timing games

[ltitanb]

fix #23

This PR adds support for timing games, ie. allows a node operator to fine tune how to send get_header requests to relays. This is configured via two new per-relay parameters:

• target_first_request_ms: which sets how late in the slot to send the first header request
• frequency_get_header_ms: which sets how frequently to send header requests after the first

and a global param:

• late_in_slot_time_ms: which sets how late in the slot is "late", acting as a hard upper bound for when to return headers to the CL

The requests are still bound by timeout_get_header_ms which is ~what the full get_header request can take at most from the CL perspective (plus some small overhead from the module itself).

The logic is as follows:

• if the request from the CL comes before target_first_request_ms, wait until the target time
• after that send requests every frequency_get_header_ms with decreasing timeouts
• if the request comes later into the slot, then reduce the max timeout accordingly

Examples
Assuming: timeout_get_header_ms = 950, frequency_get_header_ms = 300, target_first_request_ms = 200, late_in_slot_time_ms = 2000

1. CL request comes at 100ms in the slot (max timeout 1050ms in the slot), then:

• sleep for 100ms
• send request at 200ms with 850ms timeout
• send request at 500ms with 550ms timeout
• send request at 800ms with 250ms timeout

2. CL request comes at 1500ms in the slot (max timeout 2000ms in the slot), then:

• send request at 1500ms with 500ms timeout
• send request at 1800ms with 200ms timeout

3. CL request comes 2500ms in the slot then:

• return 204 and force local build

[cwhcheng]

Hi! If a relay intentionally delays the return of the get_header response to the max timeout value, does it make any difference whether the validators try to send multiple get_header request within the timeout?

[ltitanb]

hey @cwhcheng would depend on a number of factors such as how long the relay is waiting and how they process this internally, best way is probably to make real world tests