tsan-suppressions.txt

# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
race:_dl_deallocate_tls
race:yb::tablet::ScopedRowLock::Release

# libunwind uses some double-checked locking which isn't perfectly safe.
# Reported at http://savannah.nongnu.org/bugs/index.php?42677
#
# With TSAN in clang 3.5, it's the init() function that's flagged as a data
# race (not local_addr_space_init()), due to the former calling sigfillset()
# on an unprotected global variable. Given that init() invokes
# local_addr_space_init(), suppressing init() suppresses both races.
race:_ULx86_64_init

# libev uses some lock-free synchronization, but doesn't have TSAN annotations.
# See http://lists.schmorp.de/pipermail/libev/2013q2/002178.html or KUDU-366
# for examples.
race:evpipe_write
race:evpipe_init
race:epoll_ctl

# We're currently a little lax on using atomic instructions for mutations where
# we might need to. There are possible bugs lurking here, though likely we are
# hiding behind x86 TSO for now.
race:yb::RowChangeListDecoder::Init

# GLog's fatal signal handler isn't signal-safe -- it allocates memory.
# This isn't great, but nothing we can do about it. See
# https://code.google.com/p/google-glog/issues/detail?id=191
signal:logging_fail

# This method in Boost's UUID library operates on static state with impunity,
# triggering (harmless) data races in TSAN when boost::uuids::random_generator
# instances are created across threads (see yb::ObjectIdGenerator).
race:boost::uuids::detail::seed_rng::sha1_random_digest_

# ------------------------------------------------------------
# Known bugs below. As these JIRAs are resolved, please remove the relevant
# suppression.
# ------------------------------------------------------------

# KUDU-1283: TSAN warning from consensus OpIdPB
race:yb::OpIdPB::CopyFrom

# KUDU-186: sketchy synchronization in catalog manager
race:yb::master::CatalogManager::Shutdown
race:yb::master::CatalogManagerBgTasks::Shutdown
race:yb::master::CatalogManager::~CatalogManager

# KUDU-189: unsynchronized access to 'state_' in DeferredCallback
race:yb::master::AsyncTabletRequestTask::DeferredCallback

# KUDU-574: raft_consensus_quorum-test race on LocalTestPeerProxy destruction
race:yb::consensus::LocalTestPeerProxy::~LocalTestPeerProxy

# KUDU-569: unsynchronized access to 'state_', 'acceptor_pools_', in
# GetBoundAddresses()
race:yb::Webserver::GetBoundAddresses
race:yb::RpcServer::GetBoundAddresses

# https://yugabyte.atlassian.net/browse/ENG-354
race:gutil::SubmitSpinLockProfileData

# Setting width on a stream is not thread-safe, but some of our tests do it:
# https://jenkins.dev.yugabyte.com/job/yugabyte-centos-phabricator-clang-tsan/42/artifact/build/tsan-clang-dynamic/yb-test-logs/bin__alter_table-randomized-test/AlterTableRandomized_TestRandomSequence.log
race:std::ios_base::width

# -------------------------------------------------------------------------------------------------
# Known YB concurrency issues

# There are some race conditions in shutting down ExternalMiniCluster's log tailer threads, e.g.
# https://goo.gl/jWwsRY.
race:yb::ExternalDaemon::LogTailerThread::LogTailerThread

# Known well implemented lock-free containers
race:boost::lockfree

# -------------------------------------------------------------------------------------------------
# Supressions from libcds:

# false: LazyList potential deadlock
deadlock:cds/intrusive/impl/lazy_list.h
deadlock:cds/intrusive/lazy_list_nogc.h

# false: BronsonAVLTree potential deadlock
deadlock:cds/container/impl/bronson_avltree_map_rcu.h

# Until the third party fix for cassandra-cpp-driver lands.
# TODO(#14710)
deadlock:libcassandra
# TODO(#14677)
deadlock:MasterSnapshotCoordinator

#TODO: temporarily suppressed. Must be researched later
race:cds/container/impl/bronson_avltree_map_rcu.h

race:cds::gc::dhp::retired_array

# https://gist.githubusercontent.com/mbautin/fcb0ce293027b7bc3f93cb358c4c053c/raw
race:cds::intrusive::BasketQueue

# https://gist.githubusercontent.com/mbautin/268c31aaa4daef367d8d76809f8ca5ba/raw
race:cds::threading::pthread::Manager::attachThread

# yb_build.sh tsan --cxx-test util_lockfree-test --gtest_filter LockfreeTest.QueuePerformance
# https://gist.githubusercontent.com/mbautin/22ad5dd7378d55340821182b22ff120d/raw
race:cds::intrusive::OptimisticQueue
race:cds::container::MSQueue
race:cds::intrusive::SegmentedQueue
race:cds::container::MoirQueue
race:cds::gc::dhp::smr::free_thread_data

# And more like these, so we silence all of these warnings in our test using just this:
race:yb::QueuePerformanceHelper::RunAll

# End of suppressions from libcds.
# -------------------------------------------------------------------------------------------------


# std::locale is thread safe
race:std::locale::classic

# libbacktrace
race:backtrace_vector_grow
race:read_line_header
race:read_abbrevs
race:backtrace_alloc
race:dwarf_lookup_pc
race:read_function_entry

# False positives of TSAN, because we use atomic to check whether all responses are filled.
race:yb::redisserver::RedisInboundCall::DoSerialize
race:yb::redisserver::RedisInboundCall::DynamicMemoryUsage
race:yb::redisserver::(anonymous namespace)::BatchContext::LookupDone

# Usages of URCU
race:yb::internal::ConcurrentValue

# Bugs in glog
race:google::FlushLogFilesUnsafe
race:google::GetStackTrace
race:google::InitVLOG3__
race:google::RawLog__SetLastTime

# Bug in gflags
race:google::FlagSaverImpl::RestoreToRegistry

# On shutdown we get a lot of signal-unsafe calls originating from PostgreSQL's ServerLoop.
signal:src/postgres/src/backend/postmaster/postmaster.c

# C++ Cassandra driver uses classic thread-safe implementation of SharedPtr based on atomics.
# The implementation is correct but TSAN has false positive data races during SharedPtr
# destruction.
race:cass::RefCounted<cass::*>::dec_ref

# In the C++ Cassandra driver there is a data race in 'cass::Metadata::update_partitions'.
# Github issue: https://github.com/yugabyte/yugabyte-db/issues/13059
#               [DocDB] Race condition in Cassandra C++ driver between
#               cass::Metadata::update_partitions and cass::Metadata::clear_and_update_back
# Github issue: https://github.com/yugabyte/yugabyte-db/issues/14977
#               [YCQL][TSan Unit Test] Data race in
#               CppCassandraDriverTest.TestCreateUniqueIndexPartial
# The following lines should be removed from the supressions after the issues fixing.
race:cass::Metadata::clear_and_update_back
race:cass::Metadata::update_partitions
race:cass::Metadata::update_keyspaces

# Races in libuv.so
race:uv__io_feed
race:uv__update_time
race:uv_timer_start

# protobuf updates size cache at end of ByteSizeLong, and this update is surrounded by
# GOOGLE_SAFE_CONCURRENT_WRITES_BEGIN and GOOGLE_SAFE_CONCURRENT_WRITES_END.
# But they are defined to empty, so have to suppress it here.
race:google::protobuf::Any::ByteSizeLong

# Races on gflag changes should be strictly test-only issues, as we never update gflags from the
# code, at runtime.
race:fLI::*

# Known problem with PstackWatcher
signal:yb::PstackWatcher::DumpStacks

# Unexpected tsan warning in the getpwuid_r standard thread safe function which is called
# from the pqGetpwuid function
race:pqGetpwuid