It was a good run. Turns out we ended up not using this as much after the 2nd/3rd year. I moved back to Los Angeles and become a big fat egg. Starting 2022, I paid over 12-13$ a month for this, with the price steadily going up. This month is the last month I will pay for this - it racked up 16$. In total, I'd say my bill for this ended up being around ~350$, all said and done. The site will be left up and running, but it'll be going off of very stale data at this point.
I have
- Deleted my cloud functions. I still have the source code, but GCP has moved onto v2 functions now. V1 is pretty much obselete.
- Deleted my Postgres DB. It was a good run. I find it hilarious that I can't delete a stopped DB. I hurt mysef.
- Killed any cloud scheduler things. I shuddered at how bad this was, but hey, it WORKED!
- Persist my artifacts created in 2025, so the site still has something to read. No new updates.
What I learned:
- You can do start/stop instances, but you'll hate yourself when you come back.
- Storage is cheap. I had initially thought my rising costs were due to the legacy files I was storing. Nope.
- Running a DB, even if not very long, is not cheap. I originally did the start/stop functions trick with this to at least keep costs 1/100th of what it would have been to keep it running daily.
- Calling the maps API is also not cheap. (I'm surprised at how much it cost me)
-
Cloud function 1 - fetch map data A) fetch API key from env B) calls the (google maps) API for a given location segment C) makes a connect to DB to INSERT the segment
-
Cloud scheduler to call cloud function 1 on an interval
-
PostGres DB to store data from cloud function
-
Cloud function 2 - fetch reporting data through /GET for various segments A) connect to DB to fetch data B) return all segments