Mostafa Mokhtar created IMPALA-6406:
---------------------------------------
Summary: Consider partitioning metastoreDdlLock_ by database
Key: IMPALA-6406
URL: https://issues.apache.org/jira/browse/IMPALA-6406
Project: IMPALA
Issue Type: Improvement
Components: Catalog
Reporter: Mostafa Mokhtar
Concurrent DML operations occasionally block on metastoreDdlLock_, for cases
where a user is dropping a table with a large number of partitions other
operations trying to create a table in a different database block on
metastoreDdlLock_
Partitioning the lock by database should help a lot, but ideally it should be
hierarchal.
{code}
"Thread-197" #250 prio=5 os_prio=0 tid=0x000000000ef00000 nid=0x88af waiting
for monitor entry [0x00007f1e9eb44000]
java.lang.Thread.State: BLOCKED (on object monitor)
at
org.apache.impala.service.CatalogOpExecutor.createTable(CatalogOpExecutor.java:1666)
- waiting to lock <0x00000003c066aa90> (a java.lang.Object)
at
org.apache.impala.service.CatalogOpExecutor.createTable(CatalogOpExecutor.java:1543)
at
org.apache.impala.service.CatalogOpExecutor.execDdlRequest(CatalogOpExecutor.java:274)
at org.apache.impala.service.JniCatalog.execDdl(JniCatalog.java:147)
{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
