Redis

* Resolve a race condition with the PubSubWorkerThread. 1150
* Cleanup socket read error messages. Thanks Vic Yu. 1159
* Cleanup the Connection's selector correctly. Thanks Bruce Merry. 1153
* Added a Monitor object to make working with MONITOR output easy.
Thanks Roey Prat 1033
* Internal cleanup: Removed the legacy Token class which was necessary
with older version of Python that are no longer supported. 1066
* Response callbacks are now case insensitive. This allows users that
call Redis.execute_command() directly to pass lower-case command
names and still get reasonable responses. 1168
* Added support for hiredis-py 1.0.0 encoding error support. This should
make the PythonParser and the HiredisParser behave identically
when encountering encoding errors. Thanks Brian Candler. 1161/1162
* All authentication errors now properly raise AuthenticationError.
AuthenticationError is now a subclass of ConnectionError, which will
cause the connection to be disconnected and cleaned up appropriately.
923
* Add READONLY and READWRITE commands. Thanks theodesp. 1114
* Remove selectors in favor of nonblocking sockets. Selectors had
issues in some environments including eventlet and gevent. This should
resolve those issues with no other side effects.
* Fixed an issue with XCLAIM and previously claimed but not removed
messages. Thanks thomdask. 1192/1191
* Allow for single connection client instances. These instances
are not thread safe but offer other benefits including a subtle
performance increase.
* Added extensive health checks that keep the connections lively.
Passing the "health_check_interval=N" option to the Redis client class
or to a ConnectionPool ensures that a round trip PING/PONG is successful
before any command if the underlying connection has been idle for more
than N seconds. ConnectionErrors and TimeoutErrors are automatically
retried once for health checks.
* Changed the PubSubWorkerThread to use a threading.Event object rather
than a boolean to control the thread's life cycle. Thanks Timothy
Rule. 1194/1195.
* Fixed a bug in Pipeline error handling that would incorrectly retry
ConnectionErrors.

* Fix SentinelConnectionPool to work in multiprocess/forked environments.

* Added support for `select.poll` to test whether data can be read
on a socket. This should allow for significantly more connections to
be used with pubsub. Fixes 486/1115
* Attempt to guarantee that the ConnectionPool hands out healthy
connections. Healthy connections are those that have an established
socket connection to the Redis server, are ready to accept a command
and have no data available to read. Fixes 1127/886
* Use the socket.IPPROTO_TCP constant instead of socket.SOL_TCP.
IPPROTO_TCP is available on more interpreters (Jython for instance).
Thanks Junnplus. 1130
* Fixed a regression introduced in 3.0 that mishandles exceptions not
derived from the base Exception class. KeyboardInterrupt and
gevent.timeout notable. Thanks Christian Fersch. 1128/1129
* Significant improvements to handing connections with forked processes.
Parent and child processes no longer trample on each others' connections.
Thanks to Jay Rolette for the patch and highlighting this issue.
504/732/784/863
* PythonParser no longer closes the associated connection's socket. The
connection itself will close the socket. 1108/1085

* Connection URLs must have one of the following schemes:
redis://, rediss://, unix://. Thanks jdupl123. 961/969
* Fixed an issue with retry_on_timeout logic that caused some TimeoutErrors
to be retried. Thanks Aaron Yang. 1022/1023
* Added support for SNI for SSL. Thanks oridistor and Roey Prat. 1087
* Fixed ConnectionPool repr for pools with no connections. Thanks
Cody Scott. 1043/995
* Fixed GEOHASH to return a None value when specifying a place that
doesn't exist on the server. Thanks guybe7. 1126
* Fixed XREADGROUP to return an empty dictionary for messages that
have been deleted but still exist in the unacknowledged queue. Thanks
xeizmendi. 1116
* Added an owned method to Lock objects. owned returns a boolean
indicating whether the current lock instance still owns the lock.
Thanks Dave Johansen. 1112
* Allow lock.acquire() to accept an optional token argument. If
provided, the token argument is used as the unique value used to claim
the lock. Thankd Dave Johansen. 1112
* Added a reacquire method to Lock objects. reacquire attempts to renew
the lock such that the timeout is extended to the same value that the
lock was initially acquired with. Thanks Ihor Kalnytskyi. 1014
* Stream names found within XREAD and XREADGROUP responses now properly
respect the decode_responses flag.
* XPENDING_RANGE now requires the user the specify the min, max and
count arguments. Newer versions of Redis prevent count from being
infinite so it's left to the user to specify these values explicitly.
* ZADD now returns None when xx=True and incr=True and an element
is specified that doesn't exist in the sorted set. This matches
what the server returns in this case. 1084
* Added client_kill_filter that accepts various filters to identify
and kill clients. Thanks Theofanis Despoudis. 1098
* Fixed a race condition that occurred when unsubscribing and
resubscribing to the same channel or pattern in rapid succession.
Thanks Marcin Raczyński. 764
* Added a LockNotOwnedError that is raised when trying to extend or
release a lock that is no longer owned. This is a subclass of LockError
so previous code should continue to work as expected. Thanks Joshua
Harlow. 1095
* Fixed a bug in GEORADIUS that forced decoding of places without
respecting the decode_responses option. Thanks Bo Bayles. 1082

* Fixed regression with UnixDomainSocketConnection caused by 3.0.0.
Thanks Jyrki Muukkonen
* Fixed an issue with the new asynchronous flag on flushdb and flushall.
Thanks rogeryen
* Updated Lock.locked() method to indicate whether *any* process has
acquired the lock, not just the current one. This is in line with
the behavior of threading.Lock. Thanks Alan Justino da Silva

BACKWARDS INCOMPATIBLE CHANGES
* When using a Lock as a context manager and the lock fails to be acquired
a LockError is now raised. This prevents the code block inside the
context manager from being executed if the lock could not be acquired.
* Renamed LuaLock to Lock.
* Removed the pipeline based Lock implementation in favor of the LuaLock
implementation.
* Only bytes, strings and numbers (ints, longs and floats) are acceptable
for keys and values. Previously redis-py attempted to cast other types
to str() and store the result. This caused must confusion and frustration
when passing boolean values (cast to 'True' and 'False') or None values
(cast to 'None'). It is now the user's responsibility to cast all
key names and values to bytes, strings or numbers before passing the
value to redis-py.
* The StrictRedis class has been renamed to Redis. StrictRedis will
continue to exist as an alias of Redis for the foreseeable future.
* The legacy Redis client class has been removed. It caused much confusion
to users.
* ZINCRBY arguments 'value' and 'amount' have swapped order to match the
the Redis server. The new argument order is: keyname, amount, value.
* MGET no longer raises an error if zero keys are passed in. Instead an
empty list is returned.
* MSET and MSETNX now require all keys/values to be specified in a single
dictionary argument named mapping. This was changed to allow for future
options to these commands in the future.
* ZADD now requires all element names/scores be specified in a single
dictionary argument named mapping. This was required to allow the NX,
XX, CH and INCR options to be specified.
* ssl_cert_reqs now has a default value of 'required' by default. This
should make connecting to a remote Redis server over SSL more secure.
Thanks u2mejc
* Removed support for EOL Python 2.6 and 3.3. Thanks jdufresne
OTHER CHANGES
* Added missing DECRBY command. Thanks derek-dchu
* CLUSTER INFO and CLUSTER NODES responses are now properly decoded to
strings.
* Added a 'locked()' method to Lock objects. This method returns True
if the lock has been acquired and owned by the current process,
otherwise False.
* EXISTS now supports multiple keys. It's return value is now the number
of keys in the list that exist.
* Ensure all commands can accept key names as bytes. This fixes issues
with BLPOP, BRPOP and SORT.
* All errors resulting from bad user input are raised as DataError
exceptions. DataError is a subclass of RedisError so this should be
transparent to anyone previously catching these.
* Added support for NX, XX, CH and INCR options to ZADD
* Added support for the MIGRATE command
* Added support for the MEMORY USAGE and MEMORY PURGE commands. Thanks
Itamar Haber
* Added support for the 'asynchronous' argument to FLUSHDB and FLUSHALL
commands. Thanks Itamar Haber
* Added support for the BITFIELD command. Thanks Charles Leifer and
Itamar Haber
* Improved performance on pipeline requests with large chunks of data.
Thanks tzickel
* Fixed test suite to not fail if another client is connected to the
server the tests are running against.
* Added support for SWAPDB. Thanks Itamar Haber
* Added support for all STREAM commands. Thanks Roey Prat and Itamar Haber
* SHUTDOWN now accepts the 'save' and 'nosave' arguments. Thanks
dwilliams-kenzan
* Added support for ZPOPMAX, ZPOPMIN, BZPOPMAX, BZPOPMIN. Thanks
Itamar Haber
* Added support for the 'type' argument in CLIENT LIST. Thanks Roey Prat
* Added support for CLIENT PAUSE. Thanks Roey Prat
* Added support for CLIENT ID and CLIENT UNBLOCK. Thanks Itamar Haber
* GEODIST now returns a None value when referencing a place that does
not exist. Thanks qingping209
* Added a ping() method to pubsub objects. Thanks krishan-carbon
* Fixed a bug with keys in the INFO dict that contained ':' symbols.
Thanks mzalimeni
* Fixed the select system call retry compatibility with Python 2.x.
Thanks lddubeau
* max_connections is now a valid querystring argument for creating
connection pools from URLs. Thanks mmaslowskicc
* Added the UNLINK command. Thanks yozel
* Added socket_type option to Connection for configurability.
Thanks garlicnation
* Lock.do_acquire now atomically sets acquires the lock and sets the
expire value via set(nx=True, px=timeout). Thanks 23doors
* Added 'count' argument to SPOP. Thanks AlirezaSadeghi
* Fixed an issue parsing client_list responses that contained an '='.
Thanks swilly22