Attending: ajeffrey, nox, manishearth, jdm, jack, ssapin, negge, paul
Alan: have to pull out of October OSS EU launch announcement because trademark transition is taking longer than expected. How’s the website?
Paul: Can have the edits applied and ready for linux foundation review within 10 days.
Alan: How’s the infrastructure?
Simon: To be clear, I don’t want to lead infrastructure long term, just want to help with transition. I’ve removed a bunch of Amazon volumes which should reduce the monthly billing. We’ll see what that looks like in the future. I now have access to a tool called Cloud Health. We’ve discussed making Mac CI only run webgl tests and reduce the number of Mac minis at macstadium. As for moving infra out of Mozilla, we should only need to change the billing information on macstadium and cloudflare. We also have domain registration, so we need to pick a registrar and coordinate with mozilla to transfer. We could use cloudflare and not have to move anything else. For the AWS account it’s under the mozilla organization, and moving it out is complicated. I was told it might be easier to create a new account and move resources out. That doesn’t work for S3 buckets, which would require copying everything out and changing things that link to them. For CI, we have options - GitHub actions? Self-hosted task cluster? Other alternatives exist that haven’t been investigated yet. I plan to write up pros and cons and we’ll need to make a call. The people who want to maintain that long-term should be involved in that decision.
Alan: State of play of setting up @servo.org email addresses?
Simon: Done. Free account with alwaysdata, a French hosting company, and there’s a GitHub issue about this. We have addresses that redirect to other addresses for people who asked for it. We should decide who is eligible for this - TSC? Reviewers? Others? If new people ask for it, we should have an answer. We can set up aliases to multiple people, too, like [email protected]. There were questions about whether we should have GSuite, but if we already have email do we still need anything else?
Alan: Still hypothetical since there’s no money yet. Didn’t realize we’re not eligible for free tier when I filed the issue about it. Do we care about separation between personal/project stuff?
general lack of strong response
Alan: I think we can close the issue.
Simon: Speaking of payment, for transferring to cloudflare they want payment information. Do you expect we will have a credit card for the servo project?
Alan: Not sure how LF’s billing works. Does CF want a credit card?
Simon: Either credit card or PayPal.
Alan: Jack, any experience with this kind of thing?
Jack: For a startup, a founder would keep track of the amounts and expense it later if it matters. I could do that for Servo.
Simon: Would LF want to be the formal owner of the domain?
Jack: I don’t think they would care. The Launch Subcommittee should push for anything being owned by the Servo entity.
Alan: Cloudflare isn’t so bad. I get nervous about AWS, due to chance of infinite liability.
Simon: AWS docs for decoupling from an organization mention needing to talk to a lawyer because some part of some agreement no longer applies. That might be one reason why it’s easier to make a new account.
Jack: I haven’t found the AWS recovery codes yet. Still have one more place to check. Any idea when the account was created?
Alan: Don’t spend too much time on this. If we create a new account, are we going to keep doing all of the nightly builds?
Simon: S3 storage cost is $11/month for some number of terabytes of nightly builds.
Alan: When moving, do you incur costs for network traffic?
Simon: No moving, just copying, but I think there are optimizations for within the same region.
Alan: The old builds are useful for bisecting.
Simon: More worried about the name of the bucket, since it can’t be renamed. Have to copy to temporary bucket, remove the old one, create third bucket with original name, copy again.
Alan: What points to them?
Josh: Download.servo.org and the build system.
Simon: Maybe not bad to rename in that case. For CDN, some domains are marked as proxies. So we are using cloudflare as a CDN for GitHub pages for websites. Is that useful?
Alan: Might be useful for downloads page.
Simon: Downloads is on AWS CDN. Maybe an advantage? But Cloudflare’s CDN is for the github pages that aren’t the downloads. It’s all on the free plan so it doesn’t cost us anything. If we want to use another registrar besides cloudflare, it might be easier to move the DNS away as well.
lack of input
Simon: What we need now is to wait on the agreement with Mozilla for transferring the domain, and see if LF has any opinions on how to handle that domain, and make a decision for the CI platform.
Simon: Taskcluster is interesting because if we self-host, it’s just a matter of changing a few URLs. Of course then we have no commercial support and need to deal with staying up to date. For Github Actions, Macs get expensive quickly, and there’s the obvious conversion costs as well.. For self-hosted Github Actions, there’s a big warning about public repos mean that anybody who forks can run arbitrary code on your machines.
Jack: Github does have the ability to control what containers get launched. It’s an invite-only program right now, not public. Rust gets theirs donated and uses this ability. Libra does this as well, since it’s public repos. It’s worth looking at; the self-hosted runners work fine. You throw some stuff up on AWS and GitHub actions has instructions for how to run it.
Simon: One thing I wondered - with task cluster we can have short-lived instances on AWS that are spun-up on demand and shut down when idle. I expect this is less expensive than a fixed number of machines running all the time and sitting idle when the queue is empty. Can GitHub actions do this?
Jack: Don’t know; it’s not something that matters for Libra. Depends on the workload; can you estimate how long the bors queue is idle?
Josh: 20-22 hours most days.
Alan: Manish, do you know how rust got GitHub support for CI? Contact?
Manish: Complicated. Github open source team is in contact, and MS is good friends with the project, so there are multiple complementary tiers. Probably not a useful model to try to follow.
Simon: Probably hybrid as well, since there are multiple companies providing different machines for Rust.
Manish: You can talk to Pietro if you’d like.
Alan: That would be a good reason for picking GitHub Actions over a self-hosted task cluster, if the CI time were being donated.
Simon: If that happens, that’s a form of sponsorship.
Alan: The quid pro quo for that would be they would be a corporate member and we’d be getting payment in lieu. That would require estimating whether that would be premium or normal level of sponsorship. There’s something they would get in exchange for that donation.
Simon: Maybe it’s tricky to talk with them before we announce?
Alan: Possible to hold 1:1 conversations where we point out that we’re not public yet.
Jack: Free tier for GitHub actions is 2000 minutes per month. Won’t help with self-hosted runners. Mac builds are 10x the price of the other ones.
Simon: Given Mac and parallel builds, minutes get used very fast.
Jack: If you set up task cluster, you could have people like Nathan with lots of machines under his desk for a task cluster cloud instead of AWS.
Simon: Possible. Remember talking to Lars about desks on machines, and he was very against the idea.
Jack: I think it’s the maintenance rather than the costs. For Mac in particular, have to wipe the machine and reimage the machine. Macstadium handles that for you.
Simon: That happened several times. We had machines that stopped responding, so they replaced hardware for us.
Jack: If it was just linux we could probably do it, since they’re pretty reliable. Mac isn’t like that.
Paul: LF’s format is to use netlify for website hosting. We won’t do that, right?
Simon: We’ve been using GitHub pages and it’s been working fine.
Paul: There’s a build system for the website now.
Simon: Can you make a call? Look at netlify and see if it looks interesting.
Paul: With infra we have in place, we have nothing that could replace it?
Simon: Github already supports building static pages.
Paul: I will look into it.
Simon: For subcommittees, is it up to everyone to volunteer for those?
Alan: Yes. I volunteer for both infra and technical direction.
Simon: What does it mean to be on TSC but not technical direction?
Alan: TSC is too large to meet regularly and hold discussions. A proposed roadmap will come from the technical direction subcommittee, and the TSC will (likely) approve the recommendations after discussing them.
Simon: For infra subcommittee, I’d prefer to have asynchronous communication in GitHub issues, unless there are specific status updates necessary.
Alan: Other items - we’ve got press quotes lined up, there’s a press release that’s been edited. It’s mostly on hold because the deadline disappeared. We should probably give up on anybody who hasn’t responded to a request for a quote yet. This group should probably have a say on setting the membership rates for the project, too.
Simon: When I looked at the draft budget, every number seemed high to me. I have no experience with something like that, though.
Alan: Every non-technical number came from the LF. I suspect we won’t require the full legal cost budget, but it’s easy to run up nontrivial costs with lawyers. The technical numbers were envelope calculations.
Simon: Based on that, I feel like setting membership to $100k would be fine, assuming we get two sponsors at that level.
Alan: The only way to find out the real numbers are to pull the lever! That’s all for today!