[Merged by Bors] - Fix possible deadloop in beacon

[fasmat]

Motivation

I found a problem in the beacon protocol where if the node wakes up from sleep or syncs its clock via NTP at the wrong moment the go routine running the beacon protocol might end up in a deadloop without ever recovering.

Description

listenEpochs is supposed to wait until the start of an epoch to then start the beacon protocol for that epoch. When the new epoch starts the select case <-pd.clock.AwaitLayer(layer) will unlock. This can happen any time after that layer was reached - usually ms, but if the runtime was very busy or the host was hibernating the time between reaching the layer and this case unlocking can extend much longer. The problem is the if a bit further below:

if !current.FirstInEpoch() {
	continue
}

if for any reason the signal from the select case was received after the 2nd layer of the epoch already started, this if statement will continue the for loop without updating current or layer resulting in a deadloop.

I fixed the issue by updating layer before checking, so that if the node is late it skips participating in the beacon protocol for this epoch and waits for the next.

Test Plan

Existing tests pass

TODO

• Explain motivation or link existing issue(s)
• Test changes and document test plan
• Update documentation as needed
• Update changelog as needed

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.9%. Comparing base (fbca87d) to head (7f8251a).
Report is 1 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #6451   +/-   ##
=======================================
  Coverage     79.9%   79.9%           
=======================================
  Files          352     352           
  Lines        46099   46098    -1     
=======================================
+ Hits         36850   36867   +17     
+ Misses        7154    7143   -11     
+ Partials      2095    2088    -7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[poszu]

What do you mean by "deadloop"? As far as I understand, it would spin on:

		case <-pd.clock.AwaitLayer(layer):
			current := pd.clock.CurrentLayer()
			if !current.FirstInEpoch() {
				continue
			}

for a long time until the next epoch (when the current becomes the first layer finally) because layer is in the past and not updated.

[fasmat]

What do you mean by "deadloop"? As far as I understand, it would spin on:

		case <-pd.clock.AwaitLayer(layer):
			current := pd.clock.CurrentLayer()
			if !current.FirstInEpoch() {
				continue
			}

for a long time until the next epoch (when the current becomes the first layer finally) because layer is in the past and not updated.

Right, it will only loop for a full epoch.

[fasmat]

bors merge

[spacemesh-bors]

Pull request successfully merged into develop.

Build succeeded:

• ci-status
• systest-status