Closed gedw99 closed 1 year ago
For example
https://github.com/maxpert/marmot
It just demonstrates a control plane to watch may db instances .,
this ones a little different in that it’s multi master. So your http proxy would I guess just load balance in that can and watch health data and then remove db instances.
Btw, not sure yet if this is applicable to duckdb since that uses s3 which is by nature clustered. your duckdb instance can of course also die, so I guess it is. But not sure yet . I don’t actually know if I can have 3 duckdb I stances pointing to 1 s3. Do you know ?
Is there a way to force writes to one db and read to another ?
this would enable scaling patterns where writes go to origin db and reads go to slave dbs.
it’s definitiv possible since you have a http proxy .
i guess the thing is that the concept of many db instances of the same db might need to be added.
Basically a map of db end points with a tag of what’s origin and what’s slave.
hope this is not out of scope of your projects.
Background:
Under the hood it’s common to setup using CDC based replication of db so that you can scale out your db. The classic way is to do the old master / slave pattern. if origin falls over then an election happens ( using some control plane like NATS agent ) and one of the slaves is voted to be the new master.
with worum of 3 db , then your down to 2 and have time ( hopefully ) to get your cloud to create another one off the latest snapshot off s3.