Availability Distribution: Consider occupied cores from last k blocks within same session

[rphmeier]

A Before and after of availability-recovery metrics on Kusama would be really interesting (number of NoSuchChunk):
https://grafana.parity-mgmt.parity.io/goto/sjkh8Upnz?orgId=1 .. currently I see 3-5 NoSuchChunk per second.

Discussion here: #2423 (comment)

This was the previous behavior before using request/response and should be reinstated. A permalink to the previous implementation can be found here:

polkadot/node/network/availability-distribution/src/lib.rs

Line 952 in c66049d

async fn query_live_candidates<Context>(

.

It is not crucial to the functioning of the network but is useful for maximizing availability and increasing throughput by increasing likelihood of peers having chunks later on.

It is important to only consider blocks within the same session, and this should recover gracefully from runtime API errors as historical state may be (but is unlikely to be) pruned.

[eskimor]

Up to 7% of requests fail due to "NoSuchChunk" right now on Kusama (normally in the 2% range).

[eskimor]

Situation with fixes, still in the same range of failures.

[slumber]

Situation with fixes, still in the same range of failures.

It's very likely that the source of this issue is #4792 and it's fixed in #4850 (not merged yet)