This is the order of events that we discussed leading to release 20.11. Links to test plans will be added here.
- Testing on Badger October branch. Dgraph/Badger Changes for 20.11
- Release Badger 20.10
- Dgraph 20.11 feature, regression, performance and upgrade testing
- Cut rc1
- Slash testing on staging - 1 week
- Phased rollout to slash production
6.1. New Clusters are in 20.11, old are in 20.07 - 1 week
6.2. Free Customers are auto upgraded - 1 week (given 1 week notice when we do 6.1)
6.3. Paying customers are upgraded including Newsela who is enterprise at that point (again given 1-2 weeks notice) - Release 20.11 externally
Things that Slash would like to see tested before going live:
- Test with the top schemas. Slash will provide a set of schemas to validate that GraphQL works correctly. We should design some simple automated tests with these.
- Safe to deploy via kube. We need it to be possible to just update the image and for kube to roll out dgraph. This means that technically, you could be running some combinations of 20.07 and 20.11 on zeros and alphas for upto 5 minute during the golive process, and data should not get corrupted.
- Ease of Rollback. In case the 20.011 does not work, i’ll need a script to convert the data back into a format that 20.07 can read, and restart the cluster. I think it’s probably acceptable to have ~1minute of downtime per say 5gb of data here.
@gja, @Paras, @vvbalaji and @Sankalan13 please add anything i missed.