Run Time Upgrade

[josietyleung]

Goal:

1. Oracle + multihop fixes

Date: 8 Aug 2023

Progress:

• [x] review required PRs @kkast
• [x] local testing @NOOB6942069
• [x] finalise CI to merge @kkast
• [x] build release candidate (including wasm file) @kkast
• [x] upgrade + test on chopsticks both composable and picasso + ibc transfers @NOOB6942069"
• [x] test on devnet or rococo @fl-y
• [x] picasso-roccoco: @rjonczy
  • [x] node upgrade
  • [x] RPC
  • [x] runtime upgrade
  • [x] put up maintanence pages on Pablo, Centauri & Picasso @fl-y
  • [x] discord + telegram announcement @NOOB6942069
• [x] composable: @rjonczy
  • [x] all nodes upgrade
  • [x] all RPCs
  • [x] runtime upgrade
• [x] picasso-kusama: @rjonczy
  • [x] all nodes upgrade
  • [x] all RPCs
  • [x] runtime upgrade
• [ ] confirm stability @rjonczy
• [x] smoke test production @NOOB6942069
• [ ] remove maintenance pages on Pablo & Picasso @fl-y
• [ ] runtime upgrade hyperspace - all relayers @vmarkushin @rjonczy
• [ ] remove maintenance page on Centauri @fl-y
• [x] communicate to collators to upgrade to latest version @NOOB6942069
• [x] discord + telegram announcement @NOOB6942069

[kkast]

sre team needs to update all nodes +rpcs for composable + picasso as part of RTU

[kkast]

Need to change "upgrade + test on chopsticks @NOOB6942069" to "upgrade + test on chopsticks + ibc transfers @NOOB6942069" + tests from fe team on devnet or rococo @fl-y

[kkast]

then we upgrade all relayers all the time

[blasrodri]

Robert will move our infra to Kubernetes. This will be a long work, but it will payoff. It will provide us with a simple mechanism to update our infra much faster, and to react upon things like a new docker image being added into the registry. It will certainly speed up deployment of RPCs + Hyperspace.

[kkast]

for fe and parachain common testing we 1) start which.dev with previous release version 2) populate it with data 3) check fe works(it should) 4) do runtime upgrade with release wasm file 5) run automated tests @fl-y @blasrodri

[dzmitry-lahoda]

k8s sucks for single nodes and p2p software and software which special purposes hardware(more dances around than doing usefull work). terraform, nix, amis, cloud inits, cloud formation are better. k8s by no means simple. k8s also sucks with state with identity. k8s edge cases are our cases. feel free to waste half year and suffer migration head ackes :) btw k8s can be done as bad as any cloud, it is not magic bullet.

[josietyleung]

closing as completed