Network Partition and CAP therrem (Distributed Systems)

CAP

Consistency

read latest data

example:

write value 1 and sytem returns done, and read returns 1

Availability

always returns info

Partition tolerence

Network between nodes can be cut off

Still return something

CxA

sacrifice partition tolerence => can always talk to peer nodes

Can always return up to date info because peer is always available => consistent

Can always be available because peer is available => available

BUT, you will always be dealing with network paritions!

pick availability or consistency

PxC

Stop accepting trafic

Mitigation?

PxA

Risk of split brain

Algorithm to mitigate?

compensation

reconcilliation

ATM

withdrawal bounds check

allow overdraft by some number

amazon shopping cart

just take union (e.g. item deleted from version 1 re-appear when merged with version 2)

e.g. git

merge conflicts

Allow duplicate records

https://www.youtube.com/watch?v=gkg-FAEXIkY