RocksDB3.5 crash when doing manual compacting.

[yinqiwen]

There is another crash in rocksdb when application doing a manual compacting.
This is the crash stack trace.
(gdb) bt
#0 0x00007fadfc62b3e9 in raise () from /home/opt/compiler/gcc-4.8.2.09/lib64/libc-2.18.so
#1 0x00007fadfc62c7e8 in abort () from /home/opt/compiler/gcc-4.8.2.09/lib64/libc-2.18.so
#2 0x00007fadfc624516 in __assert_fail_base () from /home/opt/compiler/gcc-4.8.2.09/lib64/libc-2.18.so
#3 0x00007fadfc6245c2 in __assert_fail () from /home/opt/compiler/gcc-4.8.2.09/lib64/libc-2.18.so
#4 0x000000000051f852 in rocksdb::DBImpl::RunManualCompaction (this=this@entry=0x7fadfbebf400, cfd=cfd@entry=0x7fadfbc17f00, input_level=input_level@entry=3, output_level=output_level@entry=4, output_path_id=output_path_id@entry=0,

begin=begin@entry=0x0, end=end@entry=0x0) at db/db_impl.cc:1881

#5 0x0000000000529e40 in rocksdb::DBImpl::CompactRange (this=this@entry=0x7fadfbebf400, column_family=, begin=begin@entry=0x0, end=end@entry=0x0, reduce_level=reduce_level@entry=false,

target_level=target_level@entry=-1, target_path_id=target_path_id@entry=0) at db/db_impl.cc:1673

#6 0x000000000053bcc1 in rocksdb::DB::CompactRange (this=0x7fadfbebf400, begin=0x0, end=0x0, reduce_level=, target_level=-1, target_path_id=0) at ./include/rocksdb/db.h:360
#7 0x0000000000517983 in ardb::RocksDBEngine::CompactRange (this=, begin=..., end=...) at engine/rocksdb_engine.cpp:299
#8 0x00000000004a0778 in ardb::Ardb::DoCompact (this=0x7fffa07e8670, start=..., end=...) at command/server.cpp:749
#9 0x000000000044ac77 in ardb::CompactTask::Run (this=) at cron.cpp:117
#10 0x0000000000497698 in ardb::Timer::Routine (this=this@entry=0x7fadf05cc500) at common/channel/timer/timer.cpp:217
#11 0x0000000000496afc in ardb::TimerChannel::TimeoutCB (eventLoop=, id=, clientData=0x7fadf05cc500) at common/channel/timer/timer_channel.cpp:39
#12 0x000000000049f721 in processTimeEvents (eventLoop=0x7fadfbc10a68) at common/channel/redis/ae.c:305
#13 aeProcessEvents (eventLoop=eventLoop@entry=0x7fadfbc10a68, flags=flags@entry=3) at common/channel/redis/ae.c:441
#14 0x000000000049fa3b in aeMain (eventLoop=0x7fadfbc10a68) at common/channel/redis/ae.c:485
#15 0x0000000000477520 in ardb::Thread::ThreadFunc (data=0x7fffa07e8bc8) at common/thread/thread.cpp:38
#16 0x00007fadfd7f71c1 in start_thread () from /home/opt/compiler/gcc-4.8.2.09/lib64/libpthread-2.18.so
#17 0x00007fadfc6dd0ad in clone () from /home/opt/compiler/gcc-4.8.2.09/lib64/libc-2.18.so

[yinqiwen]

This bug seems introduced by d977e55

[igorcanadi]

Hmm. This is interesting. Are you running manual compactions in parallel?

[igorcanadi]

Tnx for reporting this @yinqiwen. It turns out that the assertion that caused this was wrong. It was fixed by removing the assert here: 2fb1fea

If you're running in production, please use make release that will remove all asserts from the code :)