and developer guides
Attendees: Reynir Björnsson, Amir Chaudhry (chair), Justin Cormack, Thomas Gazagnaire, Masoud Koleini, Thomas Leonard, Hannes Mehnert, Richard Mortier, Mindy Preston and Jeremy Yallop
Been provisioning a number of backends on one of our machines (punk), which is serving https://mirage.io, including tuntap and unix. Would also like to have an ARM backend running but that's blocked on Scaleway as they haven't finished looking into the Xen support (scaleway/kernel-tools#14). We will also need a logging interface for unikernels but that can wait until the other work is done.
Ultimately, we'd like to use Mirage.io as the main site and load-balance the incoming connections across all our supported backends (x86/ARM, Unix/Xen). Would also like this to work across multiple providers, including AWS. We'd need a method to inform the visitor which version of the site they're looking at in order to triage and fix incoming bug reports.
This is working but we need to fix a memory leak in Entropy and maybe also
xentropyd. Anil mentioned that he'd write up instructions on how
to go from getting certificates, all the way to a deployed https website —
likely based on his experience with certs from Gandi.net
Amir will coordinate a set of blog posts for the TLS related releases and we should publish things when we're all back. For the moment we should focus on testing things with the new stack.
ThomasL mentioned that he announced Cuekeeper and published a blog post about it and today he also got it working with Safari! It turns out that Safari's database implementation can't cope with concurrent reads, so when it throws an exception, Cuekeeper has to back off and try again.
Anil will be speaking at an OCaml user group meeting (Silicon Valley OCaml Meetup) and will be showing the crowd an Irmin demo and Cuekeeper.
ThomasG is putting together docs on how to run some of our existing demos and it should be almost a step by step, repeatable guide. Would appreciate feedback on this.
It turns out that lots of people have been trying out MirageOS and have found their way to Amir's Jekyll to Unikernel post and Mindy's post on deploying unikernels to EC2. Getting deployments up to EC2 is still quite tricky and a nicer way to do this would be very welcome. Lots of people seem to have scripts for this including Anil, Mindy and Jyostna, and it would be good to publicise or consolidate these. We'd also have to ensure they work when we have SSL support.
This was discussed in a previous call and was meant to be brought up this time. Given how busy we're likely to be this summer, we'll probably put this off until later in the year.
Jitsu/NSDI - We should consider how we can improve our release management process. At the moment we have no good way to do this as it requires the ability to manage issues across repos. Most of our releases involve cross-cutting changes and we have many libraries. We could possibly put something together ourselves using the GitHub API but this would be a project in its own right. Perhaps David Sheets can look at this once he's back.
UROP/Pioneer Projects. We have two summer students taking part, with one project on a DHCP server and another on an NTP server. If anyone has written a DHCP server before, please let us know.