. We have to call `via` twice in the implementation—once to gate the result and again to make sure the caller will get a future that executes in that executor's context also.

NB this is a slight semantic change - if the executor is, say, a threadpool then now it will go through the threadpool queue (and maybe execute in two different threads) whereas before the whole later chain would execute in the same thread. I don't *think* this is a problem, but something to think about.

Test Plan:
stuff we fbconfig'd builds
contbuild
unit tests
thinking really hard

Reviewed By: hannesr@fb.com

Subscribers: trunkagent, fbcode-common-diffs@, net-systems@, ldbrandy, hannesr, fugalh, zhuohuang, exa, watashi, smarlow, akr, bnitka, jcoens, darshan, njormrod, anfarmer, folly-diffs@

FB internal diff: D1644012

Tasks: 5409538

Signature: t1:1644012:1417625401:99b1b7df6de4cfcdd945eed7104d4c82e8c0b78f

10 years agoFixing static initilize fiasco in Benchmark.cpp
Maxim Sokolov [Tue, 2 Dec 2014 00:03:36 +0000 (16:03 -0800)]
Fixing static initilize fiasco in Benchmark.cpp

Summary:
Benchmark.cpp code is dependent on initialization order, which leads
to benchmarks not running. The root cause is that on some linkers the code which
adds benchmarks to the benchmarks vector is run before the vector initialization.

The original bug report: https://github.com/facebook/folly/issues/18
See: http://www.parashift.com/c++-faq-lite/static-init-order.html

Test Plan: run benchmark_test and compare result with master output

Reviewed By: andrei.alexandrescu@fb.com, njormrod@fb.com

Subscribers: trunkagent, sdwilsh, folly-diffs@

FB internal diff: D1710588

Signature: t1:1710588:1417468175:fd2705a573cef5c4ff020b60c6aec3d83bcdbbab

10 years agoReadd unit test
Andrei Bajenov [Mon, 1 Dec 2014 21:52:38 +0000 (13:52 -0800)]
Readd unit test

Summary:
Description:
Last time a merge error broke stuff, fixing and readding

Test Plan:
Ran unit test

This reverts commit 1075ab9d6f62cad36a228bd3fc7203a75fdf5baf.

Reviewed By: davejwatson@fb.com

Subscribers: njormrod, folly-diffs@, wstefancik

FB internal diff: D1695920

Signature: t1:1695920:1416529609:3f726ac5df4e33f254075738ea6655a6fc01bb7f

10 years agoFix dynamic::insert
Pavlo Kushnir [Thu, 27 Nov 2014 03:28:18 +0000 (19:28 -0800)]
Fix dynamic::insert

Summary: folly::dynamic::insert is broken. When we try to insert an rvalue and the key already exists in dynamic, insert will replace the value with empty dynamic.

Test Plan: unit test attached

Reviewed By: stepan@fb.com

Subscribers: trunkagent, njormrod, folly-diffs@

FB internal diff: D1704995

Signature: t1:1704995:1417053631:8e0163693df721c54b016f8c27e81e7bf60194cb

10 years agofix varint decoding in case the first buffer is smaller than encoded varint
Mainak Mandal [Wed, 26 Nov 2014 02:27:13 +0000 (18:27 -0800)]
fix varint decoding in case the first buffer is smaller than encoded varint

Summary: This is not a super clean solution, given that I have just copied the code over form Varint.h. Another way would be to add a peek(IOBuf&, size_t) method to CursorBase and use that to peek deeper into the chain.

Test Plan: unit tests

Reviewed By: tulloch@fb.com

Subscribers: trunkagent, njormrod, dcapel, benr, folly-diffs@

FB internal diff: D1689872

Signature: t1:1689872:1416964989:b7f12a2686233f161401288ffcb8c51926b01fdf

10 years agoIOBuf::moveToFbString legacy 4GiB limit
Nicholas Ormrod [Wed, 26 Nov 2014 01:42:05 +0000 (17:42 -0800)]
IOBuf::moveToFbString legacy 4GiB limit

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

10 years agozlib compression fails on large IOBufs
Nicholas Ormrod [Wed, 26 Nov 2014 00:36:20 +0000 (16:36 -0800)]
zlib compression fails on large IOBufs

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

10 years agoMake SSLContext's exit() handling more graceful
Subodh Iyengar [Tue, 25 Nov 2014 21:07:16 +0000 (13:07 -0800)]
Make SSLContext's exit() handling more graceful

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

10 years agofix Makefile.am
James Sedgwick [Fri, 21 Nov 2014 23:23:56 +0000 (15:23 -0800)]
fix Makefile.am

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

10 years agofix build
Dave Watson [Fri, 21 Nov 2014 23:06:53 +0000 (15:06 -0800)]
fix build

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

10 years agoChannelPipeline tests and fixes
James Sedgwick [Fri, 21 Nov 2014 21:06:05 +0000 (13:06 -0800)]
ChannelPipeline tests and fixes

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

10 years agoAsyncSocketHandler
James Sedgwick [Fri, 21 Nov 2014 20:59:49 +0000 (12:59 -0800)]
AsyncSocketHandler

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

10 years agotest for OutputBufferingHandler
James Sedgwick [Fri, 21 Nov 2014 19:55:49 +0000 (11:55 -0800)]
test for OutputBufferingHandler

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

10 years agomodifiable channel pipelines
James Sedgwick [Fri, 21 Nov 2014 19:30:14 +0000 (11:30 -0800)]
modifiable channel pipelines

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

10 years agoAdd process group leader option
Alexey Spiridonov [Tue, 12 Aug 2014 19:24:40 +0000 (12:24 -0700)]
Add process group leader option

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

10 years agoFix build break
Nicholas Ormrod [Thu, 20 Nov 2014 19:06:39 +0000 (11:06 -0800)]
Fix build break

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

10 years agomerge event base tests
James Sedgwick [Thu, 20 Nov 2014 18:02:13 +0000 (10:02 -0800)]
merge event base tests

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

10 years agoBump version to 17:0
Dave Watson [Thu, 20 Nov 2014 17:32:17 +0000 (09:32 -0800)]
Bump version to 17:0

10 years agofolly/io/async documentation
Dave Watson [Wed, 19 Nov 2014 17:41:06 +0000 (09:41 -0800)]
folly/io/async documentation

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

10 years agoHHWheelTimer - fix scheduling timeouts within callbacks
Andrei Bajenov [Thu, 20 Nov 2014 01:47:07 +0000 (17:47 -0800)]
HHWheelTimer - fix scheduling timeouts within callbacks

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

10 years agoBump version to 16:0
Dave Watson [Wed, 19 Nov 2014 20:54:03 +0000 (12:54 -0800)]
Bump version to 16:0

10 years agoOS merges
Dave Watson [Mon, 10 Nov 2014 18:14:41 +0000 (10:14 -0800)]
OS merges

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

10 years agoserver bootstrap
Dave Watson [Tue, 28 Oct 2014 21:46:19 +0000 (14:46 -0700)]
server bootstrap

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

10 years agoRun the HHWheelTimer and EventBase tests
Daniel Sommermann [Tue, 18 Nov 2014 22:15:57 +0000 (14:15 -0800)]
Run the HHWheelTimer and EventBase tests

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

10 years agoadd custom getLoad() for multifeed leaf
Edward Chin [Tue, 18 Nov 2014 21:21:00 +0000 (13:21 -0800)]
add custom getLoad() for multifeed leaf

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

10 years agoClean up runBeforeLoop API
Dave Watson [Thu, 13 Nov 2014 20:51:43 +0000 (12:51 -0800)]
Clean up runBeforeLoop API

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

10 years agoMove some tests to folly
Daniel Sommermann [Thu, 13 Nov 2014 18:49:31 +0000 (10:49 -0800)]
Move some tests to folly

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

10 years agotop-level via()
James Sedgwick [Tue, 18 Nov 2014 18:05:10 +0000 (10:05 -0800)]
top-level via()

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

10 years agoImprove initialization of openssl
subodh [Sun, 16 Nov 2014 01:17:53 +0000 (17:17 -0800)]
Improve initialization of openssl

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

10 years agoFix build break
Nicholas Ormrod [Mon, 17 Nov 2014 23:47:05 +0000 (15:47 -0800)]
Fix build break

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

10 years agoIndefinite observers and inline observers
James Sedgwick [Mon, 17 Nov 2014 23:26:00 +0000 (15:26 -0800)]
Indefinite observers and inline observers

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

10 years agoOutputBufferingHandler
James Sedgwick [Mon, 17 Nov 2014 23:23:18 +0000 (15:23 -0800)]
OutputBufferingHandler

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

10 years agoCompactProtocol: readVarint optimizations
Alex Landau [Fri, 14 Nov 2014 04:50:30 +0000 (20:50 -0800)]
CompactProtocol: readVarint optimizations

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

10 years agoimplicit constructor exception_wrapper(std::exception)
James Sedgwick [Mon, 17 Nov 2014 23:04:23 +0000 (15:04 -0800)]
implicit constructor exception_wrapper(std::exception)

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

10 years agofix potential race/memory corruption in IOThreadPoolExecutor
James Sedgwick [Mon, 17 Nov 2014 23:04:35 +0000 (15:04 -0800)]
fix potential race/memory corruption in IOThreadPoolExecutor

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

10 years agopriority CPU thread pool
James Sedgwick [Mon, 17 Nov 2014 22:44:10 +0000 (14:44 -0800)]
priority CPU thread pool

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

10 years agoFix format bug
Dave Watson [Thu, 13 Nov 2014 20:31:47 +0000 (12:31 -0800)]
Fix format bug

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

10 years agoMove Acceptor to wangle
Dave Watson [Thu, 23 Oct 2014 17:26:08 +0000 (10:26 -0700)]
Move Acceptor to wangle

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

10 years agofix potential memory leak in ThreadLocal
Philip Pronin [Fri, 14 Nov 2014 22:00:47 +0000 (14:00 -0800)]
fix potential memory leak in ThreadLocal

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

10 years agoCompactProtocol: more improvements
Alex Landau [Thu, 13 Nov 2014 01:19:15 +0000 (17:19 -0800)]
CompactProtocol: more improvements

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

10 years agofolly::Singleton-based McrouterManager
Andrii Grynenko [Tue, 11 Nov 2014 19:13:30 +0000 (11:13 -0800)]
folly::Singleton-based McrouterManager

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

10 years agoHave EventBase implement wangle::Executor
Hans Fugal [Wed, 12 Nov 2014 18:01:50 +0000 (10:01 -0800)]
Have EventBase implement wangle::Executor

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

10 years agofolly::EventBase: wrap libevent calls to prevent race-condition
Nick Burrett [Wed, 12 Nov 2014 13:30:43 +0000 (05:30 -0800)]
folly::EventBase: wrap libevent calls to prevent race-condition

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

10 years agoUse folly::IPAddress::hash instead of gethostid
Alecs King [Tue, 11 Nov 2014 21:53:09 +0000 (13:53 -0800)]
Use folly::IPAddress::hash instead of gethostid

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

10 years agoClean up RequestContext
Dave Watson [Thu, 6 Nov 2014 17:59:27 +0000 (09:59 -0800)]
Clean up RequestContext

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

10 years agoOS merges
Dave Watson [Mon, 10 Nov 2014 18:14:41 +0000 (10:14 -0800)]
OS merges

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

10 years agoAdd optional parameter pruneHook to EvictingCacheMap::set(..)
Dmitry Panin [Sat, 8 Nov 2014 01:31:02 +0000 (17:31 -0800)]
Add optional parameter pruneHook to EvictingCacheMap::set(..)

Summary:
Inside `set()` we can do pruning, but it will happen
with default pruneHook.
Adding it as an optional param makes API more convenient.
(Instead, the users of API could just call `setPruneHook(pruneHook)` before `set`,
and then `setPruneHook(nullptr)` afterwards -- but it looks too ugly)

Test Plan:
```
fbconfig -r folly/ && fbmake runtests
```
passes:
```
Summary (total time 60.11s):
PASS: 1758
FAIL: 0
SKIP: 0
FATAL: 0
TIMEOUT: 0
```

Reviewed By: njormrod@fb.com

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

FB internal diff: D1665690

Tasks: 5551091

Signature: t1:1665690:1415391406:e4d2a956f9212aed70ab518159dbb19553764ce4

10 years agoBump version to 15:0
Pavlo Kushnir [Sat, 8 Nov 2014 02:43:32 +0000 (18:43 -0800)]
Bump version to 15:0

10 years agoFix folly lint errors
Nicholas Ormrod [Fri, 31 Oct 2014 17:48:19 +0000 (10:48 -0700)]
Fix folly lint errors

Summary:
Fix EOF whitespace
for file in `find folly -type f` ; do if [ -z "`tail -n1 $file`" ] ; then sed -i '$d' $file ; fi ; done

Test Plan: unit tests

Reviewed By: robbert@fb.com

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

FB internal diff: D1644130

Tasks: 5486739

Signature: t1:1644130:1414715392:b6c783851aa030ad1148f84a98139a5dca207da0

10 years agoisDetachable for HHWheelTimer
Misha Shneerson [Fri, 7 Nov 2014 20:33:14 +0000 (12:33 -0800)]
isDetachable for HHWheelTimer

Summary: title

Test Plan: unit tests

Reviewed By: andrei.bajenov@fb.com

Subscribers: trunkagent, mcduff, hitesh, alandau, bmatheny, njormrod, mshneer, folly-diffs@, andreib, davejwatson

FB internal diff: D1666499

Tasks: 5563183

Signature: t1:1666499:1415347256:d58c8bbe952385c1c96f7f8cc6ae7f02216c56bb

10 years agoMove SSL socket to folly
Dave Watson [Tue, 14 Oct 2014 18:10:48 +0000 (11:10 -0700)]
Move SSL socket to folly

Summary: One of the last thrift -> folly moves.  The only change was the exception types - there are small wrapper classes in thrift/lib/cpp/async left to convert from AsyncSocketException to TTransportException.

Test Plan: run unit tests

Reviewed By: dcsommer@fb.com

Subscribers: jdperlow, trunkagent, doug, bmatheny, ssl-diffs@, njormrod, mshneer, folly-diffs@, fugalh, jsedgwick, andrewcox, alandau

FB internal diff: D1632425

Signature: t1:1632425:1414526483:339ae107bacb073bdd8cf0942fd0f6b70990feb4

10 years agoRemove volatile from MemoryMappingTest
Tom Jackson [Fri, 7 Nov 2014 01:20:41 +0000 (17:20 -0800)]
Remove volatile from MemoryMappingTest

Summary: Not needed, I don't know why I put them there before.

Test Plan: Run the unit test

Reviewed By: njormrod@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1665689

Tasks: 5487902

Signature: t1:1665689:1415323144:0998e7f700a3b40652615a36c3b9c9f661fbdadf

10 years agoAdd some SocketAddress tests
Daniel Sommermann [Thu, 6 Nov 2014 21:44:13 +0000 (13:44 -0800)]
Add some SocketAddress tests

Summary: There was a bit of gap in test coverage for bracketed ipv6

Test Plan: unit tests

Reviewed By: viswanath@fb.com

Subscribers: doug, ps, bmatheny, njormrod, folly-diffs@

FB internal diff: D1664783

Signature: t1:1664783:1415316694:b17e0dc1fcfece06c6e04e5f65e2095c66d28cc4

10 years agoFix integer overflow in folly/Benchmark.cpp
Dmitry Panin [Thu, 6 Nov 2014 20:47:43 +0000 (12:47 -0800)]
Fix integer overflow in folly/Benchmark.cpp

Summary:
`bm_max_secs` has type int32
So when this value is more than 2, expression
```
FLAGS_bm_max_secs * 1000000000;
```
overflows

Test Plan:
I did put a
```
LOG(INFO) << timeBudgetInNs;
```
and observed correct values (before fix they were overflowed)

Reviewed By: antonl@fb.com

Subscribers: njormrod, folly-diffs@

FB internal diff: D1663247

Signature: t1:1663247:1415261814:c9154ffde183b2a4f5403e534e47e52e8276c61b

10 years agoBMI1 support in EliasFanoCoding
Philip Pronin [Tue, 4 Nov 2014 18:21:34 +0000 (10:21 -0800)]
BMI1 support in EliasFanoCoding

Summary:
This diff updates `folly::CpuId` with support of extended features (EAX =
7, ECX = 0) to provide detection logic for BMI1 introduced in Haswell, and
provides support for `BLSR` instruction in `EliasFanoReader`.

Test Plan:
I used clang to compile the logic and run unittests

Reviewed By: lucian@fb.com

Subscribers: fbcode-common-diffs@, trunkagent, chaoyc, search-fbcode-diffs@, unicorn-diffs@, njormrod, folly-diffs@

FB internal diff: D1658100

Signature: t1:1658100:1415126635:d1820b8eb41c9e9786b5c8062b801cf1e2049a97

10 years ago(wangle) fix a race condition in Core::maybeCallback
Hans Fugal [Wed, 5 Nov 2014 23:44:06 +0000 (15:44 -0800)]
(wangle) fix a race condition in Core::maybeCallback

Summary:
`calledBack_` could be seen as true by both threads in this conditional. Classic rookie mistake. :-/

Test Plan: run unit tests

Reviewed By: darshan@fb.com

Subscribers: trunkagent, hannesr, net-systems@, fugalh, exa, njormrod, folly-diffs@

FB internal diff: D1661199

Tasks: 55429385506504

Signature: t1:1661199:1415215840:fb69f56c8cf6f59beeca809724ce015b5260d9ad

Blame Revision: D1636487

10 years agoRemove memory leak from readRandomDevice by using a raw fd
Brian Watling [Wed, 5 Nov 2014 23:17:10 +0000 (15:17 -0800)]
Remove memory leak from readRandomDevice by using a raw fd

Summary: Switch from dynamically allocating a File to using a raw fd for readRandomDevice. This prevents races at shutdown while also eliminating the memory leak

Test Plan: run unit tests

Reviewed By: meyering@fb.com

Subscribers: njormrod, folly-diffs@, tao-eng@

FB internal diff: D1662151

Signature: t1:1662151:1415229242:525b6294b27bb68b5dda70aadb8d3ba1cc61b815

10 years agoFix crashes in readRandomDevice()
Brian Watling [Wed, 5 Nov 2014 22:50:37 +0000 (14:50 -0800)]
Fix crashes in readRandomDevice()

Summary: The comment in readRandomDevice() lies - the PCHECK fails with "Bad file descriptor" in some tests. This is most likely due to a race where a background thread is calling readRandomDevice() while the main thread shuts down. Another possibility is bad destruction order, where rnadomDevice is destroyed before some other static object whose destructor calls readRandomDevice(). Either way, this fixes it without having to chace down every instance.

Test Plan: run unit tests

Reviewed By: meyering@fb.com

Subscribers: antonl, trunkagent, meyering, simpkins, njormrod, folly-diffs@, tao-eng@

FB internal diff: D1660903

Signature: t1:1660903:1415215895:7f1f96ab7416b45c4dc18e78173fb59eb27bd03b

10 years ago(wangle) fix a race in whenAll
Hans Fugal [Wed, 5 Nov 2014 19:21:49 +0000 (11:21 -0800)]
(wangle) fix a race in whenAll

Summary:
Race in `++ctx->count == ctx->total`. This ordering, while not very obvious or likely, is possible:

T1            T2
--            --
++ctx->count
++ctx->count
ctx->total
==
setValue
delete ctx
ctx->total
==
setValue
delete ctx

Test Plan:
That's the idea, anyway. I need some sleep, and it takes 20 minutes to build and test. I had a more convoluted fix (using `shared_ptr`) and it did seem to fix the error we were seeing, but I was seeing another error.

Reviewed By: darshan@fb.com

Subscribers: trunkagent, net-systems@, fugalh, exa, njormrod, folly-diffs@

FB internal diff: D1660663

Tasks: 5506504

Signature: t1:1660663:1415207632:49dc224363cec27736fc71fb211fa846be9e170f

Blame Revision: D1636487