[KUDU-1682] Lock contention on table locations cache in Java client

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.0.0
• Fix Version/s: 1.1.0
• Component/s: client, perf
• Target Version/s: 1.1.0
• Code Review: https://gerrit.cloudera.org/#/c/4706/
• Assignee: Jean-Daniel Cryans
• Reporter: Todd Lipcon
• Created: 06/Oct/16 03:07
• Updated: 13/Oct/16 15:54
• Resolved: 13/Oct/16 15:53

Description

Running YCSB with 16 threads, I see a lot of contention on the table locations cache:

"Thread-15" #36 prio=5 os_prio=0 tid=0x00007f2300391000 nid=0xfb2 waiting for monitor entry [0x00007f229a452000]
   java.lang.Thread.State: BLOCKED (on object monitor)
        at org.apache.kudu.client.TableLocationsCache.get(TableLocationsCache.java:62)
        - locked <0x00000001da767c78> (a java.lang.Object)
        at org.apache.kudu.client.AsyncKuduClient.getTableLocationEntry(AsyncKuduClient.java:1429)
        at org.apache.kudu.client.AsyncKuduClient.loopLocateTable(AsyncKuduClient.java:1184)
        at org.apache.kudu.client.AsyncKuduClient.locateTable(AsyncKuduClient.java:1240)
        at org.apache.kudu.client.AsyncKuduClient.getTabletLocation(AsyncKuduClient.java:1449)

Since this changes rarely, we should make it a read-write lock.