apache/cassandra · critical · IllegalStateException

Tried to commit when in gossip mode

Error message

Tried to commit when in gossip mode

What it means

Commit.checkCMSState validates the sender's ClusterMetadataService state before processing a commit via Commit.doVerb. When the local CMS state is GOSSIP, the node is operating in gossip-based cluster-metadata mode rather than TCM mode, and metadata commits are invalid; the code logs the message via logger.error and throws IllegalStateException. This is an internal-mode invariant check: a TCM commit verb should never arrive (or be processed) while the node is in gossip mode.

Source

Thrown at src/java/org/apache/cassandra/tcm/Commit.java:424

                logger.info("Responding with full result {} to sender {}", result, message.from());
            }

            messagingService.accept(message.responseWith(result), message.from());
        }

        private void checkCMSState()
        {
            switch (cmsStateSupplier.get())
            {
                case RESET:
                case LOCAL:
                    break;
                case REMOTE:
                    throw new NotCMSException("Not currently a member of the CMS, can't commit");
                case GOSSIP:
                    String msg = "Tried to commit when in gossip mode";
                    logger.error(msg);
                    throw new IllegalStateException(msg);
                default:
                    throw new IllegalStateException("Illegal state: " + cmsStateSupplier.get());
            }
        }
    }

    public interface Replicator
    {
        Replicator NO_OP = (a,b) -> {};
        void send(Result result, InetAddressAndPort source);
    }

    public static class DefaultReplicator implements Replicator
    {
        public static class RoutingHelper
        {
            private final Directory directory;
            private final EndpointLookup endpoints;

View on GitHub (pinned to 88fd0f6a0e)

Solutions

1. Verify the cluster is running in TCM mode: check ClusterMetadataService state and that the target node is a current CMS member before sending commits.
2. Ensure all nodes are upgraded and ClusterMetadataService.instance().isCurrentMember(address) is true for the recipient; re-run the CMS placement/registration if membership is missing.
3. Check for mixed-version or misconfigured nodes routing Commit verbs across mode boundaries; align node versions and retry after the node joins the CMS.
4. If this appears during a controlled gossip->TCM transition, allow the transition to complete before issuing metadata changes.

Example fix

// guard before sending a commit
if (ClusterMetadataService.instance().isCurrentMember(FBUtilities.getBroadcastAddressAndPort()))
{
    Commit.instance().commit(proposed);
}
else
{
    logger.warn("Skipping TCM commit: node is not a CMS member (gossip mode)");
}

if (ClusterMetadataService.instance().isCurrentMember(FBUtilities.getBroadcastAddressAndPort())
    && ClusterMetadataService.instance().state() != ClusterMetadataService.CMSState.GOSSIP)
{
    Commit.instance().commit(proposed);
}

try { Commit.instance().commit(proposed); } catch (IllegalStateException e) { if (e.getMessage().contains("gossip mode")) logger.warn("Node in gossip mode; TCM commit rejected"); else throw e; }

When it happens

Trigger: A TCM Commit verb is dispatched to doVerb while ClusterMetadataService state resolves to CMSState.GOSSIP (cluster not using TCM / CMS, or the node hasn't been registered as a CMS member); checkCMSState's switch on cmsStateSupplier.get() hits the GOSSIP case at Commit.java:424.

Common situations: Clusters migrated from or started in gossip mode receiving stale/misrouted TCM messages; a node whose CMS membership was lost or not yet established handling an in-flight commit; mixed-version rolling upgrade where an old node routes metadata commits to a gossip-mode node.

Understand the failure class

Background: "Invalid state transition" errors: "status must be X, actually Y", "already rejected/charging/uninstalled", "cannot ... while running" — what they mean when a library rejects your call — this error's family across 31 libraries.

Related errors

• Illegal state:
• Node %s is not a CMS member in epoch %s; members=%s
• Can only initialize cluster identifier during epoch %d, but
• Initial CMS node needs to be fully joined, not: %s
• Can't upgrade from gossip since CMS is already initialized

AI-assisted analysis of apache/cassandra@88fd0f6a0e (2026-09-10). Data as JSON: /api/errors/75a86552ad84df3d. Report an issue: GitHub.