## Issue Addressed
Addresses https://github.com/sigp/lighthouse/issues/2926
## Proposed Changes
Appropriated from https://github.com/sigp/lighthouse/issues/2926#issuecomment-1039676768:
When a node returns *any* error we call [`CandidateBeaconNode::set_offline`](c3a793fd73/validator_client/src/beacon_node_fallback.rs (L424)) which sets it's `status` to `CandidateError::Offline`. That node will then be ignored until the routine [`fallback_updater_service`](c3a793fd73/validator_client/src/beacon_node_fallback.rs (L44)) manages to reconnect to it.
However, I believe there was an issue in the [`CanidateBeaconNode::refesh_status`](c3a793fd73/validator_client/src/beacon_node_fallback.rs (L157-L178)) method, which is used by the updater service to see if the node has come good again. It was holding a [write lock on the `status` field](c3a793fd73/validator_client/src/beacon_node_fallback.rs (L165)) whilst it polled the node status. This means a long timeout would hog the write lock and starve other processes.
When a VC is trying to access a beacon node for whatever purpose (getting duties, posting blocks, etc), it performs [three passes](c3a793fd73/validator_client/src/beacon_node_fallback.rs (L432-L482)) through the lists of nodes, trying to run some generic `function` (closure, lambda, etc) on each node:
- 1st pass: only try running `function` on all nodes which are both synced and online.
- 2nd pass: try running `function` on all nodes that are online, but not necessarily synced.
- 3rd pass: for each offline node, try refreshing its status and then running `function` on it.
So, it turns out that if the `CanidateBeaconNode::refesh_status` function from the routine update service is hogging the write-lock, the 1st pass gets blocked whilst trying to read the status of the first node. So, nodes that should be left until the 3rd pass are blocking the process of the 1st and 2nd passes, hence the behaviour described in #2926.
## Additional Info
NA
5a0b049049