[CALCITE-1594] ConventionTraitDef.getConversionData() is not thread-safe

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.12.0
• Assignee: Julian Hyde
• Reporter: Julian Hyde
• Created: 19/Jan/17 21:21
• Updated: 24/Mar/17 03:20
• Resolved: 23/Jan/17 19:30

Description

Test suite hangs with several threads in WeakHashMap.get, all called from ConventionTraitDef.getConversionData and using the same table.

It is not a deadlock; all but one threads are waiting for a lock, but the thread that has the lock and is runnable.

Abbreviated stack trace:

Full thread dump Java HotSpot(TM) 64-Bit Server VM (25.72-b15 mixed mode):
...
"pool-1-thread-7" #20 daemon prio=5 os_prio=0 tid=0x00007f3cf4003000 nid=0x1f12 runnable [0x00007f3d3f8f4000]
   java.lang.Thread.State: RUNNABLE
	at java.util.WeakHashMap.expungeStaleEntries(WeakHashMap.java:341)
	- locked <0x00000000a08d14a8> (a java.lang.ref.ReferenceQueue)
	at java.util.WeakHashMap.getTable(WeakHashMap.java:350)
	at java.util.WeakHashMap.get(WeakHashMap.java:397)
	at org.apache.calcite.plan.ConventionTraitDef.getConversionData(ConventionTraitDef.java:203)
	at org.apache.calcite.plan.ConventionTraitDef.convert(ConventionTraitDef.java:129)
	at org.apache.calcite.plan.ConventionTraitDef.convert(ConventionTraitDef.java:46)
	at org.apache.calcite.plan.volcano.VolcanoPlanner.changeTraitsUsingConverters(VolcanoPlanner.java:1197)
...
"pool-1-thread-8" #19 daemon prio=5 os_prio=0 tid=0x00007f3cec001000 nid=0x1f11 waiting for monitor entry [0x00007f3d3f9f5000]
   java.lang.Thread.State: BLOCKED (on object monitor)
	at java.util.WeakHashMap.expungeStaleEntries(WeakHashMap.java:319)
	- waiting to lock <0x00000000a08d14a8> (a java.lang.ref.ReferenceQueue)
	at java.util.WeakHashMap.getTable(WeakHashMap.java:350)
	at java.util.WeakHashMap.get(WeakHashMap.java:397)
	at org.apache.calcite.plan.ConventionTraitDef.getConversionData(ConventionTraitDef.java:203)
	at org.apache.calcite.plan.ConventionTraitDef.convert(ConventionTraitDef.java:129)
	at org.apache.calcite.plan.ConventionTraitDef.convert(ConventionTraitDef.java:46)
	at org.apache.calcite.plan.volcano.VolcanoPlanner.changeTraitsUsingConverters(VolcanoPlanner.java:1197)
...
"pool-1-thread-6" #17 daemon prio=5 os_prio=0 tid=0x00007f3ce8001000 nid=0x1f0f waiting for monitor entry [0x00007f3d3fbf8000]
   java.lang.Thread.State: BLOCKED (on object monitor)
	at java.util.WeakHashMap.expungeStaleEntries(WeakHashMap.java:319)
	- waiting to lock <0x00000000a08d14a8> (a java.lang.ref.ReferenceQueue)
	at java.util.WeakHashMap.getTable(WeakHashMap.java:350)
	at java.util.WeakHashMap.get(WeakHashMap.java:397)
	at org.apache.calcite.plan.ConventionTraitDef.getConversionData(ConventionTraitDef.java:203)
...

Hypothesis: we should be synchronizing access to ConventionTraitDef.plannerConversionMap.

Issue Links

• is duplicated by: CALCITE-1550 ConventionTraitDef.plannerConversionMap is not thread safe