317 lines
13 KiB
Rust
317 lines
13 KiB
Rust
|
|
//! Registry simulation tests — cluster registry CRDT behavior under gossip.
|
||
|
|
//!
|
||
|
|
//! Tests that registry names propagate, converge, and resolve correctly
|
||
|
|
//! across the cluster under various fault conditions.
|
||
|
|
|
||
|
|
use simulation::distribution::properties::{
|
||
|
|
check_registry_propagation, check_registry_tombstones,
|
||
|
|
};
|
||
|
|
use simulation::distribution::sim::{
|
||
|
|
run_simulation, run_simulation_with_nodes, DistributionSimConfig, NetworkFault, Partition,
|
||
|
|
SimAction,
|
||
|
|
};
|
||
|
|
|
||
|
|
fn default_config() -> DistributionSimConfig {
|
||
|
|
DistributionSimConfig {
|
||
|
|
actors_per_node: 0, // Registry tests don't need actors
|
||
|
|
..DistributionSimConfig::default()
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 1. Registry name converges across 5-node cluster via gossip piggyback
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn registry_name_converges_across_cluster() {
|
||
|
|
// Given: 5-node cluster, node 0 registers "counter" at round 5
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "registry-convergence".into(),
|
||
|
|
num_nodes: 5,
|
||
|
|
num_rounds: 50,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
action_schedule: vec![
|
||
|
|
(5, SimAction::RegisterName { node_idx: 0, name: "counter".into() }),
|
||
|
|
],
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: all alive nodes should have the registry entry
|
||
|
|
let result = check_registry_propagation(&trace, 1);
|
||
|
|
assert!(
|
||
|
|
result.passed,
|
||
|
|
"all nodes should see the 'counter' name: {}",
|
||
|
|
result.actual
|
||
|
|
);
|
||
|
|
|
||
|
|
// And: all nodes should resolve "counter" to the same actor
|
||
|
|
let resolutions: Vec<_> = nodes
|
||
|
|
.iter()
|
||
|
|
.filter_map(|n| n.as_ref())
|
||
|
|
.map(|n| n.resolve_name("counter"))
|
||
|
|
.collect();
|
||
|
|
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.is_some()),
|
||
|
|
"all nodes should resolve 'counter', got: {resolutions:?}"
|
||
|
|
);
|
||
|
|
|
||
|
|
// All should agree on the same actor address
|
||
|
|
let first = resolutions[0].unwrap().0;
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.unwrap().0 == first),
|
||
|
|
"all nodes should agree on the same actor for 'counter'"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 2. Split-brain naming — two sides register same name during partition
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn split_brain_naming_converges_after_partition_heals() {
|
||
|
|
// Given: 6-node cluster, partition {0,1,2} vs {3,4,5} at round 10
|
||
|
|
// Node 0 registers "leader" at round 12, node 3 registers "leader" at round 12
|
||
|
|
// Heal at round 40
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "split-brain-registry".into(),
|
||
|
|
num_nodes: 6,
|
||
|
|
num_rounds: 100,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
network_faults: vec![
|
||
|
|
NetworkFault::Partition {
|
||
|
|
round: 10,
|
||
|
|
partition: Partition {
|
||
|
|
side_a: vec![0, 1, 2],
|
||
|
|
side_b: vec![3, 4, 5],
|
||
|
|
asymmetric: false,
|
||
|
|
},
|
||
|
|
},
|
||
|
|
NetworkFault::Heal { round: 40 },
|
||
|
|
],
|
||
|
|
action_schedule: vec![
|
||
|
|
(12, SimAction::RegisterName { node_idx: 0, name: "leader".into() }),
|
||
|
|
(12, SimAction::RegisterName { node_idx: 3, name: "leader".into() }),
|
||
|
|
],
|
||
|
|
swim: distribution::swim::probe::SwimConfig {
|
||
|
|
probe_interval: 1,
|
||
|
|
probe_timeout: 3,
|
||
|
|
indirect_probes: 1,
|
||
|
|
suspicion_timeout: 5,
|
||
|
|
dead_reprobe_interval: 10,
|
||
|
|
},
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (_trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: all alive nodes should resolve "leader" to the same value (LWW winner)
|
||
|
|
let resolutions: Vec<_> = nodes
|
||
|
|
.iter()
|
||
|
|
.filter_map(|n| n.as_ref())
|
||
|
|
.map(|n| n.resolve_name("leader"))
|
||
|
|
.collect();
|
||
|
|
|
||
|
|
// All should resolve to Some (the LWW winner)
|
||
|
|
let resolved: Vec<_> = resolutions.iter().filter_map(|r| r.as_ref()).collect();
|
||
|
|
assert!(
|
||
|
|
resolved.len() >= 4,
|
||
|
|
"at least 4 of 6 nodes should resolve 'leader', got {} out of {}",
|
||
|
|
resolved.len(),
|
||
|
|
resolutions.len()
|
||
|
|
);
|
||
|
|
|
||
|
|
// All resolving nodes should agree on the same actor
|
||
|
|
if resolved.len() >= 2 {
|
||
|
|
let first_actor = resolved[0].0;
|
||
|
|
let agree = resolved.iter().all(|r| r.0 == first_actor);
|
||
|
|
assert!(
|
||
|
|
agree,
|
||
|
|
"all nodes resolving 'leader' should agree on the same actor (LWW winner)"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 3. Tombstone propagation on node death
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn tombstone_propagates_when_name_owner_dies() {
|
||
|
|
// Given: 5-node cluster, node 0 registers "svc" at round 5, killed at round 15
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "tombstone-propagation".into(),
|
||
|
|
num_nodes: 5,
|
||
|
|
num_rounds: 80,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
action_schedule: vec![
|
||
|
|
(5, SimAction::RegisterName { node_idx: 0, name: "svc".into() }),
|
||
|
|
],
|
||
|
|
kill_schedule: vec![(15, 0)],
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: surviving nodes should have tombstoned "svc"
|
||
|
|
let result = check_registry_tombstones(&trace, 1);
|
||
|
|
assert!(
|
||
|
|
result.passed,
|
||
|
|
"survivors should have tombstones after node death: {}",
|
||
|
|
result.actual
|
||
|
|
);
|
||
|
|
|
||
|
|
// And: resolve_name("svc") should return None on all survivors
|
||
|
|
let resolutions: Vec<_> = nodes
|
||
|
|
.iter()
|
||
|
|
.filter_map(|n| n.as_ref())
|
||
|
|
.map(|n| n.resolve_name("svc"))
|
||
|
|
.collect();
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.is_none()),
|
||
|
|
"all survivors should resolve 'svc' to None after owner dies, got: {resolutions:?}"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 4. Rapid re-registration converges to latest value
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn rapid_re_registration_converges_to_latest() {
|
||
|
|
// Given: 5-node cluster, node 0 registers "svc" three times rapidly
|
||
|
|
use swactor::actor::ActorAddress;
|
||
|
|
let actor_a = ActorAddress::new_random();
|
||
|
|
let actor_b = ActorAddress::new_random();
|
||
|
|
let actor_c = ActorAddress::new_random();
|
||
|
|
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "rapid-reregister".into(),
|
||
|
|
num_nodes: 5,
|
||
|
|
num_rounds: 60,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
action_schedule: vec![
|
||
|
|
(5, SimAction::RegisterNameWithActor { node_idx: 0, name: "svc".into(), actor: actor_a }),
|
||
|
|
(6, SimAction::RegisterNameWithActor { node_idx: 0, name: "svc".into(), actor: actor_b }),
|
||
|
|
(7, SimAction::RegisterNameWithActor { node_idx: 0, name: "svc".into(), actor: actor_c }),
|
||
|
|
],
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (_trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: all nodes should resolve "svc" to actor_c (the latest registration)
|
||
|
|
let resolutions: Vec<_> = nodes
|
||
|
|
.iter()
|
||
|
|
.filter_map(|n| n.as_ref())
|
||
|
|
.map(|n| n.resolve_name("svc"))
|
||
|
|
.collect();
|
||
|
|
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.is_some()),
|
||
|
|
"all nodes should resolve 'svc'"
|
||
|
|
);
|
||
|
|
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.unwrap().0 == actor_c),
|
||
|
|
"all nodes should converge to the latest registration (actor_c)"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 5. Simultaneous registration of same name on different nodes
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn simultaneous_registration_converges_deterministically() {
|
||
|
|
// Given: 5-node cluster, node 0 and node 3 both register "mutex" at round 5
|
||
|
|
use swactor::actor::ActorAddress;
|
||
|
|
let actor_a = ActorAddress::new_random();
|
||
|
|
let actor_b = ActorAddress::new_random();
|
||
|
|
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "simultaneous-register".into(),
|
||
|
|
num_nodes: 5,
|
||
|
|
num_rounds: 60,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
action_schedule: vec![
|
||
|
|
(5, SimAction::RegisterNameWithActor { node_idx: 0, name: "mutex".into(), actor: actor_a }),
|
||
|
|
(5, SimAction::RegisterNameWithActor { node_idx: 3, name: "mutex".into(), actor: actor_b }),
|
||
|
|
],
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (_trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: all nodes should agree on one winner
|
||
|
|
let resolutions: Vec<_> = nodes
|
||
|
|
.iter()
|
||
|
|
.filter_map(|n| n.as_ref())
|
||
|
|
.map(|n| n.resolve_name("mutex"))
|
||
|
|
.collect();
|
||
|
|
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.is_some()),
|
||
|
|
"all nodes should resolve 'mutex'"
|
||
|
|
);
|
||
|
|
|
||
|
|
// All should agree on the same actor (whichever won the LWW tiebreaker)
|
||
|
|
let first_actor = resolutions[0].unwrap().0;
|
||
|
|
assert!(
|
||
|
|
resolutions.iter().all(|r| r.unwrap().0 == first_actor),
|
||
|
|
"all nodes should agree on the LWW winner for 'mutex'"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
// 6. Multiple names propagate correctly
|
||
|
|
// ────────────────────────────────────────────────────────────────────────────
|
||
|
|
|
||
|
|
#[test]
|
||
|
|
fn multiple_names_from_different_nodes_all_propagate() {
|
||
|
|
// Given: 5-node cluster, each node registers a unique name
|
||
|
|
let config = DistributionSimConfig {
|
||
|
|
name: "multi-name-propagation".into(),
|
||
|
|
num_nodes: 5,
|
||
|
|
num_rounds: 60,
|
||
|
|
ticks_per_round: 3,
|
||
|
|
action_schedule: vec![
|
||
|
|
(5, SimAction::RegisterName { node_idx: 0, name: "svc-0".into() }),
|
||
|
|
(5, SimAction::RegisterName { node_idx: 1, name: "svc-1".into() }),
|
||
|
|
(5, SimAction::RegisterName { node_idx: 2, name: "svc-2".into() }),
|
||
|
|
(5, SimAction::RegisterName { node_idx: 3, name: "svc-3".into() }),
|
||
|
|
(5, SimAction::RegisterName { node_idx: 4, name: "svc-4".into() }),
|
||
|
|
],
|
||
|
|
..default_config()
|
||
|
|
};
|
||
|
|
|
||
|
|
// When: we run the simulation
|
||
|
|
let (trace, nodes) = run_simulation_with_nodes(config);
|
||
|
|
|
||
|
|
// Then: all 5 nodes should have all 5 registry entries
|
||
|
|
let result = check_registry_propagation(&trace, 5);
|
||
|
|
assert!(
|
||
|
|
result.passed,
|
||
|
|
"all nodes should have all 5 registry entries: {}",
|
||
|
|
result.actual
|
||
|
|
);
|
||
|
|
|
||
|
|
// And: each node should resolve all 5 names
|
||
|
|
for node in nodes.iter().filter_map(|n| n.as_ref()) {
|
||
|
|
for i in 0..5 {
|
||
|
|
let name = format!("svc-{i}");
|
||
|
|
assert!(
|
||
|
|
node.resolve_name(&name).is_some(),
|
||
|
|
"every node should resolve '{name}'"
|
||
|
|
);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|