Unable to recover %d partitions that were skipped. You can

Error message

Unable to recover %d partitions that were skipped.  You can attempt manual recovery from the pre-scrub snapshot.  You can also run nodetool repair to transfer the data from a healthy replica, if any

What it means

At the end of a scrub, outputSummary reports partitions that were corrupt/unreadable and had to be skipped (badPartitions > 0) while some valid data was recovered. It directs the operator to the pre-scrub snapshot for manual recovery or to run nodetool repair so a healthy replica can re-stream the lost data.

Source

Thrown at src/java/org/apache/cassandra/io/sstable/format/SortedTableScrubber.java:227

        {
            if (transaction.isOffline())
                finished.forEach(sstable -> sstable.selfRef().release());
        }

        outputSummary(finished);
    }

    protected abstract void scrubInternal(SSTableRewriter writer) throws IOException;

    private void outputSummary(List<SSTableReader> finished)
    {
        if (!finished.isEmpty())
        {
            outputHandler.output("Scrub of %s complete: %d partitions in new sstable and %d empty (tombstoned) partitions dropped", sstable, goodPartitions, emptyPartitions);
            if (negativeLocalDeletionInfoMetrics.fixedRows > 0)
                outputHandler.output("Fixed %d rows with overflowed local deletion time.", negativeLocalDeletionInfoMetrics.fixedRows);
            if (badPartitions > 0)
                outputHandler.warn("Unable to recover %d partitions that were skipped.  You can attempt manual recovery from the pre-scrub snapshot.  You can also run nodetool repair to transfer the data from a healthy replica, if any", badPartitions);
        }
        else
        {
            if (badPartitions > 0)
                outputHandler.warn("No valid partitions found while scrubbing %s; it is marked for deletion now. If you want to attempt manual recovery, you can find a copy in the pre-scrub snapshot", sstable);
            else
                outputHandler.output("Scrub of %s complete; looks like all %d partitions were tombstoned", sstable, emptyPartitions);
        }
    }

    private SSTableReader writeOutOfOrderPartitions(StatsMetadata metadata)
    {
        // out of order partitions/rows, but no bad partition found - we can keep our repairedAt time
        long repairedAt = badPartitions > 0 ? ActiveRepairService.UNREPAIRED_SSTABLE : sstable.getSSTableMetadata().repairedAt;
        SSTableReader newInOrderSstable;
        try (SSTableWriter inOrderWriter = CompactionManager.createWriter(cfs, destination, expectedBloomFilterSize, repairedAt, metadata.pendingRepair, metadata.isTransient, sstable, transaction))
        {
            for (Partition partition : outOfOrder)

View on GitHub (pinned to 88fd0f6a0e)

Solutions

Run nodetool repair on the affected range/node so a healthy replica re-streams the missing partitions
If no healthy replica exists, extract the lost partitions manually from the pre-scrub snapshot sstables
Check host hardware/filesystem health (smartctl, fsck) before trusting the node again
Verify replication factor is adequate so single-replica corruption is recoverable

Defensive patterns

Strategy: retry

Try / catch

// After scrub, check reported bad partitions and react:
if (scrubReport.badPartitions > 0) { triggerRepair(keyspace, table); verifyPreScrubSnapshotExists(); }

Prevention

Use RF >= 3 so scrub data loss can be repaired from replicas
Take snapshots before scrubbing important tables
Monitor disk SMART/health to catch corruption early
Schedule scrub during low traffic and inspect its output

When it happens

Trigger: nodetool scrub completes with some partitions unreadable due to corruption (bad checksums, truncated file, invalid structure); emitted after the new sstable(s) are written, only when finished is non-empty.

Common situations: See trigger scenarios.

Understand the failure class

Background: Checksum mismatch errors: "checksum verification failed", "digest mismatch", "expected vs actual checksum" — what they mean and how to fix them — this error's family across 41 libraries.

Related errors

AI-assisted analysis of apache/cassandra@88fd0f6a0e (2026-09-10). Data as JSON: /api/errors/291223b28785f415. Report an issue: GitHub.