[HIVE-13884] Disallow queries in HMS fetching more than a configured number of partitions

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.3.0
• Assignee: Sergio Peña
• Reporter: Mohit Sabharwal
• Created: 28/May/16 04:37
• Updated: 21/Jul/17 17:45
• Resolved: 02/Jul/16 16:27

Description

Currently the PartitionPruner requests either all partitions or partitions based on filter expression. In either scenarios, if the number of partitions accessed is large there can be significant memory pressure at the HMS server end.

We already have a config hive.limit.query.max.table.partition that enforces limits on number of partitions that may be scanned per operator. But this check happens after the PartitionPruner has already fetched all partitions.

We should add an option at PartitionPruner level to disallow queries that attempt to access number of partitions beyond a configurable limit.

Note that hive.mapred.mode=strict disallow queries without a partition filter in PartitionPruner, but this check accepts any query with a pruning condition, even if partitions fetched are large. In multi-tenant environments, admins could use more control w.r.t. number of partitions allowed based on HMS memory capacity.

One option is to have PartitionPruner first fetch the partition names (instead of partition specs) and throw an exception if number of partitions exceeds the configured value. Otherwise, fetch the partition specs.

Looks like the existing listPartitionNames call could be used if extended to take partition filter expressions like getPartitionsByExpr call does.

Issue Links

• relates to: HIVE-14298 NPE could be thrown in HMS when an ExpressionTree could not be made from a filter
• requires: HIVE-14055 directSql - getting the number of partitions is broken
• links to: Review Board