Keith Turner created ACCUMULO-1962:
--------------------------------------
Summary: Improve batch scanner throughput in failure case
Key: ACCUMULO-1962
URL: https://issues.apache.org/jira/browse/ACCUMULO-1962
Project: Accumulo
Issue Type: Improvement
Reporter: Keith Turner
Fix For: 1.7.0
The batch scanner currently does the following.
# bin ranges to tablets servers and tabelts
# if any ranges could not be binned (e.g. a tablet had no location) goto 1
# queue up work for tablet servers on a thread pool
# wait for thread pool to complete all work
# if there were any failures goto 1
In the face of failures (tablets not assigned because of migration, tablet
servers dying) it would be better if the batch scanner worked on what it could
and immediately requeued failures for processing immediately. The
ConditionalWriter and BatchWriter have failure queues and increase the retry
time if something keeps failing.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
