OceanBase Database Enterprise Edition V4.1.0

V4.1.0 BP2

Version information

• Release date: June 15, 2023
• Version: V4.1.0 BP2
• RPM version: oceanbase-4.1.0.1-102000042023061309

Parameter changes

System variable changes

Bug fixes

• Fixed the issue where Error 4138 was reported when a large transaction for full-table update was hung.
• Fixed the issue where the optimizer did not push down predicates for decoding expressions that return constant values. This issue caused the indexes on the corresponding columns to be unable to be used.
• Fixed the issue where a major compaction for the standby tenant timed out due to a TSC time exception in the system.
• Fixed the issue that in an equijoin scenario where both tables contain IN conditions, the 4019 Size Overflow error is returned during SQL execution if the query range is excessively large because duplicate IN predicates cannot be removed.
• Fixed the issue where Parallel eXcution (PX) could not be enabled for the full-table update of a table without a primary key.
• Fixed the issue where jitters would occur in network threads in the case of high concurrency of DDL operations or high concurrency of network packets that exceed 2 MB in size.
• Fixed the issue where Error 4017 was reported due to a count leak that is introduced when the database system attempts to obtain the hidden primary key of a table that does not have an explicitly defined primary key.
• Fixed the issue where the login by using 1.1.1.1 failed after the tenant allowlist (ob_tcp_invited_nodes) was set to 1.1.1.1/24.
• Fixed the issue where characters in the metadata of LOB objects returned by a UDF expression were garbled after OceanBase Database was upgraded from V4.0.0 to V4.1.0.
• Fixed the issue where errors would be returned while accessing views that contain synonyms.
• Fixed the issue where in Oracle-compatible mode, the scale and precision of the data returned by the ResultSetMetaData API were inconsistent with those returned by it in Oracle Database.
• Fixed the issue where when multiple threads concurrently perform TRUNCATE and INSERT operations on different tables, an error is returned during the INSERT operation because a later schema version is broadcast by the replica set. The error code is 1526 and the error message is Table has no partition for value.

V4.1.0 BP1

Version information

• Release date: May 8, 2023
• Version: V4.1.0 BP1
• RPM version: oceanbase-4.1.0.0-101000052023050621

Parameter changes

Bug fixes

• Fixed the issue where, in some data distribution architectures, the major compaction would be blocked if the amount of encoded data written to the memory buffer exceeds 2 MB during a major compaction.
• Fixed the issue where the macro data seq command would generate duplicate data and therefore causes the backup to fail when columns were deleted in parallel tasks.
• Fixed the issue where the PL plan cache was not hit when a stored procedure calls another stored procedure in a system package.
• Fixed the issue where the operating system would fail to obtain the IP address of the NIC and the OBServer node would fail to start if more than 16 virtual IP addresses were configured for an NIC.
• Fixed the issue where the arbitration service could not start because of insufficient resources when small-specification (2c4g) Elastic Compute Service (ECS) instances were used. This issue is fixed from two aspects. In the system aspect, the lower limit of the required resources is reduced. In the O&M aspect, O&M engineers need to pay attention to the value of the vm.min_free_kbytes parameter, so as to prevent the started process from being killed by the operating system because the remaining resources are less than the lower limit.
• Fixed the issue where the NULLABLE and IS_VISIBLE fields in the information_schema.STATISTICS view were incompatible with OceanBase Database V4.0.0.
• Fixed the issue where no error was reported or the error was not reported in the same form as MySQL when the REGEXP function contains both binary and non-binary parameters.
• Fixed the issue where the use of an index created on a generated column deteriorated the performance.

V4.1.0

Version information

• Release date: March 31, 2023
• Version: V4.1.0
• RPM version: oceanbase-4.1.0.0-100001122023040322

Overview

OceanBase Database Enterprise Edition V4.0 is a beta version with a comprehensively upgraded distributed database architecture. OceanBase Database Enterprise Edition V4.1 is oriented to public cloud and open-source scenarios. In addition to its rich variety of features, this version is further optimized and enhanced in compatibility with MySQL 8.0 as well as the performance and ease of use at reduced costs. This version is also designed to support large-scale replication, and therefore can be comprehensively commercialized. This version provides the following core capabilities:

• Enhanced features

  This version supports more online DDL operations, tenant-level input/output operations per second (IOPS) isolation, the Latin1 character set, table locks, and deadlock detection.

• Enhanced compatibility

  • MySQL-compatible mode: Compatibility with the functions, variables, and SQL modes in MySQL 8.0 is improved. DDL operations are supported for FOREIGN KEY constraints. The column information of a view can be displayed.
  • Oracle-compatible mode: This version supports JSON data types, writes to Oracle Database by using DBLinks, and Chinese character sets and collations.

• Improved performance

  • Transactional processing (TP) performance

    • Compared with OceanBase Database Enterprise Edition V4.0, this version improves the comprehensive read-write performance by about 40% in running the Sysbench benchmark with 1024 threads.
    • Compared with OceanBase Database Enterprise Edition V4.0, this version improves the performance by about 8% in terms of tpmC (NewOrders) in running the TPC-C benchmark with 1000 warehouses.

  • Analytical processing (AP) performance

    • The TPC-H query performance is improved. Compared with OceanBase Database V4.0, this version improves the overall performance by about 17% in sequentially executing 22 SQL statements involving 100 GB of data.
    • The TPC-DS query performance is improved. This version takes about 161 seconds to sequentially execute 99 SQL statements involving 100 GB of data.

  • Data import performance When you import a 74-GB lineitem table through direct load to an OBServer node with the c6a.12xlarge specification, which provides 48 vCPUs and 96 GB of memory, the import speed can reach 165 MB/s.

  • Transaction performance The routing of distributed transactions is optimized based on OceanBase Database Proxy (ODP), improving the performance of distributed transactions.

  • Other performance improvements This version also improves the performance of the NESTED LOOP JOIN operator, the execution of Index Skip Scan, the parallel execution of the TRUNCATE TABLE statement, and the compilation feedback.

• Enhanced stability

  The large transaction processing capability is improved. The dependency on the Network Time Protocol (NTP) service is eliminated. Upper limits on the number of partitions and the LOB storage specification are supported.

• Enhanced high availability

  Physical Standby Database based on archive logs is supported, achieving a recovery time objective (RTO) of less than 8s in more scenarios. Arbitration replicas and tenant-level backup and restore are supported.

• Improved O&M

  • The monitoring capability is improved. End-to-end tracing and session status monitoring and diagnostics are supported.
  • Storage persistence is optimized. Adaptive compactions are supported and space optimization is implemented for small tables.
  • Views are standardized. Totally 73 data dictionary views and dynamic performance views are improved.
  • Online upgrades of OceanBase Database V4.0 series are supported.
  • Statistics are collected in real time.

• Product forms

  A standalone edition is released.

Updated features

Enhanced features

• Enhanced online DDL operations

  The OceanBase Database architecture of an earlier version supports limited types of online DDL operations. For example, primary key modifications are not supported, which causes inconvenience in database use. More online DDL operations related to data migration are supported since OceanBase Database V4.0. The following online DDL operations are supported:

  • MySQL-compatible mode

    • Modify the name and type of a column, including the conversion between the character, numeric, date, and other data types.
    • Add, modify, and delete a primary key.
    • Modify the collation of a table or column online.
    • Add and delete a foreign key.
    • Convert a regular table to a partitioned table.

  • Oracle-compatible mode

    • Modify the name and type of a column, including the conversion between the character, numeric, date, and other data types.
    • Add, modify, and delete a primary key.
    • Convert a regular table to a partitioned table.

  In addition, the transaction commit logs of DLL operations take effect immediately upon commitment. This ensures the stability and reliability of commit log replay. The GV$SESSION_LONGOPS view displays the execution progress of DDL operations in real time to help users observe and track the internal running status of time-consuming DDL operations.

• Optimized tenant-level IOPS isolation

  OceanBase Database supports tenant-level IOPS isolation since V4.0. Resources are isolated based on resource units between tenants and based on I/O categories within each tenant. This ensures that each task in a tenant uses only the allocated resources and does not contend for resources of other tasks. However, the I/O categories are fixed and cannot be customized based on tasks, and resources cannot be further isolated for different users in the same tenant. Therefore, I/O resources and CPU resources are incorporated into the Resource Manager framework for unified resource isolation on the UI.

  OceanBase Database V4.1 allows you to use the DBMS_RESOURCE_MANAGER package to customize resource isolation strategies for I/O resources. Specifically, you can call CREATE_PLAN_DIRECTIVE to set the MIN_IOPS, MAX_IOPS, and WEIGHT_IOPS parameters to define an isolation plan. You can define IOPS isolation plans for both Oracle- and MySQL-compatible tenants.

• Support for the Latin1 character set

  As the number of Community Edition users and OceanBase Database users outside the Chinese mainland increases, an increasing number of projects require the Latin1 character set. With the Latin1 character set, complex character set conversion can be avoided for migration projects involving OceanBase Database. The Latin1 character set is supported since OceanBase Database V4.1. The latin1_bin and latin1_swedish_ci collations are supported in MySQL-compatible mode, and the latin1_bin collation is supported in Oracle-compatible mode.

• Table locks and deadlock detection

  A table lock allows you to lock a table or partition in a specified manner to avoid data corruption caused by concurrent operations on the table or partition. OceanBase Database V4.1 supports more online DDL operations. Therefore, the table lock feature is provided to ensure data accuracy in concurrent DDL and DML operations. The LOCK TABLE syntax is supported in SHARE and EXCLUSIVE lock modes. Deadlocks due to lock contention can be detected.

Enhanced compatibility

Compatibility with MySQL

• Compatibility with MySQL 8.0 in system functions, variables, and SQL modes is improved.

  Most features of MySQL 5.7 are supported in OceanBase Database V4.0 and earlier versions. As OceanBase Database gains more experienced MySQL users in ApsaraDB for OceanBase and OceanBase Database Community Edition, the business system becomes more dependent on the features of MySQL 8.0. MySQL 8.0 has been released for over six years and its features are mature. Therefore, OceanBase Database systematically supports the features of MySQL 8.0 since V4.1 in terms of:

  • System functions

    • Character operation functions: BIN_TO_UUID(), CHARACTER_LENGTH(), LOAD_FILE(), IS_UUID(), UUID_TO_BIN(), and OCTET_LENGTH()
    • Datetime functions: ADDTIME() and DAYNAME()
    • Encryption and decryption functions: DECODE(), DES_DECRYPT(), DES_ENCRYPT(), ENCODE(), and ENCRYPT()
    • Performance schema functions: FORMAT_BYTES() and FORMAT_PICO_TIME()
    • Information functions: CURRENT_ROLE() and ICU_VERSION()
    • Window functions: BIT_AND(), BIT_OR(), and BIT_XOR()
    • Other functions: NAME_CONST()

  • SQL modes

    • Independent SQL modes: REAL_AS_FLOAT and TIME_TRUNCATE_FRACTIONAL
    • Combination SQL modes: ANSI, DB2, MAXDB, MSSQL, ORACLE, POSTGRESQL, and TRADITIONAL

  • Information schemas

    • View compatibility: View fields in OceanBase Database are the same as those in MySQL 8.0, but the field data types, widths, or constraints are adjusted for compatibility. For example, OceanBase Database provides the following views: VIEW_TABLE_USAGE, VIEWS, TABLES, STATISTICS, ENGINES, TABLE_CONSTRAINTS, REFERENTIAL_CONSTRAINTS, CHECK_CONSTRAINTS, PARAMETERS, ROUTINES, and PARTITIONS.
    • Performance improvement: The overall query performance of system views such as STATISTICS is improved based on local table indexes.

  • Float(m, d)

    • OceanBase Database V4.1 is fully compatible with the data types and precisions in MySQL 8.0.

  • DDL operations are supported for FOREIGN KEY constraints.

    By default, OceanBase Database checks foreign keys. To disable or enable foreign key check, modify the FOREIGN_KEY_CHECKS tenant variable. A FOREIGN KEY constraint requires that the constrained column must contain only values from the primary key column of another table. In OceanBase Database of earlier versions, the FOREIGN_KEY_CHECKS tenant variable takes effect only on DML operations. In OceanBase Database V4.1, the FOREIGN_KEY_CHECKS tenant variable also takes effect on DDL operations in the same way as in a MySQL database.

• Column information of views

  In a MySQL database, the column information of tables and views is stored as metadata in a data dictionary. You can query the INFORMATION_SCHEMA.COLUMNS view for the column information. OceanBase Database persistently stores only the column information of tables. OceanBase Database V4.1 avoids parsing complex view dependencies by dynamically parsing view definitions and displays the column information of views in the INFORMATION_SCHEMA.COLUMNS view. In Oracle-compatible mode, you can query the DBA_TAB_COLS view for column information of views.

Compatibility with Oracle

• Support for JSON data types

  OceanBase Database V3.2 supports JSON data types in MySQL-compatible mode. In actual business scenarios of customers, JSON data types also need to be supported in Oracle-compatible mode. The use of JSON data types and regular functions is beneficial for upper-layer business logic. Therefore, OceanBase Database supports JSON data types in Oracle-compatible mode since V4.1 to improve its multi-model data capabilities. The following table describes the capabilities of this version.

  Category	Description
  Data scope	This version supports the JSON scope defined in the ECMA-404 and IETF RFC 8259 (Oracle 21C) standards in strict mode and most JSON capabilities of Oracle in non-strict mode.
  Data types	This version allows you to save data of a JSON type as JSON text, and to save text with an IS JSON constraint as JSON text.
  DDL operations	This version supports the CREATE TABLE, ADD COLUMN, and DROP COLUMN DDL operations.
  Query and DML operation	This version supports access by using the dot notation or by using the json_value/json_query expression. It also supports the use of JSON text in SELECT, INSERT, UPDATE, DELETE, and REPLACE statements. You can partially modify the JSON text in an UPDATE statement.
  Functions	This version supports most functions of Oracle 19C, including JSON_VALUE, JSON_QUERY, JSON_OBJECT, JSON_EQUAL, IS JSON, JSON_ARRAY, JSON_ARRAYAGG, JSON_OBJECTAGG, JSON_MERGEPATCH, and JSON_EXISTS.
  Type conversion	This version supports conversion between strings or other scalar values and JSON values. The TREAT statement is required for explicit conversion.

• Support for writes to an Oracle database by using a DBLink

  OceanBase Database V3.2 supports read-only access to Oracle databases through a DBLink. When it comes to real-life database replacement scenarios, businesses often need to avoid making extensive changes to their existing logic. To address this challenge, OceanBase Database V4.1 has expanded its capabilities to enable write access between OceanBase databases and Oracle databases, using Oracle Call Interface (OCI) drivers and XA transaction capabilities. This is a significant advantage for businesses as it eliminates the need to make major changes to their existing logic during database replacement. The supported DML statements include INSERT, DELETE, UPDATE, and MERGE INTO. Note that writes using a DBLink are currently not supported in PL.

• Support for Chinese character sets and collations

  This version supports Chinese character sets and collations and the NLSSORT() sorting function. You can sort data by Pinyin, strokes, or radicals of Chinese characters.

Improved performance

• The data import performance is improved.

  The data import performance is an important metric in the online analytical processing (OLAP) field. Data analysis and big data applications are often characterized by a large data volume. The capability of a database to quickly load business data in the TXT or CSV format to provide external query and analysis services is a key competitive strength that affects the choice of customers in the AP field. In the general data import procedure of a database system, the SQL layer parses the data received from the client, performs error tolerance processing, and routes and forwards the data. Then, the data is encapsulated in transactions and persisted for storage. This procedure has the following bottlenecks: data is processed row by row and is filtered and parsed in the SQL layer, and a large volume of clogs is generated. Therefore, OLAP databases are specifically optimized. The most common practice is to use the direct load technology to shield the consumption and cost in SQL processing so that data is directly persisted to the disk. OceanBase Database V4.1 uses the direct load technology to streamline the data loading path by skipping modules such as the SQL layer, transaction module, and MemTables, to directly persist data to SSTables, thereby significantly accelerating data import. To use the direct load capability, you must add a hint in the LOAD DATA statement, for example, load data /*+ direct(true , 1024 ) parallel(64)*/ infile '/data/1/hits.tsv' into table hits Fields Terminated By '\t';. You can query the V$SESSION_LONGOPS view for the import progress. Primary key conflict and error tolerance processing is supported in different modes. At present, only CSV data files are supported. When you import a 74-GB lineitem table through direct load to an OBServer node with the c6a.12xlarge specification, which provides 48 vCPUs and 96 GB of memory, the import speed can reach 165 MB/s.

• The routing of distributed transactions is optimized based on OceanBase Database Proxy (ODP), improving the performance of distributed transactions.

  During business data modeling, transactions are generally processed on the same data node to avoid compromising the performance due to cross-node access of distributed transactions. In earlier versions of OceanBase Database, only requests in the same transaction can be processed on the same OBServer node. Therefore, ODP cannot route SQL queries based on the locations of data to access. This increases the latency in SQL processing within distributed transactions, thereby compromising the performance. During performance testing, the execution sequence of SQL statements must be adjusted to improve the performance. Much time and energy are consumed to reduce the negative impact caused by the routing of distributed transactions on the performance. OceanBase Database V4.1 optimizes routing for distributed transactions. ODP synchronizes the running status of transactions to OBServer nodes. Not all the running status information needs to be sent to the coordinator. A route can be calculated for each SQL statement. This ensures that SQL requests can be executed on any OBServer node in the cluster, thereby avoiding unnecessary processing latency. You can specify the enable_distributed_route parameter of ODP to enable routing optimization for distributed transactions, which is enabled by default. At present, routing optimization is not supported for eXtended Architecture (XA) transactions.

• Compilation feedback is optimized.

  By using the compilation feedback optimization technology, the overall performance is improved by about 10% to 15% in running the Sysbench benchmark.

• Other performance improvements:

  • The performance of the NESTED LOOP JOIN operator is improved. The performance of the multistage NESTED LOOP JOIN operator is improved by about 10 times in specific Group Rescan scenarios.
  • The execution of Index Skip Scan is optimized. When the NDV operator is used to calculate more than 0.5 million rows of data, the performance can be improved by about 10 times.
  • The parallel execution of the TRUNCATE TABLE statement is optimized, improving the statement execution efficiency by about 10 times.

Enhanced stability

• The large transaction processing capability is improved.

  Based on the new adaptive log stream architecture, the transaction engine is redesigned to solve multiple issues related to large transactions of most distributed databases, such as large transaction logs, a large number of participants, and slow transaction commits. The new transaction engine stably handles various tasks such as online transactions, batch processing, and data correction. It ensures database reliability in complex business scenarios.

• The dependency on the NTP service is eliminated.

  The new automatic leader election protocol no longer depends on the NTP clock. In earlier versions, the clock offset among all OBServer nodes has to be controlled within 100 ms. OceanBase Database V4.1 allows a clock offset of up to 2s and supports dynamic clock modification without affecting data accuracy or cluster stability.

• The limit on the number of partitions is extended.

  The new adaptive log stream architecture allows all partitions in a unit to share a resource group. A partition no longer needs to separately apply for resource reservation as in earlier versions, hence the higher utilization of system resources. You no longer need to specify the maximum number of partitions on an OBServer node based on the configurations. However, the maximum number of partitions is still limited by the available physical resources of a server.

• The limit on the DBMS_LOB storage specification is extended.

  In earlier versions of OceanBase Database, the size of a stored LOB is limited to 48 MB. This poses a forcible restriction on the use of LOBs. In this upgraded architecture, an LOB macroblock is split into multiple LOB metadata records for storage on the storage layer. When the data is retrieved, these metadata records are aggregated into a continuous buffer, which is then returned to the SQL layer for processing. This breaks the limit on the data storage size and extends the LOB size to 512 MB by using regular SQL statements. In Oracle-compatible mode, the LOB size can be extended to a TB level by using the DBMS_LOB package.

Enhanced high availability

• Physical Standby Database based on archive logs is supported.

  To continuously improve database high availability is one of objectives in the design of OceanBase Database. Based on the distributed architecture and multi-replica data management mechanism, OceanBase Database can meet the high availability requirements of business systems in most scenarios. However, for industries such as banking and insurance that have extremely high requirements on the high availability of databases, when a single cluster is deployed in three IDCs across two regions for geo-disaster recovery, services may still become unavailable due to poor network communication across regions. Therefore, OceanBase Database provides the Physical Standby Database solution to resolve the issues in cross-region geo-disaster recovery. In earlier versions of OceanBase Database, Physical Standby Database uses a cluster-level log transmission strategy. In this strategy, the primary and standby clusters must detect the log synchronization status of each other, and log coupling exists between tenants. In OceanBase Database V4.1, cluster-level Physical Standby Database is replaced with tenant-level Physical Standby Databases. The primary and standby tenants are fully decoupled and unaware of each other. Archive logs are asynchronously synchronized by using a third-party network file system (NFS) or object storage service (OSS). The primary and standby tenants support resumable transmission of logs, switchover, and failover. In the tenant-level Physical Standby Database solution, an OceanBase cluster can contain both primary and standby tenants to help business applications achieve more balanced resource utilization.

• The RTO is reduced to less than 8s in more scenarios.

  Compared with OceanBase Database V4.0, OceanBase Database V4.1 is architecturally upgraded and optimized in terms of fault detection for the distributed architecture and partition election capabilities, aiming to achieve an RTO of less than 8s in most scenarios. Specifically:

  • When the primary OBServer node crashes due to a fault, the standby OBServer node works with ODP to recover services.
  • When the log disk or data disk of the primary OBServer node becomes faulty, the primary node returns an error but does not go offline. The replicas on this node are migrated to a functioning node and a new leader is elected to provide services. In this case, the number of available replicas decreases and the faulty node needs to be manually recovered.
  • The arbitration replica is supported.

  In high availability deployment mode, we recommend that you deploy five replicas in three IDCs across two regions for a private cloud, and deploy two full-featured replicas and one read-only replica for a public cloud. This can ensure continuous high availability for businesses but also involves issues of high network bandwidth overheads and deployment costs. OceanBase Database V4.1 supports the arbitration service to solve the issue that the response time (RT) increases when a replica in the same region fails in the deployment scenario of three IDCs across two regions. The arbitration service can also effectively reduce the cross-region bandwidth overheads. Before you use the arbitration service, you must configure the arbitration OBServer nodes to start in arbitration mode when you deploy the cluster. Then, run the add arbitration command to add the arbitration service. At present, you cannot change the type of the arbitration replica or deploy a backup node for the arbitration node.

• Tenant-level backup and restore is supported.

  Multitenancy is a core capability of OceanBase Database. In most customer systems, multiple tenants are created in each cluster. Each cluster represents a business unit. The backup frequency and strategy vary with the business type and importance. OceanBase Database V4.1 supports tenant-level backup and restore. You can restore the backup data to a new tenant. In addition, the backup snapshot retention strategies are optimized to reduce the impact of backup process on disk space. Data backups and log backups are stored in different directories, which allows you to separately store them on media that differ in performance.

Enhanced O&M capabilities

• Online upgrades of OceanBase Database V4.0 series are supported.

  OceanBase Database V4.0 is an architecturally upgraded version, in which data storage, internal tables, views, parameters, and some behaviors are incompatible with those in earlier versions. Therefore, OceanBase Database cannot be directly upgraded from V3.x to V4.0. We recommend that you use OceanBase Migration Service (OMS) to migrate the data logic. System data compatibility and physical online upgrades are the key capability objectives of OceanBase Database since V4.1. These capabilities also meet customer requirements for upgrades, reduce the upgrade complexity and costs, and facilitate the application and promotion of OceanBase Database V4.1.

  OceanBase Database V4.1 adopts the rotating upgrade strategy.

  • After an upgrade is initiated, the cluster enters the hybrid deployment state. The system replaces the binary programs of OBServer nodes by zone and exits the hybrid deployment state until the binary programs of all nodes are replaced. At this time, binary programs of the later database version host the data and behaviors of the earlier version in compatibility mode.

  • After you run an upgrade command in a tenant, the system verifies and corrects the variables, system tables, and data. You must run the same upgrade command for all the tenants in the cluster one by one to finish the overall upgrade of the cluster. We recommend that you use OceanBase Cloud Platform (OCP) to upgrade your cluster online.

  OceanBase Database V4.1 supports the following upgrade scenarios:

  • Online cluster upgrades
  • Separate upgrades of the primary and standby clusters. We recommend that you first upgrade the standby cluster and then the primary cluster, and that you use the switchover command to adjust the status of the tenants in the cluster before the upgrade to ensure that they are in the consistent state.
  • Continuous upgrades since OceanBase Database V4.0, including the upgrades of functional versions and patch versions
  • Upgraded OBServer nodes do not support rollbacks. Therefore, we recommend that you back up the data and binary programs before the upgrade.
  • Tenant-level upgrades. We recommend that you upgrade all tenants in the cluster to the same version.

  Do not perform the following operations during the upgrade, which are allowed after the upgrade is completed:

  • DDL operations
  • Major freezes
  • Migration, replication, and load balancing
  • Physical backup and restore
  • Switchover and failover
  • Tenant creation
  • Monitoring is optimized to support end-to-end tracing.

  OceanBase Database has been put into business operation for a long time. Its internal data access links are very complex. When a timeout error occurs, it is often impossible to quickly locate the root cause. Therefore, an experienced O&M engineer is required to troubleshoot each process, which is hardly responsive and affects the O&M efficiency. OceanBase Database V4.0 provides the end-to-end tracing mechanism to help you locate errors in all processes of the whole link, from the business application to the client driver such as the Java Database Connectivity (JDBC) client or OCI driver, ODP, and OBServer node. You can specify the MODULE, ACTION, CLIENT_INFO, and CLIENT_IDENTIFIER parameters for applications by using PL/SQL statements or OBClient to identify the link in use. An O&M engineer can use a PL/SQL package to check related views and determine whether to enable end-to-end tracing and set diagnosis information output strategies. Diagnosis logs are generated and stored in ODP and OBServer log files in the OpenTracing data model. The engineer can parse the diagnosis logs to obtain relevant diagnosis information such as the execution time of each SQL and transaction in the whole link. OceanBase Database V4.1 supports the SHOW TRACE statement. You can specify the ob_enable_show_trace variable to perform interactive analysis and show the execution path of the previous SQL statement on the OBServer node.

• The session status can be monitored and diagnosed based on the Active Sessions History (ASH) report.

  OceanBase Database of an earlier version allows you to obtain the status information of an SQL statement that is being executed, such as the wait events. However, you can obtain only the last wait event in a session by querying the V$ACTIVE_SESSION_HISTORY view. OceanBase Database V4.0 supports a comprehensive ASH report that describes the relationship between sessions and wait events. The report contains not only the status of the SQL statements being executed, but also the status history of multiple metrics, such as SESSION, USER, SQL, and WaitEvent. The ASH report collects the status of all active sessions in the system at an interval of 1s. The collection process does not involve row locking or affect the execution of SQL statements. OCP analyzes the collected status information to help you learn about the system load and wait events in the past period of time, and identify exceptions in time.

• Views are standardized.

  Views and internal tables are redesigned. Totally 73 standard views are provided. We recommend that you query views for information such as the metadata of OceanBase Database.

• Statistics are collected in real time.

  The accuracy of statistics is the basis for the optimizer to generate a proper execution plan. Earlier versions of OceanBase Database already support statistics collection in the following ways:

  • Use the DBMS_STATS package or the ANALYZE statement to manually trigger statistics collection.
  • Configure scheduled tasks to automatically collect statistics.
  • Automatically collect statistics when the data amount exceeds the specified threshold. OceanBase Database V4.1 supports online statistics collection. You can enable this feature by using the _optimizer_gather_stats_on_load system variable or hints. If the business system supports the CTAS or INSERT statement, OceanBase Database updates incremental statistics in real time. This avoids system overheads caused by full table scans performed in regular statistics collection.

• Storage persistence is optimized.

  • Adaptive compaction: A compaction consumes a large number of I/O operations and CPU resources. Therefore, different compaction strategies are needed in different business scenarios. In the log-structured merge (LSM) storage system, the core objective of compactions is to achieve an optimal balance among write amplification, read amplification, and space amplification. In the adaptive compaction strategy, the system collects statistics about and analyzes the historical minor compactions and query operations performed on a table to extract the characteristics of the table, determines whether to schedule a major compaction for this table based on these characteristics, and completes a major compaction that is imperceptible to users.
  • Table space optimization: If a small table is split into many partitions, storage amplification appears in macroblock storage. This is prone to result in insufficient disk space when a large number of partitions exist. Multiple small SSTables are stored in a physical macroblock for compaction and storage. This achieves a utilization of about 4% and a fragment ratio of about 1.5% for the macroblock when a small table is split into many partitions, significantly improving the disk utilization.

Product form

A standalone edition is released.

Based on the integrated architecture of OceanBase Database V4.0 that supports both standalone and distributed deployment, we released a standalone edition for OceanBase Database V4.1. The standalone edition can be deployed by using OCP or OceanBase Deployer (obd), enabling independent software vendors (ISVs), small-sized customers, and community users to quickly get started with the new features of OceanBase Database V4.1. When the standalone edition can no longer meet your project requirements, business requirements, or high availability requirements, you can upgrade the standalone edition to a distributed edition online with ease.

Performance test report

The following table shows the performance test data of OceanBase Database Community Edition V4.0.

Test environment specifications

Version for test

Sysbench OLTP benchmark

Test plan

1. Use obd to deploy an OceanBase cluster. Install ODP on a separate server with a specification of ecs.c7.16xlarge, which provides 64 CPU cores and 128 GB of memory, and install Sysbench tools on another server to avoid resource contention.
2. Deploy the cluster in the 1-1-1 architecture, where the cluster has three zones and each zone has one OBServer node. After successful deployment, create the tenant and user required for running the Sysbench benchmark. The sys tenant is a built-in system tenant used for managing the cluster. Do not use the sys tenant to run the benchmark. Set primary_zone to RANDOM for the tenant, which indicates that the leader of the new table partitions is randomly assigned to one of the three nodes.
3. Launch the Sysbench client and run the point_select, read_write, read_only, and write_only tests.
4. Set the time parameter to 60s for each round of test. The number of threads can be 32, 64, 128, 256, 512, or 1,024.
5. The test dataset consists of 30 non-partitioned tables with 1 million rows of data. To be specific, set tables to 30 and table_size to 1000000.

Tenant specifications

• MAX_CPU = 26
• MEMORY_SIZE = 70 GB

Test results

Point select performance

Read-only performance

Write-only performance

Read/Write performance

TPC-C benchmark test with BenchmarkSQL

Test plan

1. Use obd to deploy an OceanBase cluster. Deploy ODP and the TPC-C tools on the same server to avoid insufficient stress on the client.
2. Deploy the OceanBase cluster in the 1-1-1 architecture, where the cluster has three zones and each zone has one OBServer node. After successful deployment, create the tenant and users required for running the TPC-C benchmark. The sys tenant is a built-in system tenant for managing the cluster. Do not use the sys tenant to run the benchmark. Set primary_zone to RANDOM for the tenant.

Tenant specifications

• MAX_CPU = 26
• MEMORY_SIZE = 70 GB

Software versions mysql-connector-java-5.1.47 BenchmarkSQL V5.0

Test configuration

• warehouse = 1000
• loadWorder = 40
• terminals = 800
• runMins = 5
• newOrderWeight = 45
• paymentWeight = 43
• orderStatusWeight = 4
• deliveryWeight = 4
• stockLevelWeight = 4

Test results

TPC-H benchmark

Test plan

• Use obd to deploy an OceanBase cluster. Deploy the TPC-H client on a server for stress testing. You do not need to deploy ODP. You can connect to any server during testing.
• Deploy the OceanBase cluster in the 1-1-1 architecture, where the cluster has three zones and each zone has one OBServer node. After successful deployment, create the tenant and users required for running the TPC-H benchmark. The sys tenant is a built-in system tenant for managing the cluster. Do not use the sys tenant to run the benchmark. Set primary_zone to RANDOM for the tenant. Load the data, sequentially execute 22 SQL queries, and take the average value.
• Size of the test dataset: 100 GB.

Tenant specifications

• MAX_CPU = 26
• MEMORY_SIZE = 70 GB

Test results

TPC-DS benchmark

Test plan

• Use obd to deploy an OceanBase cluster. Deploy the client on a separate server for stress testing. You do not need to deploy ODP. You can connect to any server during testing.
• Deploy the OceanBase cluster in the 1-1-1 architecture, where the cluster has three zones and each zone has one OBServer node. After successful deployment, create the tenant and users required for running the TPC-DS benchmark. The sys tenant is a built-in system tenant for managing the cluster. Do not use the sys tenant to run the benchmark. Set primary_zone to RANDOM for the tenant. Load the data, sequentially execute 99 SQL queries, and take the average value.
• Size of the test dataset: 100 GB.

Tenant information

• MAX_CPU = 26
• MEMORY_SIZE = 70 GB

Test results

Compatibility changes

Product behavioral changes

The following table describes the changes made based on OceanBase Database Community Edition V4.0.

Views, entity tables, and virtual tables

The following table describes the changes made based on OceanBase Database Community Edition V4.0.

Parameter changes

The following table describes the changes made based on OceanBase Database Community Edition V4.0.

Note

• An online upgrade from OceanBase Database V3.x to OceanBase Database V4.x is not supported.
• You can use OMS to migrate the data logic of OceanBase Database V3.x to OceanBase Database V4.1.
• An online upgrade from OceanBase Database V4.0 to OceanBase Database V4.1 is supported.

Recommended versions of tools

The following table describes the recommended component versions for OceanBase Database V4.1.0.

Limitations

• In OceanBase Database V4.1.0, load balancing is not supported after replica scale-out. Therefore, data that already exists before the scale-out cannot be distributed to the newly added nodes. This issue will be resolved in V4.2.
• OceanBase Database V4.1.0 does not support replicated tables. This issue will be resolved in V4.2.
• OceanBase Database V4.1.0 allows you to initiate data backup from the primary cluster. You will be able to initiate data backup from the standby cluster in later versions.
• In OceanBase Database V4.1.0, the minimum specification of a node in the production environment is 4C16GB. Each tenant requires more than 2 GB of memory and more than 1 CPU core.
• OceanBase Database V4.1.0 supports only full-featured replicas. Other types of replicas such as read-only replicas will be supported in later versions.
• In OceanBase Database V4.1.0, you can only increase the number of resource units (UNIT_NUM) of a tenant. You will be allowed to decrease the number of resource units in later versions.
• In OceanBase Database V4.1.0, if you execute the DROP DATABASE statement to drop a database that contains more than 200 tables, the operation may fail. We recommend that you first execute DROP TABLE and then DROP DATABASE. This issue will be resolved in V4.2.