prestodb/presto · error · PrestoException

HIVE_INVALID_METADATA

HIVE_INVALID_METADATA

Error message

Table '%s' or partition '%s' has null columns

What it means

Schema validation during split computation: either the table's dataColumns or the partition's columns list is null in the metastore. Presto requires both column lists to build a TableToPartitionMapping, so it throws HIVE_INVALID_METADATA — the Hive metadata itself is corrupt/incomplete.

Source

Thrown at presto-hive/src/main/java/com/facebook/presto/hive/HiveSplitManager.java:633

                        if (partitionsNotReadable.size() <= 3) {
                            partitionsNotReadable.putIfAbsent(reason, new HashSet<>(ImmutableSet.of(partitionName)));
                            if (partitionsNotReadable.get(reason).size() <= 3) {
                                partitionsNotReadable.get(reason).add(partitionName);
                            }
                        }
                        continue;
                    }
                }

                // Verify that the partition schema matches the table schema.
                // Either adding or dropping columns from the end of the table
                // without modifying existing partitions is allowed, but every
                // column that exists in both the table and partition must have
                // the same type.
                List<Column> tableColumns = table.getDataColumns();
                List<Column> partitionColumns = partition.getColumns();
                if ((tableColumns == null) || (partitionColumns == null)) {
                    throw new PrestoException(HIVE_INVALID_METADATA, format("Table '%s' or partition '%s' has null columns", tableName, partitionName));
                }
                TableToPartitionMapping tableToPartitionMapping = getTableToPartitionMapping(session, resolvedHiveStorageFormat, tableName, partitionName, tableColumns, partitionColumns);

                if (hiveBucketHandle.isPresent() && !hiveBucketHandle.get().isVirtuallyBucketed()) {
                    Optional<HiveBucketProperty> partitionBucketProperty = partition.getStorage().getBucketProperty();
                    if (!partitionBucketProperty.isPresent()) {
                        throw new PrestoException(HIVE_PARTITION_SCHEMA_MISMATCH, format(
                                "Hive table (%s) is bucketed but partition (%s) is not bucketed",
                                hivePartition.getTableName(),
                                hivePartition.getPartitionId().getPartitionName()));
                    }
                    int tableBucketCount = hiveBucketHandle.get().getTableBucketCount();
                    int partitionBucketCount = partitionBucketProperty.get().getBucketCount();
                    List<String> tableBucketColumns = hiveBucketHandle.get().getColumns().stream()
                            .map(HiveColumnHandle::getName)
                            .collect(toImmutableList());
                    List<String> partitionBucketColumns = partitionBucketProperty.get().getBucketedBy();
                    if (!tableBucketColumns.equals(partitionBucketColumns) || !isBucketCountCompatible(tableBucketCount, partitionBucketCount)) {

View on GitHub (pinned to 55bb57d202)

Solutions

1. Run MSCK REPAIR TABLE or use ALTER TABLE ADD/DROP COLUMNS to restore column metadata.
2. Inspect the offending partition's metadata in the metastore (describe formatted <table> partition (...)) and recreate the partition with correct schema.
3. Recreate the table/partition via Presto or Hive with a proper schema.
4. Check for external writers or manual metastore edits that produced null columns.

Example fix

// before
ALTER TABLE t ADD IF NOT EXISTS PARTITION (dt='2026-09-01'); -- metadata written with null columns by external tool
// after
ALTER TABLE t ADD IF NOT EXISTS PARTITION (dt='2026-09-01') LOCATION '...'; -- created via Hive/Presto so column schema is populated

// pre-check via metastore client
Table t = metastore.getTable(db, name);
if (t.getSd().getCols() == null) { repairTableMetadata(); }
// or in SQL first: DESCRIBE <table>; verify columns non-empty

catch (PrestoException e) { if (e.getErrorCode().getCode() == HIVE_INVALID_METADATA.toErrorCode().getCode()) { /* repair metadata then retry */ } else throw e; }

When it happens

Trigger: computePartitionMetadata reads table.getDataColumns() or partition.getColumns() and gets null, typically because the metastore (or a direct thrift response) has a partition/table row with no SDS/column metadata.

Common situations: Partitions created by external tools that wrote incomplete metadata; metastore DB corruption or manual row edits; restore from backup missing column-schema rows.

Related errors

• HIVE_INVALID_METADATA
• HIVE_INVALID_METADATA
• SCHEMA_NOT_EMPTY
• Table not found: ${databaseName}.${tableName}
• ARROW_FLIGHT_METADATA_ERROR

AI-assisted analysis of prestodb/presto@55bb57d202 (2026-09-04). Data as JSON: /api/errors/062483ef0ff58fea. Report an issue: GitHub.