apache/iceberg · warning

Failed to get deleted files: this may cause orphaned data fi

Error message

Failed to get deleted files: this may cause orphaned data files

What it means

A WARN logged via Tasks' onFailure callback while CatalogUtil.deleteFiles scans manifests in parallel to collect files that are safe to delete after a table drop. If reading a manifest fails, the set of deleted files is incomplete — files referenced by the failed manifest will NOT be deleted, resulting in orphaned data/manifest files. The drop itself still proceeds; this warning signals storage leak, not data loss for surviving tables.

Source

Thrown at core/src/main/java/org/apache/iceberg/CatalogUtil.java:178

    if (gcEnabled) {
      deleteFile(io, metadata.metadataFileLocation(), "metadata");
    }
  }

  @SuppressWarnings("DangerousStringInternUsage")
  private static void deleteFiles(
      FileIO io, Set<ManifestFile> allManifests, Map<Integer, PartitionSpec> specsById) {
    // keep track of deleted files in a map that can be cleaned up when memory runs low
    Map<String, Boolean> deletedFiles =
        new MapMaker().concurrencyLevel(ThreadPools.WORKER_THREAD_POOL_SIZE).weakKeys().makeMap();

    Tasks.foreach(allManifests)
        .noRetry()
        .suppressFailureWhenFinished()
        .executeWith(ThreadPools.getWorkerPool())
        .onFailure(
            (item, exc) ->
                LOG.warn("Failed to get deleted files: this may cause orphaned data files", exc))
        .run(
            manifest -> {
              try (ManifestReader<?> reader = ManifestFiles.open(manifest, io, specsById)) {
                List<String> pathsToDelete = Lists.newArrayList();
                for (ManifestEntry<?> entry : reader.entries()) {
                  // intern the file path because the weak key map uses identity (==) instead of
                  // equals
                  String path = entry.file().location().intern();
                  Boolean alreadyDeleted = deletedFiles.putIfAbsent(path, true);
                  if (alreadyDeleted == null || !alreadyDeleted) {
                    pathsToDelete.add(path);
                  }
                }

                String type = reader.isDeleteManifestReader() ? "delete" : "data";
                deleteFiles(io, pathsToDelete, type, false);
              } catch (IOException e) {
                throw new RuntimeIOException(

View on GitHub (pinned to 86d9c8fc54)

Solutions

1. Re-run the drop with the table's metadata intact, or manually list and delete the files referenced by the failed manifests.
2. Fix storage permissions (FileIO credentials) so all manifests under the metadata location are readable.
3. Check object-store lifecycle rules so manifests are not deleted before the table is dropped.
4. Use the platform's orphan-file cleanup (e.g. Spark deleteOrphanFiles) to remove the leaked files.

Example fix

// before: drop fails silently leaving orphans
catalog.dropTable(identifier);
// after: sweep orphans afterwards
if (!catalog.tableExists(identifier)) {
  SparkActions.get().deleteOrphanFiles(spark, "s3://bucket/db/tbl").execute();
}

// ensure manifests readable before drop
allManifests.forEach(m -> ManifestFiles.read(m, io).close());

try { catalog.dropTable(identifier); } finally { scheduleOrphanFileSweep(tableLocation); }

When it happens

Trigger: dropTableData or deleteRemovedMetadataFiles triggers parallel manifest reads; a manifest file is unreadable — deleted/corrupted in object storage, missing permissions, missing spec entry (specsById mismatch), or transient S3/GCS errors.

Common situations: Table partially corrupted by a prior failed cleanup; IAM/permission changes between metadata read and manifest read; lifecycle policies deleting manifest files early; cross-region storage issues.

Understand the failure class

Background: "failed to read file", EACCES, ENOENT and "could not read <path>" errors: when a program can't read a file from disk — this error's family across 49 libraries.

Related errors

• Failed to read manifest file: %s
• Failed to validate replaced partitions
• Failed to write manifest
• Failed to read manifest file: %s
• Failed to read manifest file: %s

AI-assisted analysis of apache/iceberg@86d9c8fc54 (2026-09-12). Data as JSON: /api/errors/07d70234a09a53d6. Report an issue: GitHub.