On Fri, Sep 22, 2017 at 11:05:36AM -0400, Martin Owens wrote:
On Fri, 2017-09-22 at 08:44 -0500, Ted Gould wrote:
So, I think that this is probably fine for a first try. But I'm concerned that we're limiting our pool of potential sponsors saying that everything has to be on one machine, thus making that machine much more powerful. If we can get it, great
Possibly. In my research in what other open source projects of comparable scale to ours have arranged, so this arrangement is not atypical.
I'm not sure if we should detail a little more of a plan B having multiple sponsors or a set of smaller instances somewhere (asking for metal seems like asking someone to live in 1990).
What we have today is basically what you describe - services spread across a wide variety of hosts and providers. There is overhead in coordinating each of these individually, knitting them together, and constantly migrating one thing or another. And with each service individually hosted, adding a new service (or upgrading an existing service) necessitates finding yet another host. Current case in point being the hassles martin and I are having finding hosting for the new mailman3 instance.
My ulterior motive here is to gain for ourselves a beefy enough host that we gain a huge flexibility in how we manage, add, and upgrade our services. Something that'll enable us to minimize the project management overhead involved and streamline the administration to enough of a degree that we can return focus.
And yeah you're right being on metal is soo 1990, but it's a well known, solved problem, especially when we're dealing with FOSS services. Like you say, if we can get it, great, and I think we will have a solution that we can rely on for years to come.
Now for plan B's, well as I see it we have two paths. Finding a sponsor to provide VMs seems easy - OSUOSL is already giving us this. Problem is finding one scaled up enough to suit our needs; what we're seeing is it can involve a surprising amount of negotiation and fiddling to get it to ample enough resource levels. I wouldn't rule it out as an option, though. I haven't yet spotted other FOSS projects with VM setups of the level of flexibility and capacity that I'd like us to have, but I'm always open to new data.
The second path would be to just go with a commercial solution. Either metal or VM, there are tons of options at reasonable prices, and while it would be expensive, we do have the financial capacity to do it.
Well, there is a third path too - a FOSS service provider. Unfortunately we've outgrown what most can provide, although I'm following rumors that freedesktop.org is looking to massively renovate their infrastructure, so who knows. Maybe, although I don't see any slam dunks at the moment.
+1
I think we should pay for a machine to get our mailing list project done ASAP. I understand it invites fragmentation, but if other projects need a mailing list later on, it'll be easier to offer them a slice of a paid for cake than a donation box.
Again, I totally understand the near term optimization to solve this one particular hosting problem as convenient as possible, but I think we win more by looking a bit more broadly, and hopefully kill several birds with one solution.
From my perspective, the resource most precious to us is our time.
Fragmented services consume our time inefficiently and unpredictably. I do like the notion of us providing mailing service to other projects, but my main worry is that this imposes a time commitment on us - perhaps a minimal amount, but non-zero, particularly if some day we find ourselves wanting to transition yet again. Not to say I'm against it, but I think there's tangible costs to it beyond just money that we need to take into consideration.
Does this thinking make sense? I don't want to proceed with putting out a request if I'm not convincing enough to ensure we're on the same page here. But if you are willing to give it a shot and will support this plan, I'm prepared to put this call for sponsors next week.
Bryce