Table layout & maintenance

5 h
requires LKH-04

From your world. Redshift VACUUM / SORTKEY, and the small-files problem on Athena

Practice quiz with instant feedback. For a scored run that counts toward your local progress, use uv run de quiz LKH-05.

1A streaming job commits to an Iceberg table every few seconds. It produces tens of thousands of tiny data files. Queries have become slow. Which maintenance operation directly fixes this? Aexpire_snapshots — it deletes the small files Brewrite_data_files — it reads the small files and rewrites them into fewer, larger ones Cremove_orphan_files — it merges the files DVACUUM — Iceberg has no equivalent

2You run expire_snapshots with retain_last => 5 but no older_than. Disk usage barely drops, and this surprises you. Why? Aretain_last is ignored unless you also pass a table name Bexpire_snapshots also uses a default max_snapshot_age (≈5 days), so recent snapshots are protected whatever the value of retain_last CSnapshots cannot be expired in Spark, only in Flink DThe table has no snapshots

3Which Iceberg feature is the closest equivalent of a Redshift SORTKEY? It clusters rows, so that a predicate can skip whole data files with their min/max statistics. AA high-cardinality partition column BWRITE ORDERED BY (a table sort order), honored by compaction Cexpire_snapshots Dremove_orphan_files

4A teammate partitions a 10-billion-row events table by order_id "so the data is spread out". What goes wrong? ANothing. High-cardinality partitioning is a best practice BIt creates roughly one tiny file per order_id — the small-files problem by construction — and an enormous metadata/manifest footprint CIceberg rejects more than 1000 partitions DQueries become case-sensitive

5A table uses Iceberg hidden partitioning on days(event_ts). How should an analyst filter to one day, and why? AWHERE dt = '2026-01-01' using a physical partition string column they must add BWHERE event_ts >= TIMESTAMP '2026-01-01' AND event_ts < TIMESTAMP '2026-01-02' — Iceberg applies the days() transform and prunes partitions automatically CThey cannot filter on a partitioned table DWHERE days(event_ts) = 1

6You need to change an append-only table from unpartitioned to partitioned by days(ts). You want no full table rewrite. What does Iceberg let you do? ANothing — you must recreate and reload the table BPartition evolution: ALTER TABLE ... ADD PARTITION FIELD days(ts); old data keeps its spec, new data uses the new one COnly Delta supports changing partitioning DDrop and re-add every data file manually

7expire_snapshots deleted the data files of old snapshots. But some files from an aborted write still remain, and no snapshot references them. Which operation removes those files? Arewrite_data_files Bexpire_snapshots again with retain_last => 0 Cremove_orphan_files, which lists the table directory and deletes unreferenced files older than a cutoff DDROP TABLE