Is it inefficient for cadence to use the database to implement scheduled tasks?

Question

The scheduled tasks of cadence are stored in the database, so the problem of database polling must be involved. As the amount of data increases, how does cadence improve the query efficiency and the accuracy of delay time?

Answer 1

Cadence uses sharding to scale out. Each shard has its own timer queue. If a single DB instance cannot keep up, multiple DB hosts can be used and Cadence history service shards are distributed across all the hosts.

My talk Designing a Workflow Engine from First Principles goes into this in more detail. It talks about Temporal which is a fork of Cadence, but that part of the design is exactly the same.