[DRILL-4146] Concurrent queries hang in planner in ReflectiveRelMetadataProvider

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.3.0
• Fix Version/s: 1.4.0
• Component/s: Query Planning & Optimization
• Assignee: Aman Sinha
• Reporter: Aman Sinha
• Reviewer: Dechang Gu
• Created: 01/Dec/15 01:31
• Updated: 15/Dec/15 18:48
• Resolved: 01/Dec/15 07:33

Description

At concurrency levels of 30 or more for certain workloads we have seen queries hang in the planning phase in Calcite. The top of the jstack is shown below:

"29b47a17-6ef3-4b7f-98e7-a7c1a702c32f:foreman" daemon prio=10 tid=0x00007f55484a1800 nid=0x289a runnable [0x00007f54b4369000]
   java.lang.Thread.State: RUNNABLE
        at java.util.HashMap.getEntry(HashMap.java:465)
        at java.util.HashMap.get(HashMap.java:417)
        at org.apache.calcite.rel.metadata.ReflectiveRelMetadataProvider.apply(ReflectiveRelMetadataProvider.java:251)
        at org.apache.calcite.rel.metadata.ChainedRelMetadataProvider.apply(ChainedRelMetadataProvider.java:60)
        at org.apache.calcite.rel.metadata.ChainedRelMetadataProvider.apply(ChainedRelMetadataProvider.java:60)

After some investigations, we found that this issue was actually addressed by CALCITE-874 (ReflectiveRelMetadataProvider is not thread-safe). This JIRA is a placeholder to merge that Calcite fix since Drill is currently not up-to-date with Calcite and there is an immediate need for running queries in a high concurrency environment.