Sync Committee attendance performance issue

[twoeths]

Describe the bug

In prater, when any validators join sync committee period, the performance is very bad. This maybe related to #3416 where we subscribe to all sync committee subnets with subscribeAllSubnets flag

On Nov 17, some validators joined the sync committee

More FINDNODES dscv5 query were issues, more dial rate

At the same time, there are a lot of Goodbye sent/received, a lot of ping/metadata/status request/response too

Gossip block is delayed a lot

Expected behavior

No performance issue

[twoeths]

This could also be linked to #2490

When we dial a lot of peers, it's more likely we have to serve beacon_blocks_by_range

[dapplion]

This could also be linked to #2490

When we dial a lot of peers, it's more likely we have to serve beacon_blocks_by_range

Could be but I'll be surprised it had such an impact on performance, would be worth it to stay vigilant of our nodes and whenever we get selected again to participate in a sync committee take a CPU profile

[twoeths]

vigilant

we can just do a hard code to always subscribe all sync committees and deploy to one of our servers

[twoeths]

issue is too old, lodestar can publish sync committee messages well with v1.2.1