Skip to content

Amazon RDS :: Aurora provisioned

  • optimal read : upto ram 1024 GB, 40,000 mbps
  • burstable : 2 Gb ram , 4 cpu unit
  • Memory optimized -- SSD, eni-speed-high
  • single, multi-az mode

Aurora RDS :: serverless v2

A. Intro

  • engine : Postgres 3x and MySQL 5x
  • serverless v2 improvement : instant auto-scaling with no warm-up time.
  • forwarding writes from secondary to primary region. developer dont nned to write logic.
  • no capacity planning : configure min/max ACU (1acu === 2 gb ram , equ n/w and cpu)
  • scale to 128TB per db instance 👈
  • OLTP | rdbms
  • cluster design, fault tolerance by design:
  • cluster spans multiple AZ in region.
  • 1 primary + ( 5-16 )read replica(assign priority, can change at any time.)
  • if no reader, the primary itself re-created.

B. DR

  • RPO : 1 sec | RTP < 1 min 🎯
  • aurora has no stand-by instance thing 👈
  • has read replicas only in Aurora cluster, only replica can promote as primary, during DR.
  • read replica has 2 purpose : reader endpoint + availability
  • self-healing,
  • continuous s3 backup,
  • PITR

C. Advantages

  • include RDS adv.

1 Global Aurora database

  • cross region replicas in less than a sec. | single Database spans over multiple region.
  • 1 Primary Region (read / write)
  • Up to 5 secondary (read-only) in each region, replication lag is less than 1 second 👈
  • Up to 16 Read Replicas per secondary region
  • RPO: less than second
  • RTO: less than a minute
  • img_3.png
  • provides more comprehensive failover capabilities 🎯
  • Managed planned failover
  • Unplanned failover ("detach and promote")

2 integration ML service

  • SageMaker and Comprehend
  • fraud detection, ads targeting, sentiment analysis, product recommendations

3 Auto-scaling (storage and compute are separate)

  • storage scaling :: EBS volume - 10 GB to 128 TB . /64TB?
  • compute Instance :: type( eg: d.r3.large,etc), --RAM++, --cpu++.
  • Read replicas: (built-in, dont need to create CW + ASG, auto happens bts)
  • can add, CW metric --> triggers --> auto up/down read replicas

4 performance

  • AWS cloud optimized and claim 3x Performance improvement (on Postgres)
  • master + 6-15 Read Replica, with fast replication

5 Availability (cluster arch)

  • 6 copies for data access 3 AZ : cluster ( with reader and writer endpoint)
  • instant fail-over (<30s) + self healing from peer2peer replication.
  • img.png
  • img_2.png

6 backtracking 👈

  • rewind the DB cluster to any time you specify
  • faster compared to restoring a DB cluster via PITR or snapshot

7 cloning

  • usecase : create test env from prod.
  • faster than backup
  • restore uses copy-on-write
  • use same volume
  • for new changes additional storage allocated and data copied to it.

8 more

  • Isolation and security
  • Industry compliance
  • Push-button scaling
  • Advanced Monitoring
  • Routine Maintenance + Automated Patching with Zero Downtime
  • Backtrack: restore data at any point of time without using backups. (earliest :5 mim ago)

B pricing

  • 20% extra cost than RDS.

C. demo

- select engine
- select versions (so many available)
- template - prod (allow to configure everything)
    - admis + password
    - max i/o or standard
    - ec2 instance or serverless
    - choose : avialability - replica,etc
    - vpc, subnet, Ibv4
    - public access
    - sg
    - port
    - authentication
    - db name
    - ...

=== READY ====

- add more read replica
- add cross az replica 
- add region (global database)
- horizontal scaling policy (trigger : metric -CPU usage)
    - max 15 and min 1