After we rolled out tegola to serve all production traffic and after a couple of hiccups with overzoom not working, tile serving and pregeneration looks like is very stable.
The only problem we have identified so far is related to cache latency.

• When tegola is serving both the regular maps traffic and tile pregeneration is happening cache latency is high.
  • https://grafana.wikimedia.org/goto/qRciyEonk

To be investigated:

• What's the minimum number of replicas we can have for tegola-tile-pregeneration in order to have reasonable performance ?
• Do we have any quotas on resource usage from swift ?
• Can we prioritize live traffic serving over tile pregeneration when it comes to swift requests ?