The idea is to have a BE middleware that knows if the etag for a resource has changed and be able to reply "Not Modified" without actually doing the work to do the processing. Ideally we're able to entirely skip database hits at all on many requests.
Complications are: changes of resources without receiving requests for that resource itself. This would be triggered both by e.g. updates to related resources but also for resources that represent non-db or not-entirely database driven things. Likewise, background jobs would disrupt the easiest approaches.
One option would simply be to make the etag cache be TTL based on a very short cycle so that clusters of requests for the same resource could be handled together. Since this would tend to correlate with high server load, that might be all we really need. As the smarts of the cache improve, the TTLs could increase.
The idea is to have a BE middleware that knows if the etag for a resource has changed and be able to reply "Not Modified" without actually doing the work to do the processing. Ideally we're able to entirely skip database hits at all on many requests.
Complications are: changes of resources without receiving requests for that resource itself. This would be triggered both by e.g. updates to related resources but also for resources that represent non-db or not-entirely database driven things. Likewise, background jobs would disrupt the easiest approaches.
One option would simply be to make the etag cache be TTL based on a very short cycle so that clusters of requests for the same resource could be handled together. Since this would tend to correlate with high server load, that might be all we really need. As the smarts of the cache improve, the TTLs could increase.