Log Message:
Update databases/apache-cassandra2 to 2.1.16.

2.1.16
- Avoid infinitely looping result set when paging SELECT queries with
  an IN clause with duplicate keys by treating the IN values as a set
  instead of a list
- Add system property to set the max number of native transport requests
  in queue
- Include column family parameter when -st and -et are provided
- Fix queries with empty ByteBuffer values in clustering column
  restrictions
- Disable passing control to post-flush after flush failure to prevent
  data loss
- Allow STCS-in-L0 compactions to reduce scope with LCS
- cannot use cql since upgrading python to 2.7.11+
- Fix filtering on clustering columns when 2i is used
- Improve digest calculation in the presence of overlapping tombstones

2.1.15
- Account for partition deletions in tombstone histogram
- Avoid stalling paxos when the paxos state expires
- Remove finished incoming streaming connections from MessagingService
- Don't try to get sstables for non-repairing column families
- Prevent select statements with clustering key > 64k
- Avoid marking too many sstables as repaired
- Fix clock skew corrupting other nodes with paxos
- Remove distinction between non-existing static columns and existing
  but null in LWTs
- Support mlockall on IBM POWER arch
- Cache local ranges when calculating repair neighbors
- Allow LWT operation on static column with only partition keys
- Create interval tree over canonical sstables to avoid missing sstables
  during streaming
- cqlsh COPY FROM: shutdown parent cluster after forking, to avoid
  corrupting SSL connections
- Updated cqlsh Python driver to fix DESCRIBE problem for legacy tables
- cqlsh: apply current keyspace to source command
- Backport CASSANDRA-11578
- Clear out parent repair session if repair coordinator dies
- Set default streaming_socket_timeout_in_ms to 24 hours
- Do not consider local node a valid source during replace
- Avoid holding SSTableReaders for duration of incremental repair
- Add message dropped tasks to nodetool netstats
- Don't compute expensive MaxPurgeableTimestamp until we've verified
  there's an expired tombstone
- Fix paging on DISTINCT queries repeats result when first row in
  partition changes
- Add option to disable use of severity in DynamicEndpointSnitch
- cqlsh COPY FROM fails for null values with non-prepared statements
- Make cython optional in pylib/setup.py
- Change order of directory searching for cassandra.in.sh to favor local
  one
- cqlsh COPY FROM fails with []{} chars in UDT/tuple fields/values
- clqsh: COPY FROM throws TypeError with Cython extensions enabled
- cqlsh: COPY FROM ignores NULL values in conversion
- Validate levels when building LeveledScanner to avoid overlaps with
  orphaned sstables

2.1.14
- Start L0 STCS-compactions even if there is a L0 -> L1 compaction going
- (cqlsh) Fix potential COPY deadlock when parent process is terminating
  child processes
- Replace sstables on DataTracker before marking them as non-compacting
  during anti-compaction
- Checking if an unlogged batch is local is inefficient
- Fix paging for COMPACT tables without clustering columns
- Fix out-of-space error treatment in memtable flushing
- Backport CASSANDRA-10859
- COPY FROM fails when importing blob
- Backport CASSANDRA-10679
- Don't do defragmentation if reading from repaired sstables
- Fix streaming_socket_timeout_in_ms not enforced
- Avoid dropping message too quickly due to missing unit conversion
- COPY FROM on large datasets: fix progress report and debug performance
- InvalidateKeys should have a weak ref to key cache
- Don't remove FailureDetector history on removeEndpoint
- Only notify if repair status changed
- Add partition key to TombstoneOverwhelmingException error message
- Use logback setting for 'cassandra -v' command
- Fix sstableloader to unthrottle streaming by default
- Fix incorrect warning in 'nodetool status'
- Properly release sstable ref when doing offline scrub
- Improve nodetool status performance for large cluster
- Make it clear what DTCS timestamp_resolution is used for
- Gossiper#isEnabled is not thread safe
- Avoid major compaction mixing repaired and unrepaired sstables in DTCS
- test_bulk_round_trip_blogposts is failing occasionally
- Add a -j parameter to scrub/cleanup/upgradesstables to state how
  many threads to use

2.1.13
- Fix isJoined return true only after becoming cluster member
  (CASANDRA-11007)
- Fix bad gossip generation seen in long-running clusters
- Avoid NPE when incremental repair fails
- Unmark sstables compacting once they are done in
  cleanup/scrub/upgradesstables
- Revert CASSANDRA-10012 and add more logging
- Allow simultaneous bootstrapping with strict consistency when no
  vnodes are used
- Log a message when major compaction does not result in a single file
- (cqlsh) fix cqlsh_copy_tests when vnodes are disabled
- (cqlsh) fix formatting bytearray values
- (cqlsh) Add request timeout option to cqlsh
- Avoid AssertionError while submitting hint with LWT
- If CompactionMetadata is not in stats file, use index summary instead
- Retry sending gossip syn multiple times during shadow round
- Fix pending range calculation during moves
- Sane default (200Mbps) for inter-DC streaming througput
- Match cassandra-loader options in COPY FROM
- Fix binding to any address in CqlBulkRecordWriter
- Fix the way we replace sstables after anticompaction
- cqlsh fails to decode utf-8 characters for text typed columns
- Log error when stream session fails
- Fix bugs in commit log archiving startup behavior
- (cqlsh) further optimise COPY FROM
- Allow CREATE TABLE WITH ID
- Make Stress compiles within eclipse
- Cassandra Daemon should print JVM arguments
- Allow cancellation of index summary redistribution
- sstableloader will fail if there are collections in the schema tables
- Disable reloading of GossipingPropertyFileSnitch
- Fix Stress profile parsing on Windows

2.1.12
- Fix incremental repair hang when replica is down
- Avoid writing range tombstones after END_OF_ROW marker
- Optimize the way we check if a token is repaired in anticompaction
- Add proper error handling to stream receiver
- Warn or fail when changing cluster topology live
- Status command in debian/ubuntu init script doesn't work
- Some DROP ... IF EXISTS incorrectly result in exceptions on
  non-existing KS
- DeletionTime.compareTo wrong in rare cases
- Force encoding when computing statement ids
- Properly reject counters as map keys
- Fix the sstable-needs-cleanup check
- (cqlsh) Print column names before COPY operation
- Add Native-Transport-Requests back to tpstats
- Make paging logic consistent between searcher impls
- Fix CompressedInputStream for proper cleanup
- (cqlsh) Support counters in COPY commands
- Try next replica if not possible to connect to primary replica on
  ColumnFamilyRecordReader
- Limit window size in DTCS
- sstableloader does not use MAX_HEAP_SIZE env parameter
- (cqlsh) Improve COPY TO performance and error handling
- Don't remove level info when running upgradesstables
- Create compression chunk for sending file only
- Make buffered read size configurable
- Forbid compact clustering column type changes in ALTER TABLE
- Reject incremental repair with subrange repair
- Add a nodetool command to refresh size_estimates
- Shutdown compaction in drain to prevent leak
- Invalidate cache after stream receive task is completed
- Reject counter writes in CQLSSTableWriter
- Remove superfluous COUNTER_MUTATION stage mapping
- Improve json2sstable error reporting on nonexistent columns
- (cqlsh) fix COPY using wrong variable name for time_format
- Do not run SizeEstimatesRecorder if a node is not a member of the ring
- Improve handling of dead nodes in gossip
- Fix logback-tools.xml incorrectly configured for outputing to
  System.err
- Fix streaming to catch exception so retry not fail
- Add validation method to PerRowSecondaryIndex
- Support encrypted and plain traffic on the same port
- Do STCS in DTCS windows
- Don't try to get ancestors from half-renamed sstables
- Avoid repetition of JVM_OPTS in debian package
- Fix potential NPE from handling result of SIM.highestSelectivityIndex
- Fix paging issues with partitions containing only static columns data
- Fix conditions on static columns
- AssertionError: attempted to delete non-existing file CommitLog
- Merge range tombstones during compaction
- (cqlsh) Distinguish negative and positive infinity in output
- (cqlsh) allow custom time_format for COPY TO
- Don't allow startup if the node's rack has changed
- Fix sorting for queries with an IN condition on partition key columns