The role of a DBA in the NoSQL world has not diminished, it has only become different. The difference is based on the differences in the database model and the environment in which a database is used. RDBMS (relational database management system) databases are based on a fixed data model in comparison to the flexible data model of NoSQL databases. RDBMS databases are typically used in an enterprise environment in which the data management requirements do not vary much. NoSQL databases in comparison are used in a dynamic web environment. Because of the varying loads in a NoSQL web environment a NoSQL database needs to be highly scalable & available. Because a NoSQL database service must be provided without interruption NoSQL databases are designed to be failure (fault) tolerant. While some of these features such as high availability, scalability and fault tolerance are advanced level features in some RDBMS databases, these are features to be expected in a NoSQL database. Some of the main differences in DBA tasks for a NoSQL database are as follows. Fewer Configurations As NoSQL databases are primary used in web scale applications that must adapt to application needs in real-time most of the configuration has been automated. For example, in Couchbase server new nodes may be added from a console and auto-sharding automatically distributes the data across the nodes in the cluster. MongoDB provides automatic horizontal scaling and native replication. Provision for automatic failover exists in most NoSQL databases. No/Less Schema Modeling While modeling and designing the database schema is one of the main tasks for a DBA in a relational database, schema modeling is not required for a NoSQL database. A relational database DBA would typically also normalize database design and develop Normal Forms such as first normal form (1NF) and second normal form (2NF). While normalization is still relevant, NoSQL databases store data in a non-normalized form. Agile Methodologies Not having a fixed schema has another advantage. NoSQL characteristics are more suitable for agile development. A NoSQL DBA would need to be familiar with agile methodologies to adapt a schema or data model to evolving requirements. Backup and Recovery Backup and Recovery is still a DBA task in NoSQL databases. Just different commands and utilities are used. For example, MongoDB makes use of mongodump and mongorestore utilities and Couchbase server makes use of cbbackup and cbrestore. Backup and Recovery could be automated using tools similar to Oracle database’s RMAN. For example, MongoDB Cloud Manager could be used to schedule automated backups. Cloud Aware DBA A NoSQL DBA is a cloud aware DBA. Scalability, distributed big data, and high availability make NoSQL databases well designed for a cloud environment. A DBA would need to familiar with cloud providers such as Amazon Web Service (AWS) and Google Cloud Platform. Different Roles and Segregation of Duties In the NoSQL ecosystem DBAs would be assigned different roles as compared a RDBMS. The segregation of duties would also be different in NoSQL. For example, a Couchbase DBA could be a Full Admin with full access to all the database resources. A Couchbase Cluster Admin would only be concerned with managing a cluster and not have the top level access to the database. A Bucket Admin could be assigned to only administer Couchbase buckets such as setting bucket memory quotas. A View Admin could be concerned only with administering the views. No similar concepts of bucket and view apply in a RDBMS. Different Technology A Hadoop Admin would need to know about a completely new technology if previously a relational database Admin. Hadoop daemons such as NameNode, DataNode, ResourceManager, NodeManager, and ApplicationMaster have no parallels in RDBMS. A Hadoop Admin would need to know how the Hadoop cluster runs both in data storage and data processing. The number of nodes in a cluster would be significantly more, usually thousands of nodes. As early as 2011 Yahoo had 42,000 Hadoop nodes and hundreds of petabytes of storage. Networking would involve a different set up. Oracle database installation is typically on a single rack unless Exadata Database Machine is used. Hadoop installation could be on hundreds if not thousands of racks and an Admin would need to know about rack-aware replica placement policy, Rack Switches and Core Switches, and DNS to Switch Mapping. The Hardware and Software requirements are different in a Hadoop cluster in comparison to a relational database and would be variable based on whether the cluster is computation intensive, storage intensive, or balanced. Oracle database is typically configured in a Shared Server Architecture in which a database uses a common pool of resources. Hadoop cluster is based on Shared Nothing Architecture in which each node is independent; nodes do not share disk storage or memory. The tools used for cluster monitoring and metrics collection would be different. Conclusion The role of a NoSQL DBA has a similar importance as a RDBMS DBA, only the tasks would be different. While some admin tasks would be different some of the other admin tasks such as installation, backup and recovery stay the same; just different tools and commands are used. A Hadoop Admin would be using a completely different technology and on a much larger scale in terms of number of nodes in a cluster and the quantity of data. While some admin tasks have been automated, or the provision to automate exists, some admin tasks would still need to performed by a DBA such as designing and installing a new deployment, monitoring, backup & recovery, developing periodically running scripts to collect data for reports and analysis, debugging applications, test querying a database on staging and production, recommending & implementing indexing strategies, creating indexes, performance tuning & reporting.
↧