apache/cassandra · warning

Could not discover CMS from

Error message

Could not discover CMS from 

What it means

RemoteProcessor.tryDiscover attempts to discover the Cluster Metadata Service (CMS) by contacting remote nodes; when the discovery promise does not complete within DatabaseDescriptor.getCmsAwaitTimeout(), the timeout/exception is logged as a warning and an empty DiscoveredNodes with Kind.KNOWN_PEERS is returned so the caller (cms) can fall back to known peers. This is a degraded-discovery warning, not a fatal error.

Solutions

1. Verify the CMS hosts are running and reachable (nodetool status, ping/check rpc_address and broadcast_address).
2. Increase cms_await_timeout in cassandra.yaml (or the corresponding system property) to allow slower discovery.
3. Confirm internode connectivity and TLS settings; check for firewall or wrong seed/address configuration.
4. If this recurs at every startup, check the system.cluster_metadata logs on candidates to confirm which node actually hosts the CMS and re-run CMS relocation if needed.

Example fix

// cassandra.yaml
// before
cms_await_timeout: 10s
// after
cms_await_timeout: 60s

// Before relying on discovery, check CMS reachability
if (!ClusterMetadataService.instance().isCurrent() && ClusterMetadataService.instance().isRunning()) {
    // discovery should succeed; otherwise expect timeout fallback
}

// Discovery degrades gracefully; handle the empty/KNOWN_PEERS result
DiscoveredNodes found = processor.tryDiscover(ep);
if (found.kind() == DiscoveredNodes.Kind.KNOWN_PEERS && found.nodes().isEmpty()) {
    logger.warn("CMS discovery timed out; retrying after connectivity check");
}

When it happens

Trigger: Calling ClusterMetadataService.discover() (via RemoteProcessor.tryDiscover) when remote candidates do not respond before the cms_await_timeout elapses; network partition, firewall, or the target CMS host being down while the node is booting or re-registering with the cluster.

Common situations: Node joining a cluster whose CMS host is stopped or unreachable; misconfigured broadcast/rpc addresses so discovery RPCs go to the wrong IP; cms_await_timeout set too low for slow networks; starting a node while rest of cluster is still down.

Understand the failure class

Background: Request timed out: what client-side request timeouts mean across libraries (Request timed out, TIMED_OUT, APITimeoutError) — this error's family across 39 libraries.

Related errors

• Configured " " could not be found
• Configured " " was found, but had no addresses
• Could not catchup with peers
• Could not fetch log entries from peer, remote =
• Failed to bind port on .

AI-assisted analysis of apache/cassandra@88fd0f6a0e (2026-09-10). Data as JSON: /api/errors/313f0ff7b14e24af. Report an issue: GitHub.

Appendix: source

Thrown at src/java/org/apache/cassandra/tcm/RemoteProcessor.java:416

            public void onResponse(Message<DiscoveredNodes> msg)
            {
                promise.setSuccess(msg.payload);
            }

            @Override
            public void onFailure(InetAddressAndPort from, RequestFailure failureReason)
            {
                // "success" - this lets us just try the next one in cmsIter
                promise.setSuccess(new DiscoveredNodes(Collections.emptySet(), DiscoveredNodes.Kind.KNOWN_PEERS));
            }
        });
        try
        {
            return promise.get(DatabaseDescriptor.getCmsAwaitTimeout().to(TimeUnit.MILLISECONDS), TimeUnit.MILLISECONDS);
        }
        catch (Exception e)
        {
            logger.warn("Could not discover CMS from " + ep, e);
        }
        return new DiscoveredNodes(Collections.emptySet(), DiscoveredNodes.Kind.KNOWN_PEERS);
    }

    public static class CandidateIterator extends AbstractIterator<InetAddressAndPort>
    {
        private final Deque<InetAddressAndPort> candidates;
        private final Set<InetAddressAndPort> elements;
        private final boolean checkLive;

        @SuppressWarnings("resource")
        public CandidateIterator(Collection<InetAddressAndPort> initialContacts)
        {
            this(initialContacts, true);
        }

        @SuppressWarnings("resource")
        public CandidateIterator(Collection<InetAddressAndPort> initialContacts, boolean checkLive)

View on GitHub (pinned to 88fd0f6a0e)