Mastodon Feed
dysfun@treehouse.systems ("gaytabase") wrote:
okay so i am reading the page where they allege to describe how they achieve performance and i have questions.
yes, batching is very good. i can't help but notice that it is carefully designed to only deal with heavily batchable data. it feels like cheating but okay fine, whatever, you pick your domain and banks have lots of money.
but the single core... i assume they mean single writer. they must mean single writer, surely? my readers are effectively synchronisation free after some lightweight startup dance. to me this is like half of the point of having immutable data blocks and a single writer.
and it would be remiss of me not to point out that to actually handover the data properly, the client is required to supply a unique correlation key (indeed this is built into their client libraries!), so you're going to need some way to persist that id in your real database if you're going to correctly recover from it. so how can you ever unlock the performance they allege? why don't you just do it all in your proper database at that point?
i guess it only works if you go big on the microservices with own database thing?