Skip to content

Commit 250a4d5

Browse files
committed
[yugabyte#24026] docdb: Fix SIGSEGV from MaxPersistentOpId after flush
Summary: It is possible for tablet peer's `tablet_` to be null when a rocksdb flush finishes. We call `tablet_->MaxPersistentOpId()` after flush to clean up recently applied transaction state, and this causes a SIGSEGV: ``` * thread #1, name = 'yb-tserver', stop reason = signal SIGSEGV * frame #0: 0x000055885b97311d yb-tserver`yb::ScopedRWOperation::ScopedRWOperation(yb::RWOperationCounter*, yb::StatusHolder const*, std::__1::chrono::time_point<yb::CoarseMonoClock, std::__1::chrono::duration<long long, std::__1::ratio<1l, 1000000000l>>> const&) [inlined] std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>::basic_string(this="", __str=<unavailable>) at string:898:9 frame #1: 0x000055885b97311d yb-tserver`yb::ScopedRWOperation::ScopedRWOperation(yb::RWOperationCounter*, yb::StatusHolder const*, std::__1::chrono::time_point<yb::CoarseMonoClock, std::__1::chrono::duration<long long, std::__1::ratio<1l, 1000000000l>>> const&) [inlined] yb::RWOperationCounter::resource_name(this=0x0000000000000378) const at operation_counter.h:95:12 frame #2: 0x000055885b97311d yb-tserver`yb::ScopedRWOperation::ScopedRWOperation(this=0x00007f9455305d58, counter=0x0000000000000378, abort_status_holder=<unavailable>, deadline=0x00007f9455305d98) at operation_counter.cc:190:62 frame #3: 0x000055885b247ea6 yb-tserver`yb::tablet::Tablet::MaxPersistentOpId(bool) const [inlined] yb::ScopedRWOperation::ScopedRWOperation(this=0x00007f9455305d58, counter=<unavailable>, deadline=0x00007f9455305d98) at operation_counter.h:140:9 frame #4: 0x000055885b247e9f yb-tserver`yb::tablet::Tablet::MaxPersistentOpId(bool) const [inlined] yb::tablet::Tablet::CreateScopedRWOperationBlockingRocksDbShutdownStart(this=0x0000000000000000, deadline=yb::CoarseTimePoint @ 0x00007f9455305d98) const at tablet.cc:3375:10 frame #5: 0x000055885b247e90 yb-tserver`yb::tablet::Tablet::MaxPersistentOpId(this=0x0000000000000000, invalid_if_no_new_data=<unavailable>) const at tablet.cc:3540:32 frame #6: 0x000055885b277f5e yb-tserver`yb::tablet::TabletPeer::MaxPersistentOpId(this=<unavailable>) const at tablet_peer.cc:946:23 frame #7: 0x000055885b278e52 yb-tserver`non-virtual thunk to yb::tablet::TabletPeer::MaxPersistentOpId() const at tablet_peer.cc:0 frame #8: 0x000055885b2dec44 yb-tserver`yb::tablet::TransactionParticipant::Impl::DoProcessRecentlyAppliedTransactions(this=0x0000153123151500, retryable_requests_flushed_op_id=<unavailable>, persist=<unavailable>) at transaction_participant.cc:2186:22 frame #9: 0x000055885b2e0a8e yb-tserver`yb::tablet::TransactionParticipant::ProcessRecentlyAppliedTransactions() [inlined] yb::tablet::TransactionParticipant::Impl::ProcessRecentlyAppliedTransactions(this=0x0000153123151500) at transaction_participant.cc:1440:27 frame #10: 0x000055885b2e0a63 yb-tserver`yb::tablet::TransactionParticipant::ProcessRecentlyAppliedTransactions(this=<unavailable>) at transaction_participant.cc:2629:17 frame #11: 0x000055885b226093 yb-tserver`yb::tablet::Tablet::RocksDbListener::OnFlushCompleted(this=0x0000153110c2da58, (null)=<unavailable>, (null)=<unavailable>) at tablet.cc:503:34 frame yugabyte#12: 0x000055885af0e507 yb-tserver`rocksdb::DBImpl::BackgroundCallFlush(rocksdb::ColumnFamilyData*) at db_impl.cc:2121:19 frame yugabyte#13: 0x000055885af0e275 yb-tserver`rocksdb::DBImpl::BackgroundCallFlush(rocksdb::ColumnFamilyData*) [inlined] rocksdb::DBImpl::FlushMemTableToOutputFile(this=0x0000153123150a80, cfd=0x000015317d651600, mutable_cf_options=0x00007f94553077d8, made_progress=<unavailable>, job_context=0x00007f9455306938, log_buffer=0x00007f9455306048) at db_impl.cc:2008:3 frame yugabyte#14: 0x000055885af0d859 yb-tserver`rocksdb::DBImpl::BackgroundCallFlush(rocksdb::ColumnFamilyData*) [inlined] rocksdb::DBImpl::BackgroundFlush(this=0x0000153123150a80, made_progress=<unavailable>, job_context=0x00007f9455306938, log_buffer=0x00007f9455306048, cfd=0x000015317d651600) at db_impl.cc:3399:10 frame yugabyte#15: 0x000055885af0d21f yb-tserver`rocksdb::DBImpl::BackgroundCallFlush(this=0x0000153123150a80, cfd=<unavailable>) at db_impl.cc:3470:31 frame yugabyte#16: 0x000055885b024a53 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator()() at thread_posix.cc:133:5 frame yugabyte#17: 0x000055885b024900 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator()() [inlined] rocksdb::ThreadPool::StartBGThreads(this=<unavailable>)::$_0::operator()() const at thread_posix.cc:172:5 frame yugabyte#18: 0x000055885b024900 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator()() [inlined] decltype(__f=<unavailable>)::$_0&>()()) std::__1::__invoke[abi:ue170006]<rocksdb::ThreadPool::StartBGThreads()::$_0&>(rocksdb::ThreadPool::StartBGThreads()::$_0&) at invoke.h:340:25 frame yugabyte#19: 0x000055885b024900 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator()() [inlined] void std::__1::__invoke_void_return_wrapper<void, true>::__call[abi:ue170006]<rocksdb::ThreadPool::StartBGThreads(__args=<unavailable>)::$_0&>(rocksdb::ThreadPool::StartBGThreads()::$_0&) at invoke.h:415:5 frame yugabyte#20: 0x000055885b024900 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator()() [inlined] std::__1::__function::__alloc_func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator(this=<unavailable>)[abi:ue170006]() at function.h:192:16 frame yugabyte#21: 0x000055885b024900 yb-tserver`std::__1::__function::__func<rocksdb::ThreadPool::StartBGThreads()::$_0, std::__1::allocator<rocksdb::ThreadPool::StartBGThreads()::$_0>, void ()>::operator(this=<unavailable>)() at function.h:363:12 frame yugabyte#22: 0x000055885b9c1543 yb-tserver`yb::Thread::SuperviseThread(void*) [inlined] std::__1::__function::__value_func<void ()>::operator(this=0x000015313de3b380)[abi:ue170006]() const at function.h:517:16 frame yugabyte#23: 0x000055885b9c152d yb-tserver`yb::Thread::SuperviseThread(void*) [inlined] std::__1::function<void ()>::operator(this=0x000015313de3b380)() const at function.h:1168:12 frame yugabyte#24: 0x000055885b9c152d yb-tserver`yb::Thread::SuperviseThread(arg=0x000015313de3b320) at thread.cc:866:3 frame yugabyte#25: 0x00007f94994d81ca libpthread.so.0`start_thread + 234 frame yugabyte#26: 0x00007f9499729e73 libc.so.6`__clone + 67 ``` This diff adds a null check and returns `OpId::Min()` (i.e. don't clean anything up) if `tablet_` is null and we cannot call `MaxPersistentOpId`. Jira: DB-12915 Test Plan: Jenkins Reviewers: sergei, rthallam Reviewed By: sergei, rthallam Subscribers: rthallam, ybase Differential Revision: https://phorge.dev.yugabyte.com/D38323
1 parent 9f90819 commit 250a4d5

File tree

1 file changed

+6
-1
lines changed

1 file changed

+6
-1
lines changed

src/yb/tablet/tablet_peer.cc

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -943,7 +943,12 @@ void TabletPeer::GetInFlightOperations(Operation::TraceType trace_type,
943943
}
944944

945945
Result<OpId> TabletPeer::MaxPersistentOpId() const {
946-
auto flush_op_ids = VERIFY_RESULT(tablet_->MaxPersistentOpId());
946+
auto tablet = shared_tablet();
947+
if (!tablet) {
948+
// Tablet peer not yet initialized -- we could be doing tablet bootstrap still.
949+
return OpId::Min();
950+
}
951+
auto flush_op_ids = VERIFY_RESULT(tablet->MaxPersistentOpId());
947952
return OpId::MinValid(flush_op_ids.intents, flush_op_ids.regular);
948953
}
949954

0 commit comments

Comments
 (0)