swactor/crates/simulation/scenarios/topology/partition_heal.toml

# Partition the cluster into {a} | {b, c}, then heal. Asserts the cluster
# converges back to all-Alive within a bounded window AND that the isolated
# node's death is *resolved* — the Dead→Alive resurrection arc (Goal 3, "death
# is provisional"), not just the post-heal endpoint. `dead_reprobe_interval_ns`
# is non-zero so the partition-heal detector (SWIM_ACTOR_SPEC §9.8) actually
# re-probes the isolated node and lets it refute back to Alive.
# Expected verdict: pass-now.

name = "partition_heal"
seed = 7
duration_ns = 90_000_000_000

[default_tick]
period_ns = 200_000_000

[default_link]
latency_ns = 5_000_000
jitter_stddev_ns = 1_000_000
loss_prob_ppm = 0
reorder_prob_ppm = 0
bandwidth_bps = 100_000_000
cold_dial_penalty_ns = 50_000_000
cache_warm_after_ns = 100_000_000
cache_invalidate_after_idle_ns = 30_000_000_000

[[peers]]
id = "a"
kind = "swim"
initial_state = "alive"
kind_config = { probe_interval_ns = 1_000_000_000, suspicion_timeout_ns = 5_000_000_000, dead_reprobe_interval_ns = 1_000_000_000 }
[[peers]]
id = "b"
kind = "swim"
initial_state = "alive"
kind_config = { probe_interval_ns = 1_000_000_000, suspicion_timeout_ns = 5_000_000_000, dead_reprobe_interval_ns = 1_000_000_000 }
[[peers]]
id = "c"
kind = "swim"
initial_state = "alive"
kind_config = { probe_interval_ns = 1_000_000_000, suspicion_timeout_ns = 5_000_000_000, dead_reprobe_interval_ns = 1_000_000_000 }

[[links]]
from = "a"
to = "b"
[[links]]
from = "b"
to = "a"
[[links]]
from = "a"
to = "c"
[[links]]
from = "c"
to = "a"
[[links]]
from = "b"
to = "c"
[[links]]
from = "c"
to = "b"

[[mutations]]
kind = "partition"
at_ns = 10_000_000_000
peers_a = ["a"]
peers_b = ["b", "c"]

[[mutations]]
kind = "heal"
at_ns = 30_000_000_000

# Snapshots spanning the post-heal convergence window so `convergence_after`
# has data to evaluate (snapshots are only emitted at these declared times).
[[snapshots]]
at_ns = 45_000_000_000
[[snapshots]]
at_ns = 55_000_000_000
[[snapshots]]
at_ns = 60_000_000_000

[[assertions]]
kind = "convergence_after"
after_ns = 30_000_000_000
within_ns = 30_000_000_000
peers = ["a", "b", "c"]

# Goal 3 — death is provisional: the isolated node "a" is detected Dead during
# the partition, then resurrected to Alive after the heal (within the window).
[[assertions]]
kind = "dead_peer_resurrects_within"
peer = "a"
after_ns = 10_000_000_000
within_ns = 60_000_000_000