Google SWE teaches systems design | EP17: Consistent Hashing

Two more videos until you guys get to meet my new girlfriend, her name's Cassandra (wow im lonely)

Recommended Viewing:

[username]

The relevant part of hashing is the digestive nature of it. You take an arbitrary length input and have a constant length output. That helps you direct it to pre-determined places

[gokusaiyan]

It's so lonely here just like my interview prep life. Making comment appreciating the video

P.S. love the self-trolling in the videos

[_gunna]

Something like HashModN is a fine strategy in the sense that it evenly distributes either data (partitioning) or requests (load balancing) among the nodes. For partitioning, new entrants into the cluster require balancing of data, which is why Consistent Hashing can be useful, since it minimizes required rebalancing (to k % n?). I don't see though why Consistent Hashing is an obviously good fit for load balancing using the same logic -- there'sd just requests being routed to nodes in the load balancer pool, not any state that needs to be rebalanced on new entrants to the pool, right? Or is the assumption that you're doing some kind of caching on the actual application servers in the LB pool, and that's the state that needs to be shuffled on changes in group membership?

[shuaisan]

I encountered a coding problem about implementing the consistent hashing and I couldn't figure it out. Can you point me to some sample codes? Thank you.

[utkarshgupta]

Can we use sharding in RDBMS also ? Or in RDBMS we have to pre-decide number of shards