frequent cache writes don't propagate to some observers

reproduction: https://codesandbox.io/s/apollo-client-frequent-cache-writes-lybjy (if empty wait 20 sec for backend to spin up)

The button calls writeFragment twice in separate tasks. Cache looks good always. The issue is some observers don’t get notified if updates come in too fast.

This happens in my app where you get a couple of WS messages with minimal interval. Some components get called twice and therefore have up-to-date cache data, while others get called only once, and their data is stale. This happens with any write method (I tried writeQuery, writeData, writeFragment). Also tried @apollo/client beta, same result.

If there’s more <Box />es, all but the last two update as expected.

[alex3d]

You are right.

[lsnch]

Thanks, the new @beta seems to be working in sandbox, but I can’t make it work in my app.

Hit a couple of roadblocks upgrading…

• Request redirection. We currently use cacheRedirects, but it seems to be missing in beta.
• Fragments. Some types of queries fail (maybe field names collide? I don’t know).

Would you be able to expand migration guide on those areas so I can test it in a real app?