Secure Cloud Services — Week 5: Main Lecture

What four main factors should you consider when choosing a database?

Scalability, storage requirements, data characteristics and access patterns, and durability/availability/recoverability. Also consider cost and regulatory obligations.

What is capacity planning?

Analysing current capacity, predicting future needs, and deciding how to scale resources.

What is vertical scaling?

Using a bigger server with more resources, such as CPU or memory. It usually involves downtime.

What is horizontal scaling?

Adding more servers or instances to distribute the workload. It usually avoids downtime.

Why should you avoid underprovisioning and overprovisioning?

Underprovisioning can cause poor performance or application failure. Overprovisioning wastes money.

Why does data residency matter when choosing a database location?

Regulatory obligations, such as GDPR-related requirements for EU workloads, can affect where data should be stored.

How does a relational database organise data?

In tables containing rows and columns, with a defined schema.

How does a non-relational database organise data?

Using models such as key-value, document, graph or in-memory, often with a flexible schema.

What are the main strengths of relational databases?

Data integrity, ACID transactions, SQL and joins.

What does ACID stand for?

Atomicity, Consistency, Isolation and Durability.

What are the main strengths of non-relational databases?

Scalability, schema flexibility and high throughput.

Can relational databases scale horizontally?

Yes. For example, RDS read replicas distribute read workloads across multiple instances.

What is the main benefit of using a managed AWS database service?

AWS automates infrastructure and database operations, reducing the customer's maintenance workload.

What remains the customer's responsibility with a managed database?

Optimising the application and queries, and correctly configuring access and security.

What is Amazon RDS?

A managed relational database service that automates provisioning, patching, backups, failure detection and repair.

Which database engines are listed for RDS in the lecture?

Aurora with MySQL or PostgreSQL compatibility, MySQL, MariaDB, PostgreSQL, Oracle, SQL Server and Db2.

How should you decide which RDS instance resources to increase?

Identify the constrained resource, such as CPU or memory, and choose an upgrade that addresses it.

Where should the database sit in the lecture's application architecture?

Inside a VPC in a private subnet, rather than directly facing the internet.

What is the main purpose of a traditional RDS Multi-AZ deployment with a standby?

High availability through automatic failover to a standby database.

What type of replication does a traditional RDS Multi-AZ standby deployment use?

Synchronous replication.

Does the standby in the lecture's traditional RDS Multi-AZ deployment serve read queries?

No. It is a standby for failover, not a read-scaling replica.

What is the main purpose of RDS read replicas?

Scaling read performance by offloading read-only queries from the primary database.

What type of replication do RDS read replicas use?

Asynchronous replication.

How do traditional Multi-AZ deployments and read replicas differ during failure?

Multi-AZ automatically fails over to the standby. A read replica can be manually promoted to a standalone database.

Can RDS read replicas operate across AWS Regions?

Yes. The traditional Multi-AZ standby deployment described in the lecture stays within one Region.

What is the simplest way to remember Multi-AZ versus read replicas?

Multi-AZ means availability. Read replicas mean read scalability.

What is Amazon Aurora?

A cloud-native relational database managed by RDS, compatible with MySQL and PostgreSQL.

How is Aurora storage distributed?

Its cluster storage volume is replicated across three Availability Zones.

What Aurora storage limit does the lecture give?

Aurora storage automatically grows up to 256 TiB.

How many Aurora Replicas can an Aurora cluster have?

Up to 15 Aurora Replicas.

What are the roles of Aurora's primary instance and replicas?

The primary handles reads and writes. Replicas handle read-only queries and can act as failover targets.

When is Aurora Serverless v2 useful?

For variable or unpredictable workloads, new applications, and development or testing, because it automatically scales capacity.

What is Amazon RDS Proxy?

A fully managed, highly available database proxy for RDS and Aurora.

How does RDS Proxy help with database connections?

It pools and shares connections so the database handles fewer, longer-lived connections.

When should you consider RDS Proxy?

When an application has too many connections, frequently opens and closes connections, or holds many connections open.

How does RDS Proxy improve resilience and security?

It reduces failover disruption, supports IAM authentication, and can keep database credentials in AWS Secrets Manager.

What is the main purpose of RDS automated backups?

Point-in-time recovery using daily backups and transaction logs.

What is the maximum RDS automated backup retention stated in the lecture?

Up to 35 days.

How do manual RDS snapshots differ from automated backups?

Manual snapshots are user-initiated, restore a known state, remain until deleted, and can be shared subject to applicable restrictions.

Which controls protect access to an RDS database?

A private subnet, security groups restricting connections, and IAM controlling who can manage RDS.

How should RDS data be encrypted?

Use TLS for data in transit and AWS KMS encryption for data at rest.

How can you encrypt an existing unencrypted RDS database using the lecture's method?

Create a snapshot, copy the snapshot with encryption enabled, and restore a new database from the encrypted copy.

What is Amazon DynamoDB?

A fully managed, serverless NoSQL database supporting key-value and document data models.

What performance characteristic does the lecture associate with DynamoDB?

Single-digit millisecond performance at any scale, with automatic scaling.

How does DynamoDB handle encryption and access?

Data is encrypted at rest by default, and access is controlled through IAM rather than database usernames and passwords.

What are DynamoDB tables, items and attributes?

A table contains items. An item is similar to a row. Attributes are the item's key-value data fields.

What keys can form a DynamoDB primary key?

A partition key alone, or a composite primary key consisting of a partition key and a sort key.

How does DynamoDB support a flexible schema?

Items must contain the required primary-key attributes, but other attributes can differ between items.

Why use Device ID as the partition key and Timestamp as the sort key for sensor readings?

It groups readings by device and distinguishes or orders readings using their timestamps.

What is a DynamoDB global secondary index, or GSI?

An index that provides an alternate query pattern using a different partition key and optionally a different sort key.

What is a DynamoDB local secondary index, or LSI?

An index using the same partition key as the base table but a different sort key.

When can GSIs and LSIs be created?

GSIs can be created after the table exists. LSIs must be created when the table is created.

What GSI and LSI limits does the lecture give?

Up to 20 GSIs and up to 5 LSIs per table.

How do GSI and LSI read consistency differ?

GSIs support eventually consistent reads only. LSIs support eventually consistent or strongly consistent reads.

How do GSI and LSI capacity differ?

A GSI has its own capacity. An LSI uses the base table's capacity.

How could you query sensor readings with Error status = High without scanning the whole table?

Create a GSI with Error status as its partition key, then query that index.

What are DynamoDB global tables?

Multi-Region, multi-active replicated tables where every replica accepts reads and writes.

What are the benefits of DynamoDB global tables?

Fast local access across Regions and resilience against a Region outage.

What are DynamoDB Streams used for?

Supporting event-driven applications by exposing changes made to table items.

Which DynamoDB recovery window does the lecture give?

Point-in-time recovery covering up to 35 days.

Which AWS database services suit structured transactional workloads, or OLTP?

Amazon RDS and Amazon Aurora.

Which AWS service suits analytics over huge datasets, or OLAP?

Amazon Redshift.

Which AWS database suits JSON document workloads?

Amazon DocumentDB.

Which AWS database suits wide-column or Cassandra workloads?

Amazon Keyspaces.

Which AWS services suit in-memory workloads?

Amazon MemoryDB and Amazon ElastiCache.

Which AWS database suits highly connected data, such as social-network relationships?

Amazon Neptune, a graph database.

Which AWS database does the lecture associate with time-stamped IoT or operational data?

Amazon Timestream.

What is AWS Database Migration Service, or AWS DMS?

A service for migrating database data, with support for ongoing replication while the source remains online.

What is a homogeneous database migration?

A migration between the same database engine, such as MySQL on EC2 to RDS for MySQL.

What is a heterogeneous database migration?

A migration between different database engines, such as Oracle to Aurora PostgreSQL.

What must happen before migrating data between different database engines?

Convert the schema and code using AWS SCT or DMS Schema Conversion, then migrate the data using DMS.

What endpoint requirement does the lecture give for AWS DMS?

At least one endpoint must be on AWS.

Besides a one-time migration, what can AWS DMS do?

Continuously replicate data, for example into an Amazon S3 data lake.

How does the Well-Architected performance efficiency pillar apply to databases?

Choose the database based on data characteristics and access patterns, load test, and evaluate trade-offs such as eventual consistency.

How does the Well-Architected security pillar apply to databases?

Use secure key management with AWS KMS and enforce encryption at rest.

How does the Well-Architected cost optimisation pillar apply to databases?

Right-size the type, size and number of resources, and consider Aurora Serverless to avoid overprovisioning.

In the sample exam question, which database supports a highly available relational workload starting at 8 TB, growing daily, and requiring at least eight read replicas?

Amazon Aurora, because it is relational, its storage grows automatically, and it supports up to 15 read replicas.

Why are DynamoDB and Neptune wrong answers in the lecture's sample exam question?

DynamoDB is non-relational, and Neptune is a graph database rather than the required relational database.

Why is Redshift wrong in the lecture's sample exam question?

It is designed for data warehousing and does not provide the required read-replica arrangement.