Description:

Describe the issue.

We are leveraging the Updates field in the Snapshot

This allows us to perform smaller unit of work ( per key instead of map) ops like deleting an IR in the xds subscriber but this means that for a cases where the message is in state of the world form, like ControllerResources , we get multiple updates for the same key (e.g. deployment rollout of 100 pods, causing 100 endpoints to be gradually updated) which causes multiple iterations of translation, each iteration delaying convergence .

We can do better here by merging the updates based on key and operation (mainly to avoid skipping a delete) reducing duplicate operations .

[optional Relevant Links:]

Any extra documentation required to understand the issue.