Building highly available and fault tolerant sites on a budget

How to easily build systems that are fault-tolerant enough and available-enough so that you're not a slave to your servers' uptime.

Transcript

1. Building highly available
   and fault tolerant
   sites on a budget
   Senko Rašić, Good Code
   

   View Slide

2. View Slide

3. View Slide

4. If your server crashes
   I feel bad for you son
   I got 99% problems
   but the uptime ain’t one
   

   View Slide

5. Delegate
   

   View Slide

6. View Slide

7. 
   Ƭ
   

   View Slide

8. 
   Ƭ
   
   Ƭ
   

   View Slide

9. DNS
   master.example.com A 192.168.1.100
   slave.example.com A 192.168.1.101
   www.example.com CNAME master.example.com TTL 300
   

   View Slide

10. Reverse Proxy
    ǃ
    Ý
    Ý
    master
    slave
    haproxy
    nginx
    …
    

    View Slide

11. Round-robin DNS
    master.example.com A 192.168.1.100
    slave.example.com A 192.168.1.101
    www.example.com CNAME master.example.com TTL 300
    www.example.com CNAME slave.example.com TTL 300
    

    View Slide

12. Load Balancing
    ǃ
    Ý
    Ý
    master
    slave
    haproxy
    nginx
    …
    50%
    50%
    

    View Slide

13. ǃ
    
    
    master
    slave
    Ƭ
    State
    

    View Slide

14. image credit: http://myopia.podbean.com/
    

    View Slide

15. Warm Standby + Streaming Replication
    Replication Manager - http://repmgr.org/
    

    View Slide

16. Row-based replication with Global Transaction Identifiers (GTIDs)
    Automatic health monitoring & failover - mysqlfailover
    

    View Slide

17. Replica sets - http://docs.mongodb.org/manual/replication/
    

    View Slide

18. Master-slave replication (SLAVEOF) + Sentinel
    Redis Cluster
    

    View Slide

19. Budget
    Digital
    Ocean
    Linode Heroku* Azure* GCP*
    S3*
    (assets)
    DNS 2 x $5/mo 2 x $10/mo $3/mo
    Reverse
    Proxy
    3 x $5/mo 3 x $10/mo $3/mo
    PaaS $16/mo $50/mo $36/mo
    

    View Slide

20. Future considerations
    • Backup
    • Crash recovery
    • Logging
    • Datacenter-wide failures
    

    View Slide

21. CLOSE
    ENOUGH
    

    View Slide