Akka.NET, one server cut off from its group

What I ran

Three Akka.NET nodes ran in three worlds on one virtual network. Seed 7 cut node 2 off from the others, starting at 30,440 ms of virtual time and lasting 100,000 ms. That is longer than the 90,000 ms the worlds were asked to last. The switch dropped 42 frames, on 4 links.

What it shows

MIGHT. In 2 of 2 runs of the three-node cluster partition on seed 7, Akka's default split-brain resolver removed the node the seed cut off, and no other node.

In the main run's report, the cut-off node, alone on its side, applied the same rule to what it could see and removed every member, itself included.

At the end it counted 0 nodes Up, and each of the others counted 2.

The cut-off node's removal of every member is in the main run's report alone.

What it does not show

That a second run of one seed gives the same messages. The gate says the same seed gives the same window and does not say it gives the same run.

Which frame in a stretch was cut. The switch writes no line for a dropped frame, so a frame counted as dropped is one that was numbered and not handed over.

That the cluster heals when the cut ends. The window of 100,000 ms outlasted the 90,000 ms the worlds were asked to last, so the node was still cut off when the worlds stopped.

That the worlds ran on cores reserved for them. The ledger row of this gate says isolation was not applicable, and the machine's load average when the run began, a figure the runner recorded, was 3.17, 2.91 and 4.39 over one, five and fifteen minutes.

The gate's source names no lease on the world farm, and its ledger row says isolation was not applicable.

Conditions

Three worlds, one Akka.NET node in each, each node run for 70 s, on one virtual network. Run on hypervisor build A9A8850A71C5E48F25751126C95F55C2F67E5718B09263407463F09012E0E528.