[HIVE-26657] [Iceberg] Filter out the metadata.json file when migrating

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.0.0-alpha-2
• Component/s: Iceberg integration
• Labels: pull-request-available
• Assignee: László Pintér
• Reporter: László Pintér
• Created: 21/Oct/22 07:50
• Updated: 16/Nov/22 13:50
• Resolved: 24/Oct/22 11:14

Description

When migrating a hive table to an iceberg in certain cases a Runtime exception is raised

ERROR : Failed
java.lang.RuntimeException: s3a://dev-nfqe-base/cc-cdw-nfqe-q7wj9a/archive/env-8pt556/parquet/bakeoff/large/pli/metadata/00000-94fffe5c-c307-4341-9ea3-f5fa4863d301.metadata.json is not a Parquet file. Expected magic number at tail, but found [32, 93, 10, 125]

The hive-to-iceberg table migration has the following logic.

1. In order to walk through all the data files we request a file iterator from the filesystem. This iterator will provide all the references to be able to scan the data files.
2. The new iceberg table is created, meaning that a new entry is added to the hive catalog and on the file system level the metadata directory is created together with the first metadata file (*.metadata.json)
3. All the data files are scanned and the manifests are created.

The issue occurs when there are so many data files that it doesn't fit into memory in one go. So in step 3 when we walk through the data files list, the iterator has to run another round of file listing that reads up the content of the metadata directory that was created in step 2.

Issue Links

• links to: GitHub Pull Request #3695