Relational Database Service (RDS)

Question 1

What are the different types of RDS database Instances in terms of size, memory optimisation etc?

Answer 1

db. m5 - general purpose type
db. r5 - memory optimised
db. t3 - burst instance

All can be different sizes, single or multi AZ

Question 2

What are the types of storage that can be used for database instances?

Answer 2

io1 - high end performance, lots of iops and low latency
gp2 - same burst profile as EC2, should be good for most
magnetic -long term historic use

Question 3

How is database storage billed?

Answer 3

Amount of storage allocated GB/month

e.g. 100GB for 1 month costs same as 200Gb for half month

Question 4

How do you connect to RDS in a multi AZ environment? Which instance does the connection always default to?

Answer 4

Only via CNAME defaulting to the primary always. Should an issue occur on the primary the CNAME will switch to. Standby cannot be directly used and does not scale performance only gives HA. Multi AZ can only be in the same region (other AZ’s in the VPC)

Question 5

Is RDS multi AZ high availability and fault tolerant?

Answer 5

Highly available but not fault tolerant, in a failover operation there will be some impact/disruption, it takes 60-120 seconds

Question 6

In a multi AZ RDS setup where are backups taken from?

Answer 6

always from standby/replica (removes performance impact as al connection are defaulted to primary)

Question 7

What types of backups are available for RDS?

Answer 7

• Automatic Backups
• Manual Snapshots - function like EBS snapshots, 1st is full copy, onward are incremental. Causes brief interruption in flow of data between compute resource and storage. In Single AZ may cause blip in multi it will use the secondary hence no effect.

Both utilise S3 (AWS managed buckets, not visible in console) therefore region resilient (replicated across multiple AZ’s in region)

Question 8

If you delete an RDS instance what happens to existing backup snapshots?

Answer 8

They remain on S3 buckets

Question 9

How is it possible to ensure an RPO for transaction logs of 5 minutes on RDS?

Answer 9

Database transaction logs are copied to S3 every 5 minutes therefore snapshot can be restored and transaction overlaid. All automated backups expire based on retention period set

Question 10

Limitation of RDS restore?

Answer 10

Restores aren’t fast, think about RTO

Why? Restore creates a new instance which will have a NEW ADDRESS (CNAME)

Question 11

How can you protect RDS against data corruption?

Answer 11

Snapshots (auto or manual)

Replication can replicate corruption

Question 12

In what instance would RDS use synchronous vs asynchronous replication to a replica

Answer 12

Synchronous - Multi AZ, copies the data to the pri and sec disks at the same time

Asynchronous - for read replicas, copies data to replica after copying to the primary

Question 13

Whats the max number of read replicas you can have per DB instance?

Answer 13

5x

For scale out read capability, each providing additional read performance

Read replicas can have read-replicas but lag starts to be a problem e.g. chained replicas

Question 14

How can you improve your RPO and RTO (availability improvements) for RDS?

Answer 14

Snapshots & Backups improve RPO

Read replicas offer near 0 RPO, can promote read replica quickly upon primary failure (TO READ AND WRITE) - low RTO. This works for failure only, watch out for data corruption

Question 15

In what scenarios would you use Aurora Serverless?

Answer 15

• Infrequently used applications
• New applications
• Variable workloads
• Unpredictable workloads
• Development & test databases
• Multi-tenant application - you’re providing

Question 16

When would you use Aurora Global Databases?

Answer 16

Ensuring low RPO if you need to failover cross region

Cross region DR and BC

Global read scaling - low latency performance improvements

Question 17

What are the benefits of multi-master mode in Aurora?

Answer 17

Foundation in building fault tolerant applications vs regular failover mode e.g. when one writer fails the app can continue, maintaining connectivity to writer, failover and load balance can be handled by the application as logic

Question 18

Whats the difference between Aurora provisioned and serverless?

Answer 18

Aurora provisioned vs Serverless

Aurora
- Storage based on whats used (high watermark e.g. hit 50GiB reduce down to 40 will be billed for 50 but can re-use the 10)

Serverless uses ACU’s (Aurora Capacity Limits):

• can set MIN & MAX ACU values and Aurora will scale within them
• Cluster adjusts base don load and can go to 0 (paused)
• Consumption billing per second basis (cost effective)
• Same resilience as Aurora (copies across 6 replicas)

Question 19

What does the backtrack feature in Aurora enable?

Answer 19

In-place rewinds to previous point in time, useful if data corruption occurs

If enabled for cluster can rewind, benefit is don’t have to create a new cluster as with a restore

Question 20

What is the benefit of using fast clones in AWS Aurora?

Answer 20

Uses a small amount of storage(data thats changed after the clone was created

Much faster

Allows you to create a brand new db from an existing, it DOES NOT make a 1 to 1 copy it references the original storage and only stores differences between the 2 - this may be you updating the storage in our clone or data updated in the orginal source