ApsaraDB for Redis:ApsaraDB for Redis Enhanced Edition (Tair)

• DRAM-based instances use the multi-threading model and provide read and write performance approximately three times that of ApsaraDB for Redis Community Edition instances with the same specifications.
• DRAM-based instances provide multiple enhanced data structure modules such as TairString (including CAS and CAD commands), TairHash, TairGIS, TairBloom, TairDoc, TairTS, TairCpc, TairZset, TairRoaring, TairVector, and TairSearch. These instances eliminate concerns about storage structures and timeliness and allow you to focus on application development.
• DRAM-based instances provide high compatibility. DRAM-enhanced instances are fully compatible with open source Redis. You can switch from open source Redis to ApsaraDB for Redis without the need to modify application code.
• DRAM-based instances provide a variety of enterprise-grade features such as data flashback, proxy query cache, and Global Distributed Cache for Redis. For more information, see Use data flashback to restore data by point in time, Use proxy query cache to address issues caused by hotkeys, and Overview.

• Persistent memory-optimized instances provide ultra-high cost-effectiveness. The price of persistent memory-optimized instances is 30% lower than that of ApsaraDB for Redis Community Edition instances with the same capacity. The performance of persistent memory-optimized instances reaches 90% of that of native Redis databases.
• Persistent memory-optimized instances support the TairString (including CAS and CAD commands), TairHash, and TairCpc enhanced data structure modules.
• Persistent memory-optimized instances prevent data loss when power failures occur. These instances implement persistence for each command. The system will return a success response for each write operation only after the data is persistently stored. You can use persistent memory-optimized instances as in-memory databases instead of caches.
• Persistent memory-optimized instances optimize the append-only file (AOF) rewriting performance for large-sized Redis databases. These instances reduce the latency and jitters that are caused when Redis calls forks to rewrite the AOF.
• Persistent memory-optimized instances deliver high compatibility. These instances are compatible with most data structures and commands of native Redis databases.

• The issue concerning slow deletion of expired data from cluster instances after the number of data shards is changed is fixed.

• The stability of high availability is enhanced for read/write splitting instances.

• TairVector is available for cluster instances in proxy mode. Additionally, the FLOAT16 data type and queries with multiple vector indexes are supported.

• The Jaccard distance formula is supported for TairVector.
• The index_data_size and attribute_data_size parameters can be used in TairVector to calculate the memory usage of each index.

• The blocking issue of APIs is fixed to enhance the service stability.

• The stability of TairVector is enhanced.

• The stability of TairVector is enhanced.

• Redis 6.2.1.1 is supported. DRAM-based instances are compatible with Redis 6.2 and self-developed Tair data modules (except for TairSearch).
• Slow queries that involve the KEYS, SMEMBERS, HGETALL, or EXHGETALL command can be identified and isolated.
• Performance is optimized. Compared with commands run on ApsaraDB for Redis Community Edition instances, such as Pub/Sub commands and Lua commands, those run on Tair DRAM-based instances that use Redis 6.2.1.1 and have the same specifications enjoy two times the performance.
• TairVector is released. TairVector is a self-developed vector search data module that supports the Hierarchical Navigable Small World (HNSW) and Flat Search indexing algorithms. It provides high-performance real-time vector database service that combines storage and search capabilities.

• The lowercase parameter is available for term and terms queries in TairSearch.

• The IK and Pattern tokenizers are supported for TairSearch.
• The stability is enhanced.

• The stability is enhanced.

• The stability of data parsing and persistence in TairSearch is enhanced.
• The pipelining of the query builder client is supported for TairSearch.

• The MSEARCH feature of TairSearch is available.
• The stability is enhanced for multi-thread writing of TairSearch.
• The stability of instances in a cluster architecture is enhanced.

• The stability is enhanced.

• The stability is enhanced when child instances of a distributed instance are scaled.
• The accuracy of sort is improved in TairSearch.
• The query efficiency is improved for sparse documents in TairSearch.

• The stability of TairSearch is enhanced.

• The stability of TairTS is enhanced.

• The process of starting a cluster instance is optimized.

• The service stability is improved for instances that have the direct connection mode enabled.

• TairSearch is updated to improve the efficiency of JOIN queries.

• The configuration change stability is improved for cluster instances that have the direct connection mode enabled.
• The aggregation feature is supported in TairSearch to improve document write efficiency.

• The stability and execution efficiency of TairSearch are improved.

• The stability of TairTS is enhanced.

• The following issue is fixed: The migration of large keys may fail during configuration changes of cluster instances that have the direct connection mode enabled.

• The query efficiency of TairSearch is optimized.

• TairRoaring V2.2 is available.

• Statistics of Tair module commands can be collected and presented in latency histograms.

• TairRoaring V2 is available.
• TairSearch is available.

• The performance of TairRoaring is optimized.

• The following issue is fixed: Cluster instances may fail to be restarted in specific scenarios.

• The following issue is fixed: Statistics of SPOP commands are not collected when the real-time key analysis feature is used.

• The following issue is fixed to enhance stability: An excessive number of slow logs are recorded during configuration changes of cluster instances.

• Transparent Data Encryption (TDE) information can be included in the output of the INFO command.
• The stability is enhanced.

• The autonomous capabilities of implementing imperceptible slot migration during configuration changes of cluster instances are enhanced.

• The ray casting algorithm is optimized for TairGIS to yield more precise search results when you run the GIS.CONTAINS command.

• TairRoaring is available.
• The stability is enhanced.

• Memory leaks that may occur during the persistence of append-only file (AOF) rewrites are fixed.
• Statistics of queries per second (QPS) are classified. Three types of QPS can be calculated: read QPS, write QPS, and other QPS.

• The reliability of imperceptible slot migration is enhanced to strengthen stability.

• The following issue is fixed: Configuration changes may fail when the direct connection mode is enabled for cluster instances.

• The stability is enhanced.

• TDE can be used to perform real-time I/O encryption and decryption on Redis Database (RDB) files.

• The memory usage of TairHash is optimized.

• The stability is enhanced.

• Latency histograms are supported.

• The stability is enhanced.

• Specific functions of keys can be retained based on keys or key patterns when data is cleared. This feature can be used in the following scenarios:
  • Restore specific keys or key patterns when point-in-time data restoration is performed by using the data flashback feature.
  • Delete or retain the keyspace content based on keys or key patterns.

• The performance and migration speed during imperceptible scaling operations are optimized.
• Virtual IP addresses (VIPs) can be obtained over the Internet. This provides better support for clients that use direct connection mode.
• The format of large keys is optimized.

• By default, the statistics feature is enabled for large keys.
• The GT parameter is supported for TairString.

• The capabilities to migrate slots and implement imperceptible scaling are improved.

• The issue that the real-time statistics of large keys become inaccurate when keys with the same name are written is fixed.

• Statistics of large keys can be collected in real time.

• The failure detection capability of high-availability (HA) systems is improved when you call fork() in memory-intensive scenarios. This prevents long pauses that may occur.

• Metadata can be cleared after an active geo-redundancy link becomes available. The link is built based on Global Distributed Cache for Redis or Data Transmission Service (DTS). This accelerates subsequent synchronization operations.

• Memory leaks that may occur when the FLUSHALL command is frequently run when TairHash is used are fixed.

• When the illegal address error message is returned, the IP address of the client can be included in the error message.Tair An IP address whitelist can be configured for your instance based on the IP address prompt.Tair

  

• Flags such as FLAGS can be used in the syntax of TairString. This syntax is compatible with Memcached semantics.

• The failure detection capability of HA systems is improved.
  Note We recommend that you update the minor version to 1.5.0 or later for the latest optimization for this feature.

• Correct binary logs can be generated when TairString data expires. This prevents data inconsistency between the master and replica nodes.
• The abnormal switchover of HA systems is fixed. This issue occurs when TairHash still uses the active expire algorithm in read-only scenarios.
• The following issue is fixed: The system may not respond when an instance that was forcefully stopped is restarted.
• The following issue is fixed: Expired keys are deleted when RDB files are loaded to instances that have the data flashback feature enabled.

• Memory leaks that may occur when specific modules are loading are fixed.

• The output of the CLUSTER NODES command can be stored in the cache. This accelerates the command execution.

• The processing capabilities of specific modules are improved in specific scenarios.

• The IP addresses of clients can be passed through to operational logs and audit logs by proxy nodes. This helps you interpret logs and identify clients that have issues.

• The data collection capability is improved to minimize the impact on data shards when a large number of connections are queued up and I/O threads are busy.

• The latency is reduced when DTS is used to implement two-way data synchronization.

• The security group feature provided by Elastic Compute Service (ECS) is supported to simplify O&M. To allow ECS instances to access a Tair instance, you can add the security groups to which the ECS instances belong to the instance. You do not need to manually add the IP addresses of the ECS instances to the whitelists of the instance.
• The TairString module is updated to support more API operations (flags) that are compatible with Memcached semantics.

• The following issue is fixed: The BGREWRITEAOF command is interrupted when you use the data flashback feature to restore data to a point in time.
• Latency flag bits in audit logs can be modified to help you identify these bits in master and replica audit logs.

• Saved point metadata can be automatically cleared when data is cleared. This way, replicators can be quickly restored in scenarios where multi-way data synchronization is implemented based on Global Distributed Cache for Redis or DTS.

• The Replication part in the output of the INFO command shows role information, such as role:master. This allows Redisson clients to call the role information in specific scenarios.

• Statistics of hotkeys can be regularly recorded in logs.
• The data statistics feature is available for performance metrics. This feature allows the system to differentiate the QPS that are generated in read, write, and read/write synchronization operations. This improves the accuracy of the statistics.

• The kernel capacity to restore data to a point in time by using data flashback is optimized to simplify data restoration.
• Commands such as AUTH, ADMINAUTH, and CONFIG do not record sensitive information when these commands are used. This improves data security.

• The issue that deadlocks may occur when multi-threaded engines asynchronously close client connections is fixed.
• The issue that file descriptors in engines cannot linearly expand is fixed.

• The 64-bit hash algorithm can be applied to TairBloom.
• The final memory usage of TairBloom can be estimated to allow the system to record more accurate memory statistics.
• The exhgetAll2 interface can be used in TairHash to revise command output formats.

• The error message that is returned by instances when an whitelist is improperly configured is changed from (error) ERR invalid password to (error) ERR illegal address.
• Memory leaks that may occur when you use TairGIS to manage multiple polygons are fixed.
• The issue that the default path for TairDoc is incorrect is fixed.
• The issue that Pub and Sub commands may compete for resources on multi-threaded engines is fixed.

• The data flashback feature is supported. This feature allows you to restore instance data to a point in time within the last seven days. This helps prevent data loss caused by accidental operations, simplify O&M, and protect databases in real time.Tair
• TairGIS can be compatible with Redis GEO commands.
• TairBloom can be used in capacity security validation for the BFRESERVE interface.
• TairHash supports the following new features:
  • The NOACTIVE parameter is added to multiple commands including EXHSET, EXHEXPIRE, EXHINCRBY, and EXHINCRBYFLOAT. This parameter can reduce memory overheads in specific scenarios.
  • The MAX and MIN parameters are added to the EXHINCRBY command to define the upper and lower boundaries of the value range.
  • The noexp parameter is added to the EXHLEN command to return the true length of the EXHLEN command.
• The HINCRBY and HINCRBYFLOAT commands are supported for the hash structure. The transaction processing capability of these commands can be used to increase or decrease different fields that constitute a key at the same time.

• The use of data modules is optimized.
• The performance is enhanced when the JedisCluster client runs the MGET or MSET command on cluster instances.

• The following issue is fixed: Binary logs take up more space than can be provided by an instance.
• The issue that the system may not respond when hotkeys are evicted is fixed.
• The issue that the system may not respond due to double deallocation in TairHash is fixed.
• The issue that the system may not respond due to use-after-free (UAF) is fixed. This issue occurs when the audit log feature is disabled.

• The BITFIELD_RO command is added. This command significantly optimizes the performance of the BITFIELD command in read/write splitting scenarios.
  Note If the BITFIELD command contains only the get parameter, proxy nodes convert the command into the BITFIELD_RO command and route the new command to multiple backend data shards.

• The following issue is fixed: The replication process stops when specific complex commands in Lua scripts are run.

• The algorithm and performance related to traffic throttling are improved.

• The issue that the service stops because the congestion of client output buffers triggers server overload protection is fixed.

• Global Distributed Cache is supported. Global Distributed Cache is an active geo-redundancy database system developed in-house. Global Distributed Cache supports business scenarios in which multiple sites in different regions provide services at the same time. This system also helps enterprises replicate the active geo-redundancy architecture of Alibaba.
• Binary logs and their protocols are available to support capabilities such as active geo-redundancy.

• The issue that the output of the INFO command can contain the cluster_enabled information when the direct connection mode is used is fixed. This allows specific SDKs to automatically negotiate to enter the cluster mode.
• The issue that the number of clients that are being managed is calculated inaccurately is fixed.
• The issue that the system may not respond when a client is released is fixed.
• The issue that the system may not respond when a pipeline contains complex commands is fixed.

• The memory usage of TairHash is optimized.

• The issue that the system may not respond when traffic throttling is executed is fixed.

• The direct connection mode is supported. Clients can bypass proxy nodes to connect to instances by using private endpoints. This is similar to the connection to open source Redis clusters. The direct connection mode can reduce communication overheads and further improve the response speed.
• The identification logic for hotkeys is supported. This allows hotkeys to be accurately identified. The memory usage of hotkeys is also optimized.
• The EXCAS command is supported in optimistic locking scenarios.

• The core dump issue that may occur when pipelines are used is fixed.

DRAM-based instances of ApsaraDB for Redis Enhanced Edition (Tair) are suitable for business scenarios that require high concurrency, high performance, and a large number of read and write operations on hot data. Compared with ApsaraDB for Redis Community Edition instances, DRAM-based instances have the following advantages:

• Performance-enhanced instances use the multi-threading model and provide read and write performance approximately three times those of ApsaraDB for Redis Community Edition instances with the same specifications.
• Performance-enhanced instances provide multiple enhanced data modules, such as TairString (including CAS and CAD commands), TairHash, TairGIS, TairBloom, and TairDoc. These modules help you simplify code, improve business performance, and allow you to focus on innovation.

• Vulnerabilities of TairCpc V1.2.5 are fixed. These vulnerabilities may cause instance exceptions.
  Note We recommend that you update your instance that uses TairCpc V1.2.5 to this version.
• The stability is enhanced.

• The extension policies for primary hash tables are optimized.
• The performance of FLUSHALL and FLUSHDB commands is optimized.
• The performance of the rehashing function is optimized.
• The stability is enhanced.

• The issue that the GETBIT command may cause instance exceptions is fixed. This issue happens in V1.2.4.
  Note We recommend that you update your instance to this minor version.

• The RENAME and MOVE commands are supported by using the pena_rename_move_compatible_enabled parameter. The time complexity of the two commands is O(n) in Tair, which differs from that of the two commands in open source Redis.
• The stability is enhanced when cluster instances are scaled.
• The performance of specific commands is improved to address slow execution issues on large keys. The commands include SETBIT, SETRANGE, APPEND, and BITFIELD.

• The Errorstats - Selected field returned by the INFO command can be deleted.
• Latency histograms are optimized.
• The stability of semi-synchronous data replication is enhanced.
• The following issue is fixed: An error may be returned when commands such as ZUNIONSTORE or ZINTERSTORE are run on cluster instances that have the proxy mode enabled.
• The following issue is fixed: The SORT command may cause inaccurate calculation of persistent memory usage in specific scenarios.

• The redis_version value is changed to 6.0.9 in the INFO command output.

• TairCpc is available.

• The following issue is fixed: The key eviction order is abnormal when the MAXMEMORY_VOLATILE_TTL policy is specified for key eviction.
• The process of adding or deleting nodes for a cluster instance is optimized.
• The stability of semi-synchronous data replication is enhanced.

• The following issue is fixed: CAS commands cannot be persisted.
• The following issue is fixed: Client IP addresses cannot be displayed in slow and audit logs of cluster instances.
• The stability is enhanced.

• TairString is available.
• STREAM is supported.
• The intset encoding is supported for SETs to reduce memory overheads.
• Data eviction policies are supported.
• The performance is optimized, and the stability is enhanced.

• The following issue is fixed: SPOP command statistics are not collected when the real-time key analysis feature is used.

• Storage space occupied by large values is optimized to reduce used persistent storage.

• Tail latency that takes place during data writes is optimized.
• The stability is enhanced.

• The following issue is fixed: In specific scenarios, the usage of lists and hashes cannot be correctly calculated.

• The stability is enhanced.

• The speed at which cluster instances are scaled can be adaptively controlled.
  Note During scaling operations, the speed at which old data is cleared is well adapted to instance loads. For example, more CPU cores are used for clearance in low-load scenarios to increase the clearance speed. This feature keeps the CPU utilization at a high level during scaling operations but does not affect service availability.
• The performance of commands such as CLUSTER NODES is optimized for large cluster instances.

• The issue that abnormal slow logs are generated during scaling of cluster instances is fixed.
• The stability is enhanced.

• The storage of the LIST, HASH, SET, and ZSET data structures is optimized to reduce their usage of memory and persistent memory.

• Persistent memory can be fully utilized when memory is exhausted in specific scenarios.
• The stability is enhanced.

• Semi-synchronous data replication is supported. By default, this feature is disabled. If you want to enable this feature, submit a ticket.
  Note If this feature is enabled, logs are transmitted from a master node to a replica node after the data update that the client initiates is complete on the master node. After the replica node receives all logs, the master node returns the log transmission information to the client. If a replica node is unavailable or the communication between a master node and a replica node is abnormal, semi-synchronous replication degrades to asynchronous replication.
• In the output of the INFO command, the redis_version value is changed to 4.9.9 and the pena_version parameter is added to indicate the minor version.

• The stability is enhanced.

• Imperceptible scaling operations are supported. This allows slots to be migrated without perceptible impacts on your business.

• Ziplist encoding is supported in HASH and ZSET data structures to reduce memory overheads.
• The failure detection capability of HA systems is improved.
• Data migration of large cluster instances is optimized. Large cluster instances refer to instances that have a large number of data shards.
• The scaling stability of the cluster architecture is improved.

• Cluster instances are supported. These instances eliminate performance bottlenecks caused by the single-threading model of open source Redis. You can use cluster instances to support high-performance and large-capacity scenarios.
• The minor version of an instance can be used as the redis_version value in the output of the INFO command.

• The stability is enhanced.

• The usage of persistent memory can be included in the command output when you run the INFO command by using a standard account.

• The space occupied by embstr encoding is decreased to reduce memory overheads.
• The stability is enhanced.

Persistent memory-optimized instances of the ApsaraDB for Redis Enhanced Edition (Tair) are an in-memory database service developed based on persistent memory. These instances are compatible with open source Redis and provide large capacity. These instances have the following benefits:

• A persistent memory-optimized instance costs about 30% less than an ApsaraDB for Redis Community Edition instance.
• In scenarios where advanced memory specifications are used, specific issues can be resolved. These issues include high latency, high network jitter, and slow service data loading during fork operations that are triggered by AOF rewrites. You are not confronted with the trade-off between performance and persistence.
• Persistent memory-optimized instances support command-level persistence. A response is returned for each persisted write operation.
• Persistent memory-optimized instances are compatible with most of the data structures and interfaces of open source Redis. The persistence of data structures except for STREAM is supported.

• The following issue is fixed: The system may crash when the BRPOP, BLPOP, and BRPOPLPUSH commands are run.

• The transaction feature is supported. The MULTI and EXEC commands can be used in transactions. You can use the txn-isolation-lock parameter to manage this feature.
• The Lua scripting feature is supported. The EVAL, EVALSHA, and SCRIPT commands can be used to manage Lua scripts. You can use the txn-isolation-lock and #no_loose_lua-strict-mode parameters to manage this feature.
• Commands that manage bitmaps are supported. These commands include BITCOUNT, BITFIELD, BITOP, BITPOS, GETBIT, and SETBIT.

• The stability of data synchronization between master and replica nodes is optimized.

• The continuous read and write performance of instances that contain terabytes of data is optimized.

• The stability is enhanced.

• The stability of data synchronization between master and replica nodes is optimized.

• The stability of data synchronization between master and replica nodes is optimized.

• The efficiency of synchronization between master and replica nodes is optimized to provide higher throughput with lower latency.
• The resource usage of compact operations that are generated when a large number of large keys concurrently expire is optimized.

• The speed of master-replica switchover is optimized to improve stability. Master-replica switchover is also known as proactive HA.

• The issue that binary logs occupy a large amount of space is fixed.

• The storage parameters of ESSD-based instances that have terabytes of capacity are optimized to improve performance.
• The stability is enhanced.

• The canary release of the transaction feature is supported. The MULTI and EXEC commands can be used in transactions.
• The canary release of the Lua scripting feature is supported. The EVAL, EVALSHA, and SCRIPT commands can be used to manage Lua scripts.

• Notifications about keyspace events can be sent.

• The following issue is fixed: In specific cases, the FLUSHALL command may cause data inconsistency between master and replica nodes.

• The following issue is fixed to ensure data consistency between master and replica nodes: In specific cases, a master node and a replica node cannot be automatically reconnected after a disconnection.

• The cmd_slowlog_count metric is added to collect historical statistics of slow logs.

• Protection capabilities are improved in scenarios where disk space is fully occupied.

• Data verification for master and replica nodes is added.

• The defects of the RESTORE command that emerge in specific scenarios are fixed.

• The FLUSHALL and FLUSHDB commands cannot be run in serial.

• Memory management is optimized.

• Password-free access over a virtual private cloud (VPC) is supported.
• To prevent out of memory (OOM) errors caused by large transactions, the size of a single transaction can be limited by using the max-write-batch-size parameter.

• The redis_version parameter is added to the output of the INFO command.

• The issue that the number of connections is incorrectly calculated is fixed.
• The issue that the number of keys is incorrectly calculated when the RESTORE command is run is fixed.

• The issue that unexpected quits occur when the SETRANGE command is run is fixed.
• The issue that exceptions occur when keys are empty strings is fixed.

• The following issue is fixed: Connection management logs rapidly increase.

• The following issue is fixed: Operational logs rapidly increase.

• The stability of data synchronization between master and replica nodes is improved.
• The scanning performance for members of complex data structures is optimized.

• The following issue is fixed: The ZADD and ZINCR commands are inconsistent with those of open source Redis, and semantic errors may occur when multiple duplicate members are managed.
• The following issue is fixed: nil is returned for empty strings when the MGET command is run.

ESSD-based instances of ApsaraDB for Redis Enhanced Edition (Tair) are independently developed based on the TairDB key-value storage engine and use enhanced SSDs (ESSDs) to store data. These instances provide high-capacity, low-cost, and persistent database services. ESSD-based instances have the following benefits:

• ESSD-based instances store data in cloud disks to implement data persistence at a cost that can be as low as 15% that of ApsaraDB for Redis Community Edition instances.
• ESSD-based instances use multiple I/O threads and worker nodes to improve throughput on each server and support replication of binary logs between master and replica nodes.
• ESSD-based instances are compatible with most of the open source Redis commands.
• ESSD-based instances reduce the amount of memory that is reserved for the forks of open source Redis.