Spark SQL without the notebook.
Spark 4.1 HiveServer2 binary and HTTP connections with TLS, Kerberos, and SSH, catalog navigation, Iceberg CRUD, and partition advice — in a real desktop workbench.
Spark 4.1 · HiveServer2 · TLS/Kerberos/SSH · Iceberg
The physical plan Mongrel would show for your query.
Click any operator below to see runtime stats, shuffle sizes, and what Mongrel would suggest. Same view you get inside the workbench.
// FileScan parquet · analytics.events
Iceberg scan reads the full table: 14,203 files, 0 pruned. Filter on event_date does not match the partition spec (bucket(16, user_id)) — the whole table is read.
-- catalog.analytics · spark-thrift-prod SELECT user_id, COUNT(*) AS pv FROM catalog.analytics.events WHERE event_date >= DATE '2026-06-01' GROUP BY user_id ORDER BY pv DESC NULLS LAST LIMIT 20; -- 42.1 GB read · 14,203 files, 0 pruned -- partition spec is bucket(16, user_id); -- filter on event_date can't prune -- ✓ Proposal: add a days(event_ts) partition field ALTER TABLE catalog.analytics.events ADD PARTITION FIELD days(event_ts);
What Mongrel ships for Spark SQL.
HiveServer2 connections
Binary and HTTP transport modes against Spark Thrift Server. TLS, Kerberos, and SSH tunnels are first-class profile options.
Physical plan viewer
Run EXPLAIN and walk the physical plan: scans, exchanges, aggregates, and joins with AQE runtime stats.
Catalog navigation
Browse catalogs, databases, tables, and views across your Spark session. Hive Metastore and Iceberg catalogs side by side.
Iceberg CRUD
Create, insert, update, and delete on Iceberg tables. Inspect snapshots, partition specs, and manifest layout from the table panel.
Partition advice
Mongrel flags full-table scans caused by partition-spec mismatches and proposes the partition transform that fixes them.
Shuffle & skew detection
Exchange nodes show bytes shuffled per task. Skewed partitions are highlighted with the config that rebalances them.
Kerberos & keytab support
Authenticate with a principal and keytab or a ticket cache. Works with TLS endpoints and SSH-jump reachability.
SQL file browser
Read parquet metadata directly: row groups, compression, column statistics, and file-level pruning potential.
Session & job insight
Monitor running statements, stages, and executor metrics for the queries you launch from the workbench.