rocksdb/db
Mike Kolupaev ee6f4fa6f5 Fix deadlock in ColumnFamilyData::InstallSuperVersion()
Summary:
Deadlock: a memtable flush holds DB::mutex_ and calls ThreadLocalPtr::Scrape(), which locks ThreadLocalPtr mutex; meanwhile, a thread exit handler locks ThreadLocalPtr mutex and calls SuperVersionUnrefHandle, which tries to lock DB::mutex_.

This deadlock is hit all the time on our workload. It blocks our release.

In general, the problem is that ThreadLocalPtr takes an arbitrary callback and calls it while holding a lock on a global mutex. The same global mutex is (at least in some cases) locked by almost all ThreadLocalPtr methods, on any instance of ThreadLocalPtr. So, there'll be a deadlock if the callback tries to do anything to any instance of ThreadLocalPtr, or waits for another thread to do so.

So, probably the only safe way to use ThreadLocalPtr callbacks is to do only do simple and lock-free things in them.

This PR fixes the deadlock by making sure that local_sv_ never holds the last reference to a SuperVersion, and therefore SuperVersionUnrefHandle never has to do any nontrivial cleanup.

I also searched for other uses of ThreadLocalPtr to see if they may have similar bugs. There's only one other use, in transaction_lock_mgr.cc, and it looks fine.
Closes https://github.com/facebook/rocksdb/pull/3510

Reviewed By: sagar0

Differential Revision: D7005346

Pulled By: al13n321

fbshipit-source-id: 37575591b84f07a891d6659e87e784660fde815f
2018-02-21 10:52:03 -08:00
..
builder.cc Stream 2017-11-10 09:26:24 -08:00
builder.h Stream 2017-11-10 09:26:24 -08:00
c_test.c Added save points for transactions C API 2017-09-14 14:18:59 -07:00
c.cc add support for allow_ingest_behind in C API 2018-01-08 17:26:31 -08:00
column_family_test.cc arena: derive alignment unit from std::max_align_t 2017-10-17 11:13:19 -07:00
column_family.cc Fix deadlock in ColumnFamilyData::InstallSuperVersion() 2018-02-21 10:52:03 -08:00
column_family.h Fix Flush() keep waiting after flush finish 2018-01-18 17:50:07 -08:00
compact_files_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
compacted_db_impl.cc Fix coverity issues column_family, compaction_db/iterator 2017-10-27 11:26:42 -07:00
compacted_db_impl.h Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
compaction_iteration_stats.h add counter for deletion dropping optimization 2017-08-19 14:10:08 -07:00
compaction_iterator_test.cc Added support for differential snapshots 2017-11-01 18:56:43 -07:00
compaction_iterator.cc WritePrepared Txn: fix compaction filter snapshot checks 2017-12-12 11:12:24 -08:00
compaction_iterator.h Added support for differential snapshots 2017-11-01 18:56:43 -07:00
compaction_job_stats_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
compaction_job_test.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
compaction_job.cc Fix a wrong log formatting 2018-01-09 14:58:33 -08:00
compaction_job.h Make DBOption compaction_readahead_size dynamic 2017-11-16 17:57:25 -08:00
compaction_picker_test.cc Make Universal compaction options dynamic 2017-12-11 13:27:06 -08:00
compaction_picker_universal.cc Make Universal compaction options dynamic 2017-12-11 13:27:06 -08:00
compaction_picker_universal.h Make Universal compaction options dynamic 2017-12-11 13:27:06 -08:00
compaction_picker.cc use bottommost compression when base level is bottommost 2017-11-09 17:42:00 -08:00
compaction_picker.h fix hanging after CompactFiles with L0 overlap 2017-09-13 15:41:38 -07:00
compaction.cc Make Universal compaction options dynamic 2017-12-11 13:27:06 -08:00
compaction.h Change RocksDB License 2017-07-15 16:11:23 -07:00
comparator_db_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
convenience.cc add VerifyChecksum() to db.h 2017-08-09 15:58:13 -07:00
corruption_test.cc fix corruption_test valgrind 2017-08-11 12:29:14 -07:00
cuckoo_table_db_test.cc Enable MSVC W4 with a few exceptions. Fix warnings and bugs 2017-10-19 10:57:12 -07:00
db_basic_test.cc Enable MSVC W4 with a few exceptions. Fix warnings and bugs 2017-10-19 10:57:12 -07:00
db_blob_index_test.cc fix lite build 2017-10-17 08:57:09 -07:00
db_block_cache_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_bloom_filter_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_compaction_filter_test.cc Split CompactionFilterWithValueChange 2017-10-20 15:42:07 -07:00
db_compaction_test.cc fix ASAN for DeleteFilesInRange test case 2017-12-07 11:12:43 -08:00
db_dynamic_level_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_encryption_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
db_filesnapshot.cc Remove some left-over BSD headers 2017-07-18 11:56:57 -07:00
db_flush_test.cc Fix DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge dead lock 2018-02-05 17:13:23 -08:00
db_impl_compaction_flush.cc Fix DBFlushTest::ManualFlushWithMinWriteBufferNumberToMerge dead lock 2018-02-05 17:13:23 -08:00
db_impl_debug.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
db_impl_experimental.cc Inform caller when rocksdb is stalling writes 2017-10-05 18:11:43 -07:00
db_impl_files.cc fix DBImpl instance variable naming 2018-01-09 12:56:58 -08:00
db_impl_open.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
db_impl_readonly.cc WritePrepared Txn: Iterator 2017-10-09 17:15:28 -07:00
db_impl_readonly.h Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_impl_write.cc fix DBImpl instance variable naming 2018-01-09 12:56:58 -08:00
db_impl.cc WritePrepared Txn: Return NotSupported on iterator refresh 2017-12-18 22:29:30 -08:00
db_impl.h Fix Flush() keep waiting after flush finish 2018-01-18 17:50:07 -08:00
db_info_dumper.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
db_info_dumper.h Change RocksDB License 2017-07-15 16:11:23 -07:00
db_inplace_update_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
db_io_failure_test.cc Fix LITE unit tests 2017-07-26 21:11:47 -07:00
db_iter_test.cc fix Seek with lower_bound 2017-11-29 22:56:29 -08:00
db_iter.cc WritePrepared Txn: Return NotSupported on iterator refresh 2017-12-18 22:29:30 -08:00
db_iter.h WritePrepared Txn: Return NotSupported on iterator refresh 2017-12-18 22:29:30 -08:00
db_iterator_test.cc fix Seek with lower_bound 2017-11-29 22:56:29 -08:00
db_log_iter_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
db_memtable_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_merge_operator_test.cc Make iterator invalid on Merge error 2017-12-06 11:56:39 -08:00
db_options_test.cc Make DBOption compaction_readahead_size dynamic 2017-11-16 17:57:25 -08:00
db_properties_test.cc Add DB::Properties::kEstimateOldestKeyTime 2017-10-23 15:27:27 -07:00
db_range_del_test.cc Fix wrong smallest key of delete range tombstones 2017-08-29 18:41:35 -07:00
db_sst_test.cc Fix DeleteScheduler::MarkAsTrash() handling existing trash 2017-12-12 18:17:13 -08:00
db_statistics_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
db_table_properties_test.cc fix deletion-triggered compaction in table builder 2017-09-28 18:17:30 -07:00
db_tailing_iter_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
db_test2.cc use bottommost compression when base level is bottommost 2017-11-09 17:42:00 -08:00
db_test_util.cc fix release order in validateNumberOfEntries 2017-12-18 14:27:28 -08:00
db_test_util.h make compaction_readahead_size_ thread safe 2017-11-21 20:11:38 -08:00
db_test.cc Fix deadlock in ColumnFamilyData::InstallSuperVersion() 2018-02-21 10:52:03 -08:00
db_universal_compaction_test.cc Make Universal compaction options dynamic 2017-12-11 13:27:06 -08:00
db_wal_test.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
db_write_test.cc BlobDB: Remove the need to get sequence number per write 2017-12-15 13:27:30 -08:00
dbformat_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
dbformat.cc Added support for differential snapshots 2017-11-01 18:56:43 -07:00
dbformat.h Fix coverity issues compaction_job, compaction_picker 2017-12-11 11:57:15 -08:00
deletefile_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
event_helpers.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
event_helpers.h Change RocksDB License 2017-07-15 16:11:23 -07:00
experimental.cc Replace dynamic_cast<> 2017-07-28 16:27:16 -07:00
external_sst_file_basic_test.cc optimize file ingestion checks for range deletion overlap 2017-11-28 11:27:02 -08:00
external_sst_file_ingestion_job.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
external_sst_file_ingestion_job.h optimize file ingestion checks for range deletion overlap 2017-11-28 11:27:02 -08:00
external_sst_file_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
fault_injection_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
file_indexer_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
file_indexer.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
file_indexer.h Change RocksDB License 2017-07-15 16:11:23 -07:00
filename_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
flush_job_test.cc Add a histogram stat for memtable flush 2017-12-15 18:57:00 -08:00
flush_job.cc Add a histogram stat for memtable flush 2017-12-15 18:57:00 -08:00
flush_job.h WritePrepared Txn: Compaction/Flush 2017-10-06 10:41:53 -07:00
flush_scheduler.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
flush_scheduler.h Change RocksDB License 2017-07-15 16:11:23 -07:00
forward_iterator_bench.cc fix gflags namespace 2017-12-01 10:42:05 -08:00
forward_iterator.cc fix ForwardIterator reference to temporary object 2017-12-20 16:12:04 -08:00
forward_iterator.h Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
internal_stats.cc TableProperty::oldest_key_time defaults to 0 2017-10-27 15:00:05 -07:00
internal_stats.h Add DB::Properties::kEstimateOldestKeyTime 2017-10-23 15:27:27 -07:00
job_context.h Inform caller when rocksdb is stalling writes 2017-10-05 18:11:43 -07:00
listener_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
log_format.h Change RocksDB License 2017-07-15 16:11:23 -07:00
log_reader.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
log_reader.h Suppress lint in old files 2018-02-05 16:54:59 -08:00
log_test.cc Enable MSVC W4 with a few exceptions. Fix warnings and bugs 2017-10-19 10:57:12 -07:00
log_writer.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
log_writer.h Change RocksDB License 2017-07-15 16:11:23 -07:00
malloc_stats.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
malloc_stats.h Change RocksDB License 2017-07-15 16:11:23 -07:00
managed_iterator.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
managed_iterator.h Change RocksDB License 2017-07-15 16:11:23 -07:00
manual_compaction_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
memtable_list_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
memtable_list.cc Add DB::Properties::kEstimateOldestKeyTime 2017-10-23 15:27:27 -07:00
memtable_list.h Fix Flush() keep waiting after flush finish 2018-01-18 17:50:07 -08:00
memtable.cc dynamically change current memtable size 2017-11-02 22:28:10 -07:00
memtable.h Fix Flush() keep waiting after flush finish 2018-01-18 17:50:07 -08:00
merge_context.h Change RocksDB License 2017-07-15 16:11:23 -07:00
merge_helper_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
merge_helper.cc Allow merge operator to be called even with a single operand 2017-08-16 23:42:00 -07:00
merge_helper.h Allow merge operator to be called even with a single operand 2017-08-16 23:42:00 -07:00
merge_operator.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
merge_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
options_file_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
perf_context_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
pinned_iterators_manager.h Change RocksDB License 2017-07-15 16:11:23 -07:00
plain_table_db_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
pre_release_callback.h WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
prefix_test.cc fix gflags namespace 2017-12-01 10:42:05 -08:00
range_del_aggregator_test.cc optimize file ingestion checks for range deletion overlap 2017-11-28 11:27:02 -08:00
range_del_aggregator.cc optimize file ingestion checks for range deletion overlap 2017-11-28 11:27:02 -08:00
range_del_aggregator.h optimize file ingestion checks for range deletion overlap 2017-11-28 11:27:02 -08:00
read_callback.h write-prepared txn: call IsInSnapshot 2017-09-11 09:14:48 -07:00
repair_test.cc fix file numbers after repair 2017-10-10 13:12:37 -07:00
repair.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
snapshot_checker.h WritePrepared Txn: Disable GC during recovery 2017-10-18 09:11:50 -07:00
snapshot_impl.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
snapshot_impl.h BlobDB: Remove the need to get sequence number per write 2017-12-15 13:27:30 -08:00
table_cache.cc Make DBOption compaction_readahead_size dynamic 2017-11-16 17:57:25 -08:00
table_cache.h Change RocksDB License 2017-07-15 16:11:23 -07:00
table_properties_collector_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
table_properties_collector.cc fix duplicate definition of GetEntryType() 2017-11-01 22:56:17 -07:00
table_properties_collector.h Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
transaction_log_impl.cc WritePrepared Txn: Refactor conf params 2017-11-10 17:28:12 -08:00
transaction_log_impl.h WritePrepared Txn: Refactor conf params 2017-11-10 17:28:12 -08:00
version_builder_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
version_builder.cc Fix coverity issues version, write_batch 2017-12-07 11:57:36 -08:00
version_builder.h Allow DB reopen with reduced options.num_levels 2017-08-24 16:10:54 -07:00
version_edit_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
version_edit.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
version_edit.h Fix coverity issues version, write_batch 2017-12-07 11:57:36 -08:00
version_set_test.cc Revert "comment out unused parameters" 2017-07-21 18:26:26 -07:00
version_set.cc Reduce heavy hitter for Get operation 2017-12-12 21:11:33 -08:00
version_set.h WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
wal_manager_test.cc WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00
wal_manager.cc WritePrepared Txn: Refactor conf params 2017-11-10 17:28:12 -08:00
wal_manager.h WritePrepared Txn: Refactor conf params 2017-11-10 17:28:12 -08:00
write_batch_base.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
write_batch_internal.h WritePrepared Txn: cross-compatibility test 2017-11-11 11:28:37 -08:00
write_batch_test.cc Enable MSVC W4 with a few exceptions. Fix warnings and bugs 2017-10-19 10:57:12 -07:00
write_batch.cc Fix coverity issues version, write_batch 2017-12-07 11:57:36 -08:00
write_callback_test.cc Fix write_callback_test compile error 2017-12-06 17:12:27 -08:00
write_callback.h Change RocksDB License 2017-07-15 16:11:23 -07:00
write_controller_test.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
write_controller.cc Change RocksDB License 2017-07-15 16:11:23 -07:00
write_controller.h Change RocksDB License 2017-07-15 16:11:23 -07:00
write_thread.cc Fix IOError on WAL write doesn't propagate to write group follower 2017-11-28 11:42:48 -08:00
write_thread.h WritePrepared Txn: PreReleaseCallback 2017-11-30 23:50:45 -08:00