apache/rocketmq · warning · MQClientException

Fetch consume offset from broker exception

Error message

Fetch consume offset from broker exception

What it means

DefaultLitePullConsumerImpl.committed(MessageQueue) calls the offsetStore with MEMORY_FIRST_THEN_STORE; when the store lookup returns -2 (OFFSET_OVERFLOW / no readable offset — neither memory nor broker had a committed offset for the queue), it throws this MQClientException. It means the consumer group has never committed an offset for that queue anywhere.

Source

Thrown at client/src/main/java/org/apache/rocketmq/client/impl/consumer/DefaultLitePullConsumerImpl.java:831

    private void submitConsumeRequest(ConsumeRequest consumeRequest) {
        try {
            consumeRequestCache.put(consumeRequest);
        } catch (InterruptedException e) {
            log.error("Submit consumeRequest error", e);
        }
    }

    private long fetchConsumeOffset(MessageQueue messageQueue) throws MQClientException {
        checkServiceState();
        long offset = this.rebalanceImpl.computePullFromWhereWithException(messageQueue);
        return offset;
    }

    public long committed(MessageQueue messageQueue) throws MQClientException {
        checkServiceState();
        long offset = this.offsetStore.readOffset(messageQueue, ReadOffsetType.MEMORY_FIRST_THEN_STORE);
        if (offset == -2) {
            throw new MQClientException("Fetch consume offset from broker exception", null);
        }
        return offset;
    }

    private void clearMessageQueueInCache(MessageQueue messageQueue) {
        ProcessQueue processQueue = assignedMessageQueue.getProcessQueue(messageQueue);
        if (processQueue != null) {
            processQueue.clear();
        }
        Iterator<ConsumeRequest> iter = consumeRequestCache.iterator();
        while (iter.hasNext()) {
            if (iter.next().getMessageQueue().equals(messageQueue)) {
                iter.remove();
            }
        }
    }

    private long nextPullOffset(MessageQueue messageQueue) throws MQClientException {

View on GitHub (pinned to 293f588571)

Solutions

1. Poll at least once (letting auto-commit run) before querying committed()
2. Treat this error as 'no commit yet' and fall back to minOffset/maxOffset or consumer launch configuration (fromWhere)
3. Ensure autoCommit is enabled or call commitSync() before querying

Example fix

// before
long off = consumer.committed(queue);

// after
long off;
try {
    off = consumer.committed(queue);
} catch (MQClientException e) {
    off = consumer.minOffset(queue); // no commit yet: start from earliest
}

try {
    offset = consumer.committed(q);
} catch (MQClientException e) {
    offset = consumer.minOffset(q); // group never committed for this queue
}

When it happens

Trigger: Calling committed(q) right after start() before any message has been pulled and auto-committed; a brand-new consumer group (CONSUME_FROM_LAST_OFFSET never materialized) on a queue that was just assigned.

Common situations: New group on an existing topic; checking committed offset for a queue after seek() cleared committed state; offset store wiped (new client instance with remote offsets not yet persisted).

Related errors

• Seek offset illegal, seek offset =
• offset < 0
• QUERY_NOT_FOUND
• SYSTEM_ERROR
• Failed to get max offset

AI-assisted analysis of apache/rocketmq@293f588571 (2026-08-14). Data as JSON: /api/errors/3079a953b43f7d14. Report an issue: GitHub.