cuijianwei created HBASE-11623:
----------------------------------
Summary: mutateRowsWithLocks might require updatesLock.readLock
with waitTime=0
Key: HBASE-11623
URL: https://issues.apache.org/jira/browse/HBASE-11623
Project: HBase
Issue Type: Improvement
Components: regionserver
Affects Versions: 0.98.4, 0.94.21, 0.96.1.1
Reporter: cuijianwei
Priority: Minor
mutateRowsWithLocks will acquire updatesLock.readLock by the following code:
{code}
...
lock(this.updatesLock.readLock(), acquiredRowLocks.size());
...
{code}
However, acquiredRowLocks might be empty, and then the waitTime of
HRegion.lock(...) will be set to 0, which will make mutateRowsWithLocks fail if
can not acquire updatesLock.readLock immediately.
In our environment, we implement a region coprocessor which need to hold row
locks before invoke mutateRowsWithLocks. Then, the rowsToLock(passed to
mutateRowsWithLocks) will be an empty set, and we get the following exception
occasionally:
{code}
org.apache.hadoop.hbase.RegionTooBusyException: failed to get a lock in 0ms
582 at org.apache.hadoop.hbase.regionserver.HRegion.lock(HRegion.java:6191)
583 at
org.apache.hadoop.hbase.regionserver.HRegion.mutateRowsWithLocks(HRegion.java:5126)
584 at
org.apache.hadoop.hbase.regionserver.HRegion.mutateRowsWithLocks(HRegion.java:5034)
...
{code}
Is it reasonable that we use default waitTime when rowsToLock is empty? (as the
following code)
{code}
lock(this.updatesLock.readLock(), acquiredRowLocks.size() == 0 ? 1 :
acquiredRowLocks.size() );
{code}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
