UNDF: UNDF-2026-000000570

vertx-0001 — HAManager nodeLeft() nodes List.contains O(C²) in cluster failover scan

Classification: CWE-407 Algorithmic Complexity — Inefficient Membership Test Severity: MEDIUM Component: vert.x core — io.vertx.core.impl.HAManager File: vertx-core/src/main/java/io/vertx/core/impl/HAManager.java Line: 307–317

Description

When a cluster node leaves, HAManager.nodeLeft() scans the entire clusterMap to find any in-progress failovers that were themselves abandoned. For each entry in clusterMap it calls nodes.contains(entry.getKey()) where nodes is a List<String> returned by clusterManager.getNodes(). The List.contains() is an O(N) linear scan, making the entire loop O(C × N) = O(N²) in cluster size.

Defective code

// HAManager.java:307-317
List<String> nodes = clusterManager.getNodes();

for (Map.Entry<String, String> entry: clusterMap.entrySet()) {
    if (!leftNodeID.equals(entry.getKey()) && !nodes.contains(entry.getKey())) {
        JsonObject haInfo = new JsonObject(entry.getValue());
        checkFailover(entry.getKey(), haInfo);
    }
}

nodes is declared as List<String> getNodes() in ClusteredNode.java:47. The clusterMap has one entry per cluster node. Both collections grow linearly with cluster size, so the inner nodes.contains() is called C times, each taking O(N) time → O(N²) total.

Fix

Convert nodes to a HashSet<String> before the loop:

List<String> nodesList = clusterManager.getNodes();
Set<String> nodesSet = new HashSet<>(nodesList);   // O(N) one-time build

for (Map.Entry<String, String> entry: clusterMap.entrySet()) {
    if (!leftNodeID.equals(entry.getKey()) && !nodesSet.contains(entry.getKey())) {
        JsonObject haInfo = new JsonObject(entry.getValue());
        checkFailover(entry.getKey(), haInfo);
    }
}

The same fix applies to the second call site at line 319:

// before:
if (clusterManager.getNodes().contains(nodeID) && ...
// after: use the already-constructed nodesSet
if (nodesSet.contains(nodeID) && ...

Complexity

	Before	After
`nodes.contains()` per call	O(N)	O(1)
Full `nodeLeft()` scan	O(N²)	O(N)
HashSet construction	—	O(N) one-time

Speedup estimate

At N=500 cluster nodes: ~500× operation-count reduction in the membership scan. Typically triggered once per node departure event, so the absolute time is small in small clusters. In large clusters (100–1000 nodes, common in cloud deployments) and high churn rates this becomes a significant stall on the event-loop thread.

Affected versions

All versions through current HEAD (2026-03-28).

2.7 KiB Raw Blame History Unescape Escape