Summary:
The call stack used to look like this during static initialization:
#0 0x00000000008032d1 in rocksdb::ThreadLocalPtr::StaticMeta::StaticMeta() (this=0x7ffff683b300) at util/thread_local.cc:172
#1 0x00000000008030a7 in rocksdb::ThreadLocalPtr::Instance() () at util/thread_local.cc:135
#2 0x000000000080310f in rocksdb::ThreadLocalPtr::StaticMeta::Mutex() () at util/thread_local.cc:141
#3 0x0000000000803103 in rocksdb::ThreadLocalPtr::StaticMeta::InitSingletons() () at util/thread_local.cc:139
#4 0x000000000080305d in rocksdb::ThreadLocalPtr::InitSingletons() () at util/thread_local.cc:106
It involves outer/inner classes and the call stacks goes
outer->inner->outer->inner, which is too difficult to understand. We can avoid
a level of back-and-forth by skipping StaticMeta::InitSingletons(), which
doesn't initialize anything beyond what ThreadLocalPtr::Instance() already
initializes.
Now the call stack looks like this during static initialization:
#0 0x00000000008032c5 in rocksdb::ThreadLocalPtr::StaticMeta::StaticMeta() (this=0x7ffff683b300) at util/thread_local.cc:170
#1 0x00000000008030a7 in rocksdb::ThreadLocalPtr::Instance() () at util/thread_local.cc:135
#2 0x000000000080305d in rocksdb::ThreadLocalPtr::InitSingletons() () at util/thread_local.cc:106
Test Plan:
unit tests
verify StaticMeta::mutex_ is still initialized in DefaultEnv() (StaticMeta::mutex_ is the only variable intended to be initialized via StaticMeta::InitSingletons() which I removed)
#0 0x00000000005cee17 in rocksdb::port::Mutex::Mutex(bool) (this=0x7ffff69500b0, adaptive=false) at port/port_posix.cc:52
#1 0x0000000000769cf8 in rocksdb::ThreadLocalPtr::StaticMeta::StaticMeta() (this=0x7ffff6950000) at util/thread_local.cc:168
#2 0x0000000000769a53 in rocksdb::ThreadLocalPtr::Instance() () at util/thread_local.cc:133
#3 0x0000000000769a09 in rocksdb::ThreadLocalPtr::InitSingletons() () at util/thread_local.cc:105
#4 0x0000000000647d98 in rocksdb::Env::Default() () at util/env_posix.cc:845
Reviewers: lightmark, yhchiang, sdong
Reviewed By: sdong
Subscribers: arahut, IslamAbdelRahman, yiwu, andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D60813
Summary:
When a child thread that uses ThreadLocalPtr, ThreadLocalPtr::OnThreadExit
will be called when that child thread is destroyed. However,
OnThreadExit will try to access a static singleton of ThreadLocalPtr,
which will be destroyed when the main thread exit. As a result,
when a child thread that uses ThreadLocalPtr exits AFTER the main thread
exits, illegal memory access will occur.
This diff includes a test that reproduce this legacy bug.
==2095206==ERROR: AddressSanitizer: heap-use-after-free on address
0x608000007fa0 at pc 0x959b79 bp 0x7f5fa7426b60 sp 0x7f5fa7426b58
READ of size 8 at 0x608000007fa0 thread T1
This patch fix this issue by having the thread local mutex never be deleted
(but will leak small piece of memory at the end.) The patch also describe
a better solution (thread_local) in the comment that requires gcc 4.8.1 and
in latest clang as a future work once we agree to move toward gcc 4.8.
Test Plan:
COMPILE_WITH_ASAN=1 make thread_local_test -j32
./thread_local_test --gtest_filter="*MainThreadDiesFirst"
Reviewers: anthony, hermanlee4, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D53013
Summary:
By default, RocksDB initializes the singletons of ThreadLocalPtr first, then initializes PosixEnv
via static initializer. Destructor terminates objects in reverse order, so terminating PosixEnv
(calling pthread_mutex_lock), then ThreadLocal (calling pthread_mutex_destroy).
However, in certain case, application might initialize PosixEnv first, then ThreadLocalPtr.
This will cause core dump at the end of the program (eg. https://github.com/facebook/mysql-5.6/issues/122)
This patch fix this issue by ensuring the destruction order by moving the global static singletons
to function static singletons. Since function static singletons are initialized when the function is first
called, this property allows us invoke to enforce the construction of the static PosixEnv and the
singletons of ThreadLocalPtr by calling the function where the ThreadLocalPtr singletons belongs
right before we initialize the static PosixEnv.
Test Plan: Verified in the MyRocks.
Reviewers: yoshinorim, IslamAbdelRahman, rven, kradhakrishnan, anthony, sdong, MarkCallaghan
Reviewed By: anthony
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D51789
Summary:
This patch fixes the following clang-build error in util/thread_local.cc by using a cleaner macro blocker:
12:26:31 util/thread_local.cc:157:19: error: declaration shadows a static data member of 'rocksdb::ThreadLocalPtr::StaticMeta' [-Werror,-Wshadow]
12:26:31 ThreadData* tls_ =
12:26:31 ^
12:26:31 util/thread_local.cc:19:66: note: previous declaration is here
12:26:31 __thread ThreadLocalPtr::ThreadData* ThreadLocalPtr::StaticMeta::tls_ = nullptr;
12:26:31 ^
Test Plan: db_test
Reviewers: sdong, anthony, IslamAbdelRahman, igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D44043
Summary:
This patch provides a simplier solution to the memory leak
issue identified in patch https://reviews.facebook.net/D43677,
where a static function local variable can be used instead of
using a global static unique_ptr.
Test Plan: run db_stress on mac
Reviewers: igor, sdong, anthony, IslamAbdelRahman, maykov
Reviewed By: maykov
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43995
Commit 257ee89 added a static destruction helper to avoid notional
"leaks" of TLS on main thread exit. This helper fails to compile on
OS X (and presumably Windows, though I haven't checked), which lacks
the __thread storage class StaticMeta::tls_ member.
This patch fixes the builds. Do note that the static cleanup mechanism
may be somewhat brittle and atexit(3) may be a more suitable approach
to releasing the main thread's TLS if it's highly desirable for this
memory to not be reported "reachable" by Valgrind at exit.
Summary:
MyRocks valgrind run was showing memory leaks. The fixes are mostly self-explaining.
There is only a single usage of ThreadLocalPtr. Potentially, we may think about replacing this use with thread_local, but it will be a bigger change. Another option to consider is using thread_local instead of __thread in ThreadLocalPtr implementation. This way, tls_ can be stored using std::unique_ptr and no destructor would be required.
Test Plan:
- make check
- MyRocks valgrind run doesn't report leaks
Reviewers: rven, sdong
Reviewed By: sdong
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D43677
Summary: This helps Windows port to format their changes, as discussed. Might have formatted some other codes too becasue last 10 commits include more.
Test Plan: Build it.
Reviewers: anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D41961
Summary: Make RocksDb build and run on Windows to be functionally
complete and performant. All existing test cases run with no
regressions. Performance numbers are in the pull-request.
Test plan: make all of the existing unit tests pass, obtain perf numbers.
Co-authored-by: Praveen Rao praveensinghrao@outlook.com
Co-authored-by: Sherlock Huang baihan.huang@gmail.com
Co-authored-by: Alex Zinoviev alexander.zinoviev@me.com
Co-authored-by: Dmitri Smirnov dmitrism@microsoft.com
Summary:
ThreadSanitizer complains data race of super version and version's destructor with Get(). This patch will fix those warning.
The warning is likely from ColumnFamilyData::ReturnThreadLocalSuperVersion(). With relaxed consistency of CAS, reading the data of the super version can technically happen after swapping it in, enabling the background thread to clean it up.
Test Plan: make all check
Reviewers: rven, igor, yhchiang
Reviewed By: yhchiang
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D32265
Summary: Replace runtime_error exception by abort() in thread_local
Test Plan: make dbg -j32
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29853
Summary:
make singleton a static member instead of dynamic object. This should
also avoid the race on unique_ptr
Test Plan: make all check
Reviewers: igor, haobo, sdong
Reviewed By: igor
CC: leveldb
Differential Revision: https://reviews.facebook.net/D18177
Summary:
Using ThreadLocalPtr as a flag to determine if a mutex is locked or not enables us to implement AssertNotHeld(). It also makes AssertHeld() actually correct.
I had to remove port::Mutex as a dependency for util/thread_local.h, but that's fine since we can just use std::mutex :)
Test Plan: make check
Reviewers: ljin, dhruba, haobo, sdong, yhchiang
Reviewed By: ljin
CC: leveldb
Differential Revision: https://reviews.facebook.net/D18171
Summary:
Add a check at the end of GetImpl to release SuperVersion if it becomes
obsolete. Also do Scrape() inside InstallSuperVersion so it happens more
frequent.
Test Plan:
make all check
running asan_check now
Reviewers: igor, haobo, sdong, dhruba
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16641
Summary: as title
Test Plan:
asan_check
will post results later
Reviewers: haobo, igor, dhruba, sdong
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16257
Summary:
This is not a generic thread local implementation in the sense that it
only takes pointer. But it does support multiple instances per thread
and lets user plugin function to perform cleanup when thread exits or an
instance gets destroyed.
Test Plan: unit test for now
Reviewers: haobo, igor, sdong, dhruba
Reviewed By: igor
CC: leveldb, kailiu
Differential Revision: https://reviews.facebook.net/D16131
Andrew Kryczka