Add RFC for time series gem downloads#48
Conversation
|
@segiddins this looks good to me, I think. my only remaining question is how we will ensure the main app doesn't break if e.g. timescale disappears. for the work outlined in this RFC, I think the only thing that might break is the background job that writes to timescale? if so, that means it's mainly a question for the future work that will display download counts and graphs. |
|
So my thought there is, we can have a cron job that copies all time download numbers to the main postgres database on some cadence? |
|
Oh yeah, that seems good. That plus a circuit-breaker type thing on the requests to the timescaledb so puma can continue to serve other rails requests sounds like it should work. |
I'm not quite sure what you mean by that? My thinking was we'd use stimulus to show download counts on pages, so as to be able to cache download numbers separately, and that would give us a natural place to keep reads from timescale off the critical path |
|
That's pretty much what I was thinking--something out of the critical path to show the numbers, something out of the critical path to write into timescale. Thanks for writing this up! |
| - Adding another datastore complicates contributing to the application | ||
| - Another datastore means we can't natively write joins against the download data | ||
| - Timescale cloud costs money | ||
| - There is no terraform provider for timescale cloud (yet) requiring us to write one or manually provision |
There was a problem hiding this comment.
Now, there's the official https://github.com/timescale/terraform-provider-timescale ✅
| - Another datastore means we can't natively write joins against the download data | ||
| - Timescale cloud costs money | ||
| - There is no terraform provider for timescale cloud (yet) requiring us to write one or manually provision | ||
| - This is still high cardinality data, and there's no knowing for sure how well timescale will compress it |
There was a problem hiding this comment.
I can help testing and benchmarking it. Is there any data available to seed the database? Or if anyone can share any numbers to give me some insights about how big is the cardinality.
3 participants
@indirect would appreciate your help rounding this off