What causes "Aborting SSTable index flush for {}..." in apache/cassandra?

Read the attached cause in the WARN log — it names the real failure (usually IOException or OutOfMemoryError).

Aborting SSTable index flush for {}...

Error message

Aborting SSTable index flush for {}...

What it means

SSTableIndexWriter.abort is invoked when an SAI SSTable index write is aborted, from indexSSTable (exception during build), maybeAbort (segment writer failure), or writeSegmentsMetadata. Like MemtableIndexWriter.abort, cause == null is the benign no-flush-needed path (DEBUG); a non-null cause is a real failure logged at WARN. The abort releases segment-builder memory tracked by the global limiter and deletes partial on-disk index components.

Source

Thrown at src/java/org/apache/cassandra/index/sai/disk/v1/SSTableIndexWriter.java:187

        finally
        {
            index.indexMetrics().segmentsPerCompaction.update(segments.size());
            segments.clear();
            index.indexMetrics().compactionCount.inc();
        }
    }

    @Override
    public void abort(Throwable cause)
    {
        aborted = true;

        String message = index.identifier().logMessage("Aborting SSTable index flush for {}...");

        if (cause == null)
            logger.debug(message, indexDescriptor.sstableDescriptor);
        else 
            logger.warn(message, indexDescriptor.sstableDescriptor, cause);

        // It's possible for the current builder to be unassigned after we flush a final segment.
        if (currentBuilder != null)
        {
            // If an exception is thrown out of any writer operation prior to successful segment
            // flush, we will end up here, and we need to free up builder memory tracked by the limiter:
            long allocated = currentBuilder.totalBytesAllocated();
            long globalBytesUsed = currentBuilder.release();
            logger.debug(index.identifier().logMessage("Aborting index writer for SSTable {} released {}. Global segment memory usage now at {}."),
                         indexDescriptor.sstableDescriptor, FBUtilities.prettyPrintMemory(allocated), FBUtilities.prettyPrintMemory(globalBytesUsed));
        }

        indexDescriptor.deleteColumnIndex(index.termType(), index.identifier());
    }

    /**
     * abort current write if index is dropped
     *

View on GitHub (pinned to 88fd0f6a0e)

Solutions

Read the attached cause in the WARN log — it names the real failure (usually IOException or OutOfMemoryError).
If memory-related, lower cassandra.yaml `column_index_size_in_kb`/segment size or raise `sai.segment_write_buffer_size` headroom; ensure heap and off-heap limits accommodate concurrent index builds.
Free disk space in the data directory and re-trigger the build (`nodetool scsirebuild` or recreate the index).
If aborts recur on the same SSTable, drop and recreate the index to rebuild from a clean state.

Example fix

// before: default segment write memory, builds OOM under heavy load
# cassandra.yaml
sai_coordinator_resume: true
// after: tune limits and rebuild
# cassandra.yaml
sai: { segment_write_buffer_size: 1024 }
# then: nodetool scsirebuild --index my_index keyspace table

Defensive patterns

Strategy: try-catch

Try / catch

// ensure memory tracked by the limiter is released even on unexpected paths
try { writeSegmentsMetadata(); }
catch (Throwable t) { writer.abort(t); throw t; }

Prevention

Size segment write buffers and heap for concurrent index builds; watch the SAI segment memory limiter metrics.
Ensure adequate free disk space before large index builds.
Rebuild affected indexes after repeated aborts rather than relying on partial state.

When it happens

Trigger: An exception is thrown out of any writer operation before the segment flush succeeds (disk I/O error, out-of-memory in the index write buffer); the flush is cancelled because the final segment was already written; index_sai_segment_write_memory_limit exhaustion causing writer failure.

Common situations: Disk full during index build; write buffer allocation failures under the segment write memory limit; node shutdown mid-build; concurrent SSTable lifecycle changes invalidating the in-progress build.

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

AI-assisted analysis of apache/cassandra@88fd0f6a0e (2026-09-10). Data as JSON: /api/errors/ec58aa78a2062e9d. Report an issue: GitHub.