MappingRedisConverter now accepts data for byte arrays as single property to avoid flattening out binary data into a collection form. We still accept data in a collection form to retain compatibility.
Closes: #1981
Original Pull Request: #2015
RedisConnectionProxy is now public so it can be used properly for Java proxy creation across different class loaders. Previously, creating a Java proxy from a different class loader failed because of visibility/access restrictions.
Closes#2016
Add support for Jedis Streams using JedisCluster. Add fromMany(…) for non-pipelined usage to JedisInvoker.
Reformat code, add author tags. Extract Jedis-specific stream type converters to StreamConverters. Properly convert StreamEntry and StreamEntryID into list/map. Update tests.
See #1711
Original pull request: #1977.
We now accept a type hint when calling getAllOf(…) to avoid materializing null instances when the actual type hint cannot resolve to an entity.
Closes#1995
Original Pull Request: #1996
RedisNode can now be constructed using an empty hostname. This can happen when a node is in failover state. RedisNode exposes hasValidHost() to check whether the node has a valid hostname.
Also, introduce copy constructor to avoid mutations caused by the builder.
Closes#1985
Original Pull Request: #1991
Remove ZAddArgs.empty() method to avoid confusion with none() method. Use NONE constant only internally. Adapt calling methods.
Reformat code.
See #1794
Original pull request: #1988.
We now accept unknown custom Redis commands when using the Lettuce driver. Previously, custom commands were required to exist in Lettuce's CommandType enumeration and unknown commands (such as modules) failed to run.
Closes#1979
This commit makes sure to clean up resources when a previously expiring entity is persisted by setting the time to live to zero or negative.
In case a phantom copy exists for the changed entity, it is removed to free space on the server and prevent expiration events from being sent.
Closes#1955
Original Pull Request: #1961
We now support ZREMRANGEBYLEX for Synchronous, Reactive, ZSetOperations and RedisZSet.
Also provide a Kotlin coroutines variant via the extension for ReactiveZSetOperations.
Closes#1816.
Original pull request: #1968.
This commit fixes an issue where the order of elements in a set is not preserved when converting elements of the set.
Closes#1969.
Original pull request: #1968.
Move the no op identity converter to Converters and reuse those across the connection package.
Fix nullables and minor code format issues along the way.
See: #1951
Original Pull Request: #1960
We now use JedisInvoker to call Jedis and Pipeline methods for synchronous, pipelining, and transactional execution models. JedisInvoker captures the method invocation as functional utility and allows conversion of results:
Long result = invoker.just(BinaryJedis::geoadd, MultiKeyPipelineBase:geoadd, key, point.getX(), point.getY(), member);
Closes#1951
Original Pull Request: #1960
Use lambdas where possible. Deprecate unused methods. Favor LettuceInvoker conversion methods over List to List/List to Set converters.
See #1797
Original Pull Request #1948
We now use LettuceInvoker to call Lettuce API methods for synchronous, pipelining, and transactional execution models. LettuceInvoker captures the method invocation as functional utility and allows conversion of results:
Long result = invoker.just(RedisGeoAsyncCommands::geoadd, key, point.getX(), point.getY(), member);
List<byte[]> result = invoker.fromMany(RedisGeoAsyncCommands::geohash, key, members)
.toList(it -> it.getValueOrElse(null));
Closes#1797
Original Pull Request #1948
Reading and deserialization in StreamReceiver and StreamMessageListener is now decoupled from each other to allow fine-grained control over errors and resumption.
Previously, we used the Template API to read and deserialize stream records. Now the actual read happens before the deserialization so that errors during deserialization of individual messages can be handled properly. This split also allows advancing in the stream read. Previously, the failed deserialization prevented of getting hold of the non-serialized Stream record which caused the stream receiver to remain at the offset that fetched the offending record which effectively lead to an infinite loop.
We also support a resume function in StreamReceiver to control whether stream reads should be resumed or terminated.
Original Pull Request: #576
We now correctly consider the transactional state along with the requested bindings scope when binding RedisConnection for reuse.
Previously, enabling transaction support on RedisTemplate registered the connection unconditionally with TransactionSynchronizationManager expecting a TransactionManager to close/unbind the connection upon transaction cleanup. This behavior could easily lead to lingering connections when a transactional RedisTemplate was used outside of a managed transaction.
We now distinguish between closure-scope binding and transaction binding.
Closure-scoped bindings (RedisTemplate.execute(SessionCallback)) do not interfere with transactional resources if there's no transactionally bound RedisConnection.
Closure-scope however reuses a transactionally bound connection.
Transaction-scoped binding binds only a connection to the transaction if there's an active transaction. Without an ongoing transaction, RedisTemplate does no longer bind connections to TransactionSynchronizationManager.
Original Pull Request: #573
We now consider List and subtypes of List to be returned as MULTI return type. Previously, List subtypes were considered to represent the value type.
Original Pull Request: #571
Remove overly pessimistic null checks. Use constant when channel/body are empty to avoid allocations. Remove superfluous final keywords.
Original pull request: #559.
