Riverbend Dev Days, day 1

← How to Take Notes Like a Boss

Your ten minutes:

  • Add an aside to the keynote
  • Turn the retries question into Q: and A:
  • Define !SLO
  • Quote one line
  • Write one task per talk
  • Take The Docs Nobody Reads out into a file of its own

Boring Software

Priya Raman, keynote

innovation token
one new, unproven thing a team can afford to adopt; spend them where it counts
SLO
service level objective, the reliability you promise yourselves, like 99.9% of requests under 300ms
  • most outages come from the new thing, not the old thing
  • each team gets about three innovation tokens
  • choose boring tech for everything that isn’t your product
  • she credited Dan McKinley’s Choose Boring Technology
  • graph of incidents by age of dependency, drops off after year 2
So the first two years of a dependency are where we pay for it. Worth checking against our own incident log.
  • she kept coming back to SLO
  • QaudienceWhen is it worth spending a token?
  • APriyaWhen it’s the thing your customers pay you for.

Boring is a feature you only notice when it’s gone.

Priya Raman

We’ve spent at least five tokens on the build alone.
  • Count our innovation tokens before the planning meeting on
  • Read Choose Boring Technology

Backpressure for the Rest of Us

Tomás Ortega

backpressure
a consumer telling a faster producer to slow down
  • producer faster than consumer, something has to give
  • three choices: buffer, drop, or slow the producer down
  • pick one on purpose
  • an unbounded queue is choosing to crash later
unbounded queueOur importer does exactly this. That was the outage in August.
  • QTomásWho has a queue with no max?

  • AHalf the room.

  • QaudienceDo retries count?

  • ATomásRetries are just a producer you forgot about.

That’s the line I’ll remember.
  • Put a max on the importer’s queue, and decide what happens when it’s full

The Docs Nobody Reads

→ The Docs Nobody Reads