Oracle Sharding is a scalability, availability and geo-distribution feature for applications that enables distribution and replication of data across a pool of Oracle databases that share no hardware or software. Applications elastically scale (data, transactions and concurrent users) to any level, on any platform, simply by adding additional databases (shards) to the pool. Oracle Sharding allows applications to perform high velocity relational transactions. It is applicable to applications (OLTP and Data Analytics) where the primary access pattern is via the Sharding key. Sharding supports multi-shard operations as well.

Oracle Sharding provides the advantages of an enterprise DBMS, including: relational schema, SQL, and other programmatic interfaces, support for complex data types, online schema changes, multi-core scalability, advanced security, compression, high-availability, ACID properties, consistent reads, developer agility with JSON, and much more.

In this blog, I will go over the various components of a Sharded Database (SDB) – Shards, Shard Catalog, Shard Directors (GSM), Global Service and Connection Pools. Figure 1 showcases the architectural components of Oracle Sharding:

Figure 1: Oracle Sharding Architecture with no SPOF

Sharded Database and Shards

Shards are independent Oracle databases that are hosted on database servers which have their own local resources – CPU, memory, and disk . No shared-storage is required across the shards. A sharded database is a collection of shards. Shards can all be placed in one region (datacenter[s]) or can be placed in different regions. A region in the context of Oracle Sharding represents a datacenter or a multiple datacenters that are in close network proximity.

Shards are replicated for High Availability (HA) and Disaster Recovery (DR) with Oracle replication technologies such as Active Data Guard or Oracle GoldenGate. For HA, the standby shards can be placed in the same region where the primary shards are placed. For DR, the standby shards are located in another region.

Shard Catalog

The shard catalog is a special-purpose Oracle Database that is a persistent store for SDB configuration data and plays a key role in automated deployment and centralized management of a sharded database. It also hosts the gold schema of the application and the master copies of common reference data (duplicated tables). The shard catalog database also acts as a query coordinator used to process multi-shard queries and queries that do not specify a sharding key.

All configuration changes, such as adding and removing shards and global services, are initiated on the shard catalog. All DDLs in an SDB are executed by connecting to the shard catalog.

An outage of the shard catalog does not affect the availability of the SDB. A shard catalog outage only affects the ability to perform maintenance operations or multi-shard queries during the brief period required to failover to a standby shard catalog. OLTP transactions are unaffected; they continue to be routed and executed on the shards. Oracle MAA recommends that a local Active Data Guard standby database be configured with Maximum Availability database protection mode with Fast-Start Failover and a remote physical standby database.

Shard Directors

The Shard Director is a regional network listener for clients that connect to an SDB. It maintains an up-to-date topology of the sharded database. Shard Directors route connections to the appropriate shards based on the sharding key passed during a connection request. The Shard Director is built upon the Oracle Global Service Manager (GSM). GSMs route connections based on database role, load, replication lag, and locality. For Oracle Sharding, GSMs have been enhanced to support management of SDB and routing of connections based on the location of data.

For a typical sharded database, shard directors (GSMs) are installed on dedicated low-end commodity servers in each region. Multiple shard directors should be deployed for high availability. In Oracle Database 12.2, up to 5 shard directors can be deployed in a given region. Oracle MAA recommends deploying three shard directors per region for availability and scalability.

Shard Directors provide the following set of functions

• Maintain runtime metadata about SDB configuration and availability of shards
• Publish SDB topology changes to connection pools
• Measure network latency between its own and other regions
• Act as a regional listeners for clients to connect to an SDB
• Manage global services
• Perform connection load balancing
• Publish runtime load balancing information to connection pools
• Monitor availability of database instances and global services, and notify clients via FAN HA events upon failure incidents
• Aid in schema propagation to all the shards

Global Service

A Global Service is a database service that can run across multiple databases. A global service allows clients to access data on any shard in the SDB. Global Services offer additional properties for sharded databases – e.g., database role, replication lag tolerance, region affinity, etc. You can create role-based global services: create a read-write global service that accesses data from primary shards, and a separate read-only global service for Active Data Guard shards.

Connection Pools

At runtime, connection pools act as shard directors by routing database requests across pooled connections. Key enhancements have been made to Oracle connection pools and drivers to support Sharding. Starting from 12.2, JDBC/UCP, OCI and Oracle Data Provider for .NET (ODP.NET) recognize the sharding keys as part of the connection check out. Apache Tomcat, JBoss, IBM WebSphere and Oracle WebLogic can use UCP support for sharding. PHP, Python, Perl, and Node.js can use OCI support.

I will go over the details of the data-dependent routing against an SDB in the next blog.

For more info, do visit the Oracle Sharding portal on Oracle Technology Network (OTN). Follow me on Twitter @nageshbattula

In the last of the three-part series on the capabilities of Sharding, I will cover the lifecycle management aspects of a sharded database.

Full Support for the Lifecycle Management of an SDB

Oracle Sharding platform allows shards to be added or removed online. It also supports the data reorganization at the chunk level.

Online scale-out with auto-resharding

Oracle Sharding provides the capability to elastically scale-out the sharded database via incremental deployment – shards are added online to expand the pool. In System Managed sharding, the addition of new shards to the pool will automatically trigger resharding wherein the chunks are automatically moved in order to attain balanced distribution of data.

Figure 6. Auto-resharding when shards are added

Ability to scale-in

Sharding allows to scale-in a given sharded database by removing the shards. In order to shrink the pool, you can use the “REMOVE SHARD” command. In 12.2.0.1, the chunks must be explicitly moved to the other shards before removing the shard.

Chunk Move or Split

In addition to shard addition and removal, Oracle Sharding supports splitting and moving of chunk from one shard to another. Chunks can also be moved from one shard to another when data or workload skew occurs without a change in the number of shards. Chunk migration can be initiated automatically or by the DBA.

Structural modifications to the Sharded Database are done transparently to the application. Connection pools get notified (via ONS) about split, move, add/remove shards and auto-resharding.

Patching Automation

Sharding enables patching all the shards with one command via opatchauto. OPatchauto supports all sharding schemes and replication methods. It also supports rolling mode and parallel mode.

Supports Command-Line and Graphical User Interfaces

Oracle Sharded databases can be deployed, managed and monitored with two interfaces:

GDSCTL:

GDSCTL is a command-line interface that provides a simple declarative way of specifying the configuration of an SDB and automating its deployment.  For example, just a few GDSCTL commands are required to create the SDB:

CREATE SHARDCATALOG

ADD GSM; START GSM (create and start shard directors)

CREATE SHARD (for each shard)

DEPLOY (for automated creation of shards and replication setup)

Oracle Enterprise Manager:

Enterprise Manager enables management of the life cycle of a sharded database with graphical user interface. For example, with few clicks, a shard director and shard software can be provisioned. Likewise, the deployment of shard directors and the creation of sharded database can be performed. With EM, you can also manage and monitor an SDB for availability and performance.

In summary, Oracle Sharding supports – data distribution, data colocation, data replication, direct as well as proxy routing and the complete lifecycle of a sharded database. Having studied the the key capabilities of Sharding, in the next blog post, let’s take a look at the components and architecture of the sharded database.

In the first part of the three-part blog series, we reviewed the automated data distribution and centralized schema management capabilities of Oracle Sharding. In part 2, I will cover the automated deployment of a sharded database (SDB) and data-dependent routing against an SDB.

Automated creation and replication of shards

In the Oracle Database 12.2.0.1 release, Oracle Sharding supports three automatically configured replication options: Data Guard, Active Data Guard, or Oracle GoldenGate (Oracle GoldenGate 12.3 supports Oracle Sharding).

Figure 2. Shard-level replication with Active Data Guard

Figure 3. chunkset-level replication with Oracle GoldenGate

Sharding supports two SDB deployment methods.

The first method is with the “CREATE SHARD” GDSCTL command. With this method, the shards and their respective listeners are automatically created. Once the primary shards are created, the corresponding standby shards are automatically built using the RMAN ‘duplicate’ command. After which, the Data Guard Broker configuration with Fast-Start Failover (FSFO) is automatically enabled. The FSFO observers are automatically started on the regional shard director.

The second method is with the “ADD SHARD” GDSCTL command. Many organizations have their own database creation standards and they may opt to deploy the SDB using their own pre-created databases (shards). The ADD SHARD based deployment method supports this requirement by simply adding the shards, which are pre-built by the user.

These two deployment methods support both the initial and incremental deployments.

Data-dependent routing

Oracle Sharding supports Direct and Proxy routing of database requests to shards.

Direct Routing:

Key enhancements have been made to Oracle connection pools and drivers to support Sharding. Starting from 12.2, JDBC/UCP, OCI and Oracle Data Provider for .NET (ODP.NET) recognize the sharding keys as part of the connection check out. Apache Tomcat, JBoss, IBM WebSphere and Oracle WebLogic can use UCP support for sharding. PHP, Python, Perl, and Node.js can use OCI support.

Sharding Key is used for routing the database connection requests at a user session level during connection checkout. Based on this information, connection is established to the relevant shard which contains the data pertinent to the given sharding_key. Once the session is made to a shard, all SQL queries and DMLs are supported, executed in the scope of the given shard and require no modification.

Upon the first connection to a given shard, the sharding key range mapping is collected from the shards to dynamically build the shard topology cache. This routing map is cached in the client. This allows subsequent requests for sharding keys within the cached range to be routed directly to the shard, bypassing the shard director. Such data-dependent routing of database requests eliminates an extra network-hop – thereby decreasing the latency for high volume OLTP applications.

When a connection request is made with a sharding key, connection pool looks up the corresponding shards on which this particular sharding key exists (from its topology cache). If a matching connection is available in the pool then the pool returns a connection to one of these shards by applying its internal connection selection algorithm. If a connection is not available, then forwarding the request with the sharding key (by the connection pool) to the shard director creates a new connection.

As illustrated in Figure 4, DB connection request for a given sharding key that is in any of the cached topology map, goes directly to the shard (i.e., bypassing the shard director).

Figure 4. Logical flow of direct routing – Connection pool as a Shard Director

Note: Super_sharding_key is needed only in the case of composite sharding.

The routing map automatically refreshes when a shard becomes unavailable or changes occur to the sharding topology. This is enabled by the Fast Application Notification (FAN) published by Shard Director via Oracle Notification Server (ONS).

Proxy Routing for Multi-Shard Queries:

Proxy routing is an ancillary usage pattern targeted for developer convenience. This requires connection be established to the coordinator. In Oracle Database 12.2.0.1, the shard catalog database assumes the role of the coordinator database. Once the session is made to the coordinator, SQL queries and DMLs are executed and require no modification. Proxy routing is suitable for the following scenarios:

»  When the application cannot pass the sharding key during connect

»  When the application needs to access data from multiple shards in the same query. For example, a query to aggregate data across all shards.

Figure 5. Logical flow of proxy routing

As illustrated in Figure 5, routing via the coordinator allows users to submit SQL statements without a sharding key value passed during connect. The Coordinator’s SQL compiler analyzes and rewrites the query into query fragments that are sent and executed by the participating shards. The queries are rewritten so that most of the query processing is done on the participating shards and then aggregated by the coordinator.

Applications should separate their workloads for direct vs. proxy routing. Separate connection pools must be created for these workloads.

In the second part of the three-part blog series , we have looked at the automated creation of shards and the replication setup. We also understood how routing happens against a sharded database. In the last part, we will go over the  lifecycle management aspects of a sharded database.

For more info, do visit the Oracle Sharding portal on Oracle Technology Network (OTN). Follow me on Twitter @nageshbattula