The same discussion about queue technology, for the third time in two years, with two of the three people having been present for the first.
# ADR-014: Redis for the job queue
## Status
Accepted, 2023-10-24
## Context
One application, two worker types, ~40k jobs/day, a team
of four with no dedicated operations capacity.
## Decision
Redis, with the framework's queue driver.
## Consequences
No routing topology, no per-message durability guarantee
beyond Redis persistence. Revisit if we need fanout to
independent consumers, or exceed ~1M jobs/day.
The consequences section is what makes it useful later, and the revisit condition is what makes it a decision rather than a preference. Ten records in eighteen months, each under a page, and the value is entirely in the ones that record a rejected option — the argument recurs because the reasons against the alternative were never written down, only the conclusion.