We have done the work required to get Thumbor running on kubernetes and it is now serving some production traffic. However, there are a few steps required before we can finalise the project:
- Migrate Thumbor's memcached backend away from Thumbor bare metal servers T318695
- Scale up capacity in eqiad
Currently we have scaled up enough to handle 100% of traffic in codfw on k8s. In the short term we should move towards handling 100% of traffic in codfw, but we are limited in this regard in eqiad. We can handle a 60/40 split k8s/metal there as of last testing. We should clarify and validate those numbers and then scale accordingly if capacity is available. If not, we should clearly escalate and highlight this fact to managers etc.
- Decommission the existing thumbor hosts
Once we've got the capacity, we should decommission the old hosts as soon as is possible. Four of the eight servers are out of warranty.
- clean up puppet classes/defines
- Remove apt component
- General performance improvements T333445
Find if there are areas we need to improve upon either on a per-format or general basis. Ideally bound this work within specific tasks so as to avoid taking too long or the ticket sprawling generally.