apache/cassandra · error · java.lang.IllegalStateException

0x0000

0x0000

Error message

Cannot rebuild index %s as another index build for the same index is currently in progress.

What it means

IllegalStateException thrown by SecondaryIndexManager.markIndexesBuilding when a full rebuild is requested for an index that already has an in-progress build (inProgressBuilds counter > 0). Prevents two concurrent builds of the same index.

Source

Thrown at src/java/org/apache/cassandra/index/SecondaryIndexManager.java:816

     * @param isFullRebuild {@code true} if this method is invoked as a full index rebuild, {@code false} otherwise
     * @param isNewCF {@code true} if this method is invoked when initializing a new table/columnfamily (i.e. loading a CF at startup),
     * {@code false} for all other cases (i.e. newly added index)
     */
    @VisibleForTesting
    public synchronized void markIndexesBuilding(Set<Index> indexes, boolean isFullRebuild, boolean isNewCF)
    {
        String keyspaceName = baseCfs.getKeyspaceName();

        // First step is to validate against concurrent rebuilds; it would be more optimized to do everything on a single
        // step, but we're not really expecting a very high number of indexes, and this isn't on any hot path, so
        // we're favouring readability over performance
        indexes.forEach(index ->
                        {
                            String indexName = index.getIndexMetadata().name;
                            AtomicInteger counter = inProgressBuilds.computeIfAbsent(indexName, ignored -> new AtomicInteger(0));

                            if (counter.get() > 0 && isFullRebuild)
                                throw new IllegalStateException(String.format("Cannot rebuild index %s as another index build for the same index is currently in progress.", indexName));
                        });

        // Second step is the actual marking:
        indexes.forEach(index ->
                        {
                            String indexName = index.getIndexMetadata().name;
                            AtomicInteger counter = inProgressBuilds.computeIfAbsent(indexName, ignored -> new AtomicInteger(0));

                            if (isFullRebuild)
                            {
                                needsFullRebuild.remove(indexName);
                                makeIndexNonQueryable(index, Index.Status.FULL_REBUILD_STARTED);
                            }

                            if (counter.getAndIncrement() == 0 && DatabaseDescriptor.isDaemonInitialized() && !isNewCF)
                                SystemKeyspace.setIndexRemoved(keyspaceName, indexName);
                        });
    }

View on GitHub (pinned to 88fd0f6a0e)

Solutions

1. Wait for the current build to finish (check system.built / in-progress tasks, `nodetool compactionstats`) before rebuilding
2. If the first build is genuinely stuck, restart the node to clear in-progress state, then rebuild
3. Serialize rebuild operations in automation; do not retry in a tight loop
4. Use `nodetool indexbuilds`/logs to confirm the prior build completed

Example fix

// before
cfim.rebuildIndexesBlocking(Set.of("idx")); // while already building
// after
if (!isIndexBuilding("idx")) {
    cfim.rebuildIndexesBlocking(Set.of("idx"));
} else {
    logger.info("Index idx still building; skipping rebuild");
}

// check for an in-progress build before requesting a rebuild
if (!indexBuildsComplete(keyspace, table, indexName)) {
    logger.info("Index {} still building; rebuild skipped", indexName);
    return;
}

try {
    indexManager.rebuildIndexesBlocking(indexes);
} catch (IllegalStateException e) {
    if (e.getMessage().contains("currently in progress"))
        logger.info("Rebuild skipped: build already running");
    else throw e;
}

When it happens

Trigger: Calling rebuild of an index (REBUILD INDEXES / buildIndexesBlocking / createIndex path) while the same index is still building — e.g. issuing `nodetool rebuild_index` twice, or creating an index while its initial build is running.

Common situations: Ops re-running a stuck-seeming REBUILD INDEXES without waiting; schema-change and manual rebuild racing after node restart; scripted retries firing while the first build is still active.

Related errors

• Node is still rebuilding. Check nodetool netstats.
• Bootstrap can be started exactly once, but seems to have alr
• Transfer of stream %s already completed or aborted (perhaps
• Index build of {} failed. Please run full index rebuild to f
• Queue is empty

AI-assisted analysis of apache/cassandra@88fd0f6a0e (2026-09-10). Data as JSON: /api/errors/d33bb67e65127017. Report an issue: GitHub.