I’ve had a similar awakening a few months when realising that GitHub shows you the cost of free-for-open-source Actions usage.
-
@jwildeboer @janl this is where I see the value of local-first CI. Which should always have been prioritized, but of course runs against the lock-in strategy of shitty corps like Github^Wmsft.
One promising example:
https://ambient.liw.fi/@fedops @jwildeboer @janl They all went away again or enshittified ... thinking of Dagger and earthly, there are probably countless more.
-
RE: https://grrl.me/@soph/116586772000163018
I’ve had a similar awakening a few months when realising that GitHub shows you the cost of free-for-open-source Actions usage. On of my decidedly low-traffic but big test suite (173 checks) would cost thousands of dollars per month.
I have come to the inevitable conclusion that they’ve played us all for fools. We all have infinitely fast dev machines, but we need to rent a dev machine’s worth of servers every month to run an open source project?
Absolutely not.
@janl TBF, FreeBSD has their own fleet farm to build ports. It a regular problem to fund machines for it.
-
@janl TBF, FreeBSD has their own fleet farm to build ports. It a regular problem to fund machines for it.
@mms we have donated hardware for CouchDB as well.
-
Way back when, or in fact in 2026 for some projects, people would get a mailing list message to try the most recent tests. Some similar kind of "make test" seems to be what you are talking about? Or does it specifically have to be a bespoke CI in a box?
@zygmyd we do this for CouchDB, but our CI has a wider platform matrix
-
@janl TBF, FreeBSD has their own fleet farm to build ports. It a regular problem to fund machines for it.
-
@fedops @jwildeboer @janl They all went away again or enshittified ... thinking of Dagger and earthly, there are probably countless more.
-
-
-
RE: https://grrl.me/@soph/116586772000163018
I’ve had a similar awakening a few months when realising that GitHub shows you the cost of free-for-open-source Actions usage. On of my decidedly low-traffic but big test suite (173 checks) would cost thousands of dollars per month.
I have come to the inevitable conclusion that they’ve played us all for fools. We all have infinitely fast dev machines, but we need to rent a dev machine’s worth of servers every month to run an open source project?
Absolutely not.
@janl What confuses me is that there are TW of energy used in the social media in slagging off AI use for coding and attacking those who might admit to using it, but not a sausage of FOSS effort I can see on making a better tool than these expensive tech bro LLMs. Am really perplexed that FOSS efforts seems to have been defeated by next big thing: training & providing free models that compete with the expensive stuff in speed & quality & give users what they want: off-line, all params known.
-
@fedops @jwildeboer @janl Ah, I seem to have misremembered. They still advertise "local-first", even though they have a cloud offering. IIRC what interested me was their DSL which they abandoned.
Earthly went commercial (which somewhat precludes local-first, doesn't it?) but apparently didn't make it. 🫤
Haven't seen anything except Jenkins, GitLab, GitHub in the enterprise myself.
-
@fedops @jwildeboer @janl Ah, I seem to have misremembered. They still advertise "local-first", even though they have a cloud offering. IIRC what interested me was their DSL which they abandoned.
Earthly went commercial (which somewhat precludes local-first, doesn't it?) but apparently didn't make it. 🫤
Haven't seen anything except Jenkins, GitLab, GitHub in the enterprise myself.
@OmegaPolice yeah. We did use Teamcity for a while, but of course nowadays most stuff is azure devops because msft. Plus Jenkins, but on a server and not on dev machines.
-
@diazona it’s a bit bursty, so averaging is tough
@janl Ah yeah fair enough
-
The consequence for me is working on and supporting CI solutions thay can be run anywhere and still provide a reliable build environment.
I’m so done with everything we do is renting a variable markup on the cost of AWS, Azure or GCP.
There is not reason we shouldn’t be able to (paraphrasing) run `brew install ci-server` and have a system stood up that can run tests under almost any OS/arch/env combination on any* dev box.
*macOS licensing complicates things, but the point stands.
@janl Yeah, what the heck is up with Mac dev licensing? I wanted to cross-compile a Rust crate for Mac and found I would have to sign some sort of contract.
...so I just don't compile for Mac.
-
RE: https://grrl.me/@soph/116586772000163018
I’ve had a similar awakening a few months when realising that GitHub shows you the cost of free-for-open-source Actions usage. On of my decidedly low-traffic but big test suite (173 checks) would cost thousands of dollars per month.
I have come to the inevitable conclusion that they’ve played us all for fools. We all have infinitely fast dev machines, but we need to rent a dev machine’s worth of servers every month to run an open source project?
Absolutely not.
@janl@narrativ.es I mean, the cloud's always been good only for wildly variable loads, if you have a stable load, then owning your own stuff's always better in the long run
but "the cloud" shines in "oh dang I need 10x more resources this weekend for an event" sorta stuff (or "oh no the project got on HN/slashdot/youtube/whatever and blew up in popularity and now there's so many more people doing stuff")
also some minor other benefits, like supposed 24/7 availability, but github's stuff doesn't have that anyway, lol, and unless you have a very busy project, having even a whole day of downtime is probably not gonna be all too bad anyway, lol
and technically, having a multi-user build system is a more efficient use of resources but...you can afford a bit of waste when self-hosting, who cares that you have a machine idling 24/7 in the corner, the idle usage's pretty negligible generally -
@janl@narrativ.es I mean, the cloud's always been good only for wildly variable loads, if you have a stable load, then owning your own stuff's always better in the long run
but "the cloud" shines in "oh dang I need 10x more resources this weekend for an event" sorta stuff (or "oh no the project got on HN/slashdot/youtube/whatever and blew up in popularity and now there's so many more people doing stuff")
also some minor other benefits, like supposed 24/7 availability, but github's stuff doesn't have that anyway, lol, and unless you have a very busy project, having even a whole day of downtime is probably not gonna be all too bad anyway, lol
and technically, having a multi-user build system is a more efficient use of resources but...you can afford a bit of waste when self-hosting, who cares that you have a machine idling 24/7 in the corner, the idle usage's pretty negligible generally@6 I’m older than the cloud, I don’t need it explained to me
-
-
@OmegaPolice yeah. We did use Teamcity for a while, but of course nowadays most stuff is azure devops because msft. Plus Jenkins, but on a server and not on dev machines.
@fedops @jwildeboer @janl Never got to use TeamCity, unfortunately.
Ugh, AZDO ... had to touch that briefly, what a shit experience.
Tekton et al may be interesting in this context. While representing full Kubernetes buy-in, you can self-host a small cluster, even on your dev machine. It's just ... no layer of abstraction regarding builds/pipelines at all, and lots of infra instead.
-
RE: https://grrl.me/@soph/116586772000163018
I’ve had a similar awakening a few months when realising that GitHub shows you the cost of free-for-open-source Actions usage. On of my decidedly low-traffic but big test suite (173 checks) would cost thousands of dollars per month.
I have come to the inevitable conclusion that they’ve played us all for fools. We all have infinitely fast dev machines, but we need to rent a dev machine’s worth of servers every month to run an open source project?
Absolutely not.
On of my decidedly low-traffic but big test suite (173 checks) would cost thousands of dollars per month.
That's at GitHub's prices though. GitHub uses Azure. Even from the MS financials, calculating the exact price that the VMs cost to operate vs the cost to buy is hard, but the estimates I've read place the markup on the reserved pricing at around 200% (spot price is a lot lower). GitHub's markup is wild: it's close to a factor of ten over the raw VM price. This is why there's such a thriving ecosystem of third-party CI providers around GitHub: it's easy to undercut them.
All of this means, that thousands-of-dollars-per-month cost may only be costing tens of dollars to provide.
-
-
@cararemixed number of builds run is a terrible metric, not only due to resource waste.
@janl@fedops @janl I'm not suggesting this as much as refocusing. So much is placed on sophisticated CI infrastructure when there might be ways to stage release processes more gradually. We aren't deploying into production from the main branch of most OSS (and if we were we'd probably have our own validation, ie Cloudant services vs official Apache CouchDB releases).
I don't want to trivialize the effort either. Software is hard. We still don't need to assume one right way.
