roachtest: skip perturbation-interval gate for partition

The partition test isolates an entire region (4 of 12 nodes), removing
1/3 of leaseholders. Foreground throughput drops sharply while the
partition is in effect, and the meaningful pass/fail signal is whether
the cluster returns to baseline once the partition heals. Gating the
perturbation interval at the default 1.25x throughput floor mostly
produces flakes.

Use noImpactThresholds() for the perturbation interval and keep the
default 1.25x throughput floor for the recovery interval, via the
recoveryImpact field added in the parent commit.

While here, add a comment to slowDisk explaining why the default
threshold is appropriate for the full variant: with walFailover=true
and 2 disks per node, raft log writes fail over to the non-throttled
store and foreground throughput stays close to baseline. The lenient
1.25x floor is mainly to absorb noise from the slowLiveness leg.

Release note: None

Author: tbg

pkg/cmd/roachtest/tests/perturbation/network_partition.go

@@ -31,7 +31,13 @@ var _ perturbation = partition{}
 
 func (p partition) setup() variations {
 	p.partitionSite = true
-	v := setup(p, defaultThresholds())
+	// The partition test isolates an entire region (4 of 12 nodes), removing
+	// 1/3 of leaseholders. Foreground throughput naturally drops sharply
+	// while the partition is in effect, and the meaningful pass/fail signal
+	// is whether the cluster returns to baseline once the partition heals.
+	// Skip the perturbation-interval gate; keep the default recovery gate.
+	v := setup(p, noImpactThresholds())
+	v.recoveryImpact = defaultThresholds()
 	v.leaseType = registry.ExpirationLeases
 	// TODO(baptist): Remove this setting once #120073 is fixed.
 	v.clusterSettings["kv.lease.reject_on_leader_unknown.enabled"] = "true"

pkg/cmd/roachtest/tests/perturbation/slow_disk.go

@@ -32,6 +32,15 @@ var _ perturbation = &slowDisk{}
 func (s *slowDisk) setup() variations {
 	s.slowLiveness = true
 	s.walFailover = true
+	// With walFailover=true and 2 disks per node (the default for the full
+	// variant), raft log writes fail over to the non-throttled store, so
+	// foreground throughput is expected to stay close to baseline even
+	// while the staller is active. Default thresholds apply to both
+	// intervals; we keep the 1.25x floor (rather than tightening) only to
+	// avoid flakes from the slowLiveness leg, which routes liveness
+	// heartbeats through the slow disk. The metamorphic variant exercises
+	// configurations where walFailover is off and throughput can drop
+	// substantially -- those should override impact independently.
 	return setup(s, defaultThresholds())
 }