Found a single line of byte code that cost $4M of compute per year.

Please tell us more. This sounds like a good investigative story.

We noticed that one line of byte code was crazy hot. We traced it back to the code, and realized it was a special case fallback.

They were summing 10mb of 8 bit encoded integers, in the 99.5th percentile of traffic, so no one noticed. It was using 10% of fleet compute.

We changed the encoding to 64bit; the data size stayed constant, the 99.5%tile latency plummeted, and I got a bonus. Took a week. My contribution to climate change.

A lot of stuff can hide past 99%.

That's awesome!

sleep(3)

I got a 7.5 inch wave share displaying my todos from todoist, today. Very easy. A few gotchas.

I ended up ordering a case from thingiverse because the case wave share sells doesn't fit the HAT model.

My goal is to not look at my phone in the morning, so having my todos on eink is nice. When I used paper I ended up with paper lists in 7 places.

To really scale I think the next step is something that could be run off an esp32, because power and cost are the current scaling limits.

For anyone who finds this later:

Home assistant integrates with esphome which supports waveshare if you get the esp32 interface rather than the raspberry pi HAT.

And todoist integrates with home assistant.

Basically all via yaml.