Using thrift HSHA with offheap_objects appears to corrupt data

Copying my comment from CASSANDRA-6285 to a new issue since that issue is long closed and I'm not sure if they are related...

I am getting this exception using Thrift HSHA in 2.1.0:

INFO [CompactionExecutor:8] 2015-01-26 13:32:51,818 CompactionTask.java (line 138) Compacting [SSTableReader(path='/tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-ka-2-Data.db'), SSTableReader(path='/tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-ka-1-Data.db')]
INFO [CompactionExecutor:8] 2015-01-26 13:32:51,890 ColumnFamilyStore.java (line 856) Enqueuing flush of compactions_in_progress: 212 (0%) on-heap, 20 (0%) off-heap
INFO [MemtableFlushWriter:8] 2015-01-26 13:32:51,892 Memtable.java (line 326) Writing Memtable-compactions_in_progress@1155018639(0 serialized bytes, 1 ops, 0%/0% of on/off-heap limit)
INFO [MemtableFlushWriter:8] 2015-01-26 13:32:51,896 Memtable.java (line 360) Completed flushing /tmp/cass_test/cassandra/TestCassandra/data/system/compactions_in_progress-55080ab05d9c388690a4acb25fe1f77b/system-compactions_in_progress-ka-2-Data.db (42 bytes) for commitlog position ReplayPosition(segmentId=1422296630707, position=430226)
ERROR [CompactionExecutor:8] 2015-01-26 13:32:51,906 CassandraDaemon.java (line 166) Exception in thread Thread[CompactionExecutor:8,1,RMI Runtime]
java.lang.RuntimeException: Last written key DecoratedKey(131206587314004820534098544948237170809, 800100010000000c62617463685f6d7574617465000000) >= current key DecoratedKey(14775611966645399672119169777260659240, 726f776b65793030385f31343232323937313537353835) writing into /tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-tmp-ka-3-Data.db
at org.apache.cassandra.io.sstable.SSTableWriter.beforeAppend(SSTableWriter.java:172) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.sstable.SSTableWriter.append(SSTableWriter.java:196) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.sstable.SSTableRewriter.append(SSTableRewriter.java:110) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionTask.runWith(CompactionTask.java:177) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.util.DiskAwareRunnable.runMayThrow(DiskAwareRunnable.java:48) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.utils.WrappedRunnable.run(WrappedRunnable.java:28) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionTask.executeInternal(CompactionTask.java:74) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.AbstractCompactionTask.execute(AbstractCompactionTask.java:59) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionManager$BackgroundCompactionTask.run(CompactionManager.java:235) ~[apache-cassandra-2.1.0.jar:2.1.0]
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:471) ~[na:1.7.0_40]
at java.util.concurrent.FutureTask.run(FutureTask.java:262) ~[na:1.7.0_40]
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145) ~[na:1.7.0_40]
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615) [na:1.7.0_40]
at java.lang.Thread.run(Thread.java:724) [na:1.7.0_40]

I don't think it's caused by CASSANDRA-8211, because it happens during the first compaction that takes place between the first 2 SSTables to get flushed from an initially empty column family.

Also, I've only been able to reproduce it when using both hsha for the rpc server and offheap_objects for memtable allocation. If I switch either to sync or to offheap_buffers or heap_buffers then I cannot reproduce the problem. Also under the same circumstances I'm pretty sure I've seen incorrect data being returned to a client multiget_slice request before any SSTables had been flushed yet, so I presume this is corruption that happens before any flush/compaction takes place.

nodetool scrub yielded these errors:

INFO [CompactionExecutor:9] 2015-01-26 13:48:01,512 OutputHandler.java (line 42) Scrubbing SSTableReader(path='/tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-ka-2-Data.db') (168780 bytes)
INFO [CompactionExecutor:10] 2015-01-26 13:48:01,512 OutputHandler.java (line 42) Scrubbing SSTableReader(path='/tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-ka-1-Data.db') (135024 bytes)
WARN [CompactionExecutor:9] 2015-01-26 13:48:01,531 OutputHandler.java (line 52) Out of order row detected (DecoratedKey(14775611966645399672119169777260659240, 726f776b65793030385f31343232323937313537353835) found after DecoratedKey(131206587314004820534098544948237170809, 800100010000000c62617463685f6d7574617465000000))
WARN [CompactionExecutor:9] 2015-01-26 13:48:01,534 OutputHandler.java (line 57) Error reading row (stacktrace follows):
java.lang.RuntimeException: Last written key DecoratedKey(131206587314004820534098544948237170809, 800100010000000c62617463685f6d7574617465000000) >= current key DecoratedKey(131206587314004820534098544948237170809, 800100010000000c62617463685f6d7574617465000000) writing into /tmp/cass_test/cassandra/TestCassandra/data/test_ks/test_cf-1c45da40a58911e4826751fbbc77b187/test_ks-test_cf-tmp-ka-4-Data.db
at org.apache.cassandra.io.sstable.SSTableWriter.beforeAppend(SSTableWriter.java:172) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.sstable.SSTableWriter.append(SSTableWriter.java:196) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.sstable.SSTableRewriter.append(SSTableRewriter.java:110) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.io.sstable.SSTableRewriter.tryAppend(SSTableRewriter.java:141) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.Scrubber.scrub(Scrubber.java:186) ~[apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionManager.scrubOne(CompactionManager.java:592) [apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionManager.access$300(CompactionManager.java:100) [apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionManager$3.execute(CompactionManager.java:315) [apache-cassandra-2.1.0.jar:2.1.0]
at org.apache.cassandra.db.compaction.CompactionManager$2.call(CompactionManager.java:270) [apache-cassandra-2.1.0.jar:2.1.0]
at java.util.concurrent.FutureTask.run(FutureTask.java:262) [na:1.7.0_40]
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145) [na:1.7.0_40]
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615) [na:1.7.0_40]
at java.lang.Thread.run(Thread.java:724) [na:1.7.0_40]
WARN [CompactionExecutor:9] 2015-01-26 13:48:01,534 OutputHandler.java (line 52) Row starting at position 25342 is unreadable; skipping to next
WARN [CompactionExecutor:10] 2015-01-26 13:48:01,534 OutputHandler.java (line 52) Out of order row detected (DecoratedKey(29459452031265566667651334397450214244, 726f776b65793030355f31343232323936393033323837) found after DecoratedKey(131206587314004820534098544948237170809, 800100010000000c62617463685f6d7574617465000000))

etc...

I've able to reliably reproduce by creating a new empty table, writing a few partitions containing a few rows each, forcing a flush using nodetool, writing some more, and flushing again, and letting the first compaction happen (using LCS). If I read before flushing, sometimes I will get zero rows back (with quorum between writes and reads), or corrupt bytes, though sometimes reading would yield the correct results. I haven't been able to narrow down yet the cases when the data comes back corrupt vs not, though in the same test case (one set of of writes) the results of a read appear to be consistent (successive reads are either all correct or all incorrect).