Skip to content

Instantly share code, notes, and snippets.

@deinarson

deinarson/notes.md

Last active October 30, 2015 20:48
Show Gist options
  • Save deinarson/54a5485daaa720bc3cf1 to your computer and use it in GitHub Desktop.
Save deinarson/54a5485daaa720bc3cf1 to your computer and use it in GitHub Desktop.
Download ZIP
Raw
notes.md

takeover states

  1. Rolling take-over; sawp out one node at a time
  • pros seamless migration of node replacement
  • cons no rollback
  • depending on LB config can produce end-user hiccups (RR vs load)
  1. Full takeover - aka LB swap-out ; Using DNS to swap HA/LB
  • pro quickly sawp from cluster a to cluster b - ie usage schema change or braking change
  • con DNS ttl lag time to changeover
  1. Populate to replace ; add nodes until the LB starts using them then kill old nodes.
  • pros you can watch the metrics and determines there is a success before polling out the working nodes ( like a dark-lanch in the early stages )
  • cons can cost more
  • depending on LB config can produce end-user hiccups (RR vs load)
  1. ??? I forgot

dark launch aka feature flag

Here we evaluate a conditional to determine who are canaries is and give 'unproven' code to those subset of clients. This allows us to push directly to prod with out the worry of killing everything for everyone. we merely give some clients the new feature or bit of code that we are testing.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment