LP needs instances deployed
Bug #809138 reported by
Robert Collins
This bug affects 1 person
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| txlongpoll |
Triaged
|
High
|
Unassigned | ||
Bug Description
For longpoll deployments, we can't deploy via the LP tree (because
that conflicts with the import rule); basically we want to deploy
*like* we do lp: have a staging area, run buildout there, with our
common download-cache, then rsync the resulting thing to a dedicated
tree on the servers running longpoll.
For HA we need 4 longpoll instances (2 per datacentre) and probably
want it sitting behind haproxy because apache is terrible at failover.
| Changed in lazr.amqp: | |
| status: | New → Triaged |
| importance: | Undecided → High |
| description: | updated |
Revision history for this message
| William Grant (wgrant) wrote | #1 |
To post a comment you must log in.
Since everything runs out of the egg, it's probably best to have a download-cache (potentially shared with LP) and deploy using a short buildout.cfg to grab the eggs and build twistd. Saves syncing source trees everywhere.
Something like this seems to work:
[buildout]
parts = twistd
use-dependency-
unzip = true
include-
exec-sitecustomize = true
[twistd]
recipe = zc.recipe.
interpreter = py
eggs = txlongpoll
entry-points = twistd=