Fork me on GitHub

HI ALL! Do you know ops stuff well? cljdoc could use your help revisiting some of the decisions I’ve made as I set up cljdoc. Specifically: • move away from SQLite to remove state from the machine running the service • Reconsider load balancer / deployment setup (currently using Nomad) I’d love to make it dead easy to operate cljdoc to avoid downtime and ensure long term viability. If that sounds like something you’d enjoy helping with, let us know :)

👍 1

Curious, move from SQLite to …? Just using the bucket?


Probably just to using a database that is managed on a separate machine, like Heroku Postgres kind of stuff


Sounds interesting to me, I would like to help here 🙂 Opened up a discussion with the two points you already mentioned @U050TNB9F:


I’m trying to get Traefik to re-gen the SSL certs but so far no luck


Found from @ULHUNM4NM’s great GH discussion. While I think the decision on db is correct, I'd like to point out something that sounds incorrect. > Datomic is aimed at domains where time is an important component, which is not the case for cljdoc. This doesn't sound right. While Datomic has support for going back in transaction time, it is not intended to be used as a domain level construct. Val has a on this. Perhaps I'm misinterpreting the quoted sentence though 🙂


Sounds like Exoscale might be the hero 😁


Another option, although somewhat constraining, is ddb. It's incredibly cost effective, extremely performant, and a great fit if your access patterns are stable.


Yeah, that’s probably fair feedback, I had/still have very little experience with Datomic. Probably it’d work just as well 🙂


To be fair, a lot of people talk about Datomic’s time traveling and leave out the fact that you shouldn't really use it for anything besides auditing. Easy trap to fall into. Nowadays there's also datahike and Crux. Still seems like a DBaaS is the best path for the project.


Yeah, sticking with SQL will definitely be least effort

Cora (she/her)08:01:18

I wonder if we could get sponsorship from heroku. it doesn't get much simpler than heroku


And short of that heroku is probably too expensive. We have $860 in our OpenCollective, and the estimated yearly budget is around $200. I’m sure we could increase that if needed but $100/mo is gonna be tough


AWS RDS has a free tier that we could consider, as does Planetscale


Probably we could get a good discount / some free usage with Exoscale (who use Clojure and I know some people at)


Don't know about costs. But a while back used AWS Elastic Beanstalk with AWS RDS at a startup. But somewhere we can host the db and somewhere else we release a docker container or jar would be ideal. And have the service deal with the rest.


I expect a pro of the current setup is cost? And a con complexity?


Would be interesting to compare setup and costs with @U06SGCEHJ who runs clojars on AWS.


If we get enough ops-folks interested moving to a GitHub discussion might be a good idea.

👍 1

Clojars is ~US$320 a month, but most of that is s3 costs to serve the repo. I can provide a better breakdown in a few minutes.


AWS does (did?) have an open source sponsorship program. We applied and got a year of free service out of it, but they denied us the next year w/o explanation. My employer ( currently covers the hosting costs.

❤️ 1

That's generous, I do see mentioned who pays for hosting, but I often wondered how much these things cost to run. (I guess it is not much $ compared to the volunteer labor donated, but still, generous.)


It is generous, but a drop in the bucket compared to how much Shortcut spends on AWS monthly :)

simple_smile 1

@UE21H2HHD Good idea 👌, I opened up a discussion for this:

🙏 1
❤️ 1
Alexis Schad15:01:30

Hi there, I got a SEC_ERROR_REVOKED_CERTIFICATE error with firefox and safari mobile on (chrome doesn't care). Is it a known issue?