Hi Yitz, Yes, reproducibility is part of the plan, and we've been trying to do so as sensibly as possible. As we have taken over the servers, there are *many* unautomated things, which has turned out extremely painful to fix and figure out.
For now, we've worked on having as 'small of a migration as possible'. For example, for ghc.haskell.org, we merely rsynced the existing Debian 7.6 disk to the new VM, and booted it. We will be upgrading it to a new Debian Jessie machine eventually, probably. Hackage will be relatively easy to finish migrating - I helped deploy it, and it is mostly a bare server, a Haskell webserver application, and nginx. The most painful part is it has a lot of data. www.haskell.org will be the most difficult, because it is monolithic and hosts many different sites and a few separate services (mysql, mailman, postfix, hoogle, etc). We'll probably do a migration similar to ghc.haskell.org, and we'd like to split up that configuration eventually. In the mean time, Ricky Elrod, one of our admins, is automating things using Ansible, which is nice and incremental, and can handle ad-hoc scenarios rather well.[1] I'm well aware Rackspace could change their minds at any time; until then, we might as well leverage them sensibly, and hopefully it will make moving elsewhere easier with tools like this. For the record: I've just entered a talk about securing another year of free space for us, which I expect will go just fine. (In the long run, several of the administrators actually want to move to NixOS for Haskell.org, which is really the ultimate end-game in terms of reproducible configurations. I'm slowly working towards this theoretical future myself[2].) [1] https://github.com/haskell-infra/ansible [2] https://github.com/haskell-infra/auron On Tue, Sep 9, 2014 at 2:13 AM, Yitzchak Gale <[email protected]> wrote: > Austin wrote: >> As I mentioned last year, we've gotten support from Rackspace for >> Haskell.org. In particular, this weekend we moved ghc.haskell.org to >> Rackspace... >> I plan on migrating Hackage to a new dedicated server this week... >> Furthermore, we have an older server, the one running >> lambda.haskell.org and community.haskell.org, running with Hetzner >> too. I think we can migrate these quickly as well... > > Thanks so much for this! > > We are also grateful to Rackspace for their generosity, and we > should give them plenty of public credit. HOWEVER: > > Please, please make this migration process repeatable. > Most importantly, document a complete list of all > services provided by each asset and the basic information > that is needed to set it up. Even better, do > as much of the work as possible by scripts that can be > saved for later. > > And make sure that everything is backed up somewhere > off of Rackspace. (In general, we need a good backup > and recovery plan for every service.) > > Because however generous Rackspace is being, in any > business things change. Sometimes suddenly. We cannot > risk our entire ecosystem on their continued generosity. > > Thanks, > Yitz -- Regards, Austin - PGP: 4096R/0x91384671 _______________________________________________ haskell-infrastructure mailing list [email protected] http://community.galois.com/mailman/listinfo/haskell-infrastructure
