Deployment tooling/Cabal/2015-11-23

From mediawiki.org

2015-11-23[edit]

Topics[edit]

Big Things Poppin[edit]

  • Initial Debian packaging stuff merged!
  • beta-labs RESTBase deploy on Friday
    • Bugs filed and fixed
deployment from /srv/deployment is wrong
.trebuchet directory collisions
check in ./scap directory for target list
  • TODO: file a bug: need a way to do restarts
    • Split out promote and restart
    • Just inform the user: no recourse
  • TODO: failure percentages
    • Fuzziness of problem batch_size ahead of percentage
    • etcd to determine if a server is depooled before decalring failure

TODOs from last meeting[edit]

Scap puppet work[edit]

  • Provider? Some discussion about this on Friday
    • Initial repo setup on tin via puppet
    • "Provider" looks more like directory setup then run `deploy`

git-fat[edit]

  • ????
  • The usual "deploying large binaries" problem?
    • if so, we'll need git-annex or git-lfs, git-fat is terrible.
  • Could be the context of trebuchet
  • Actual file transport the key
  • Important for jar files

etcd[edit]

  • Dynamic state: pybal
    • pybal can only depool a percentage of backends
    • not returned/obvious from bash command
    • Calculate max-batch-size inside of scap config (living in puppet)
  • First use-case, possibly, serial restart/checks of servers.
    • Have to query both pybal servers
  • Depool and repool w/single command (in a couple weeks)
  • Ideally: scap3, depool, (post-fetch)
  • Create a lock in etcd (hooray!)
    • dual-masters deploy-locks possible
  • Depooling point may have to be configurable
    • PHP and node (new children) vs. others
    • Service restart vs code changes
    • Depool, before promote, in most cases: make default
  • Ideally agnostic to etcd

As Always[edit]