Geoffrey Jacoby created PHOENIX-5344:
----------------------------------------
Summary: MapReduce Jobs Over Salted Snapshots Give Wrong Results
Key: PHOENIX-5344
URL: https://issues.apache.org/jira/browse/PHOENIX-5344
Project: Phoenix
Issue Type: Bug
Reporter: Geoffrey Jacoby
I'm modifying an existing MapReduce job to use Phoenix's MapReduce / HBase
snapshot integration. When testing, I noticed that existing tests that had
previously worked for this job when running on salted Phoenix tables began to
fail when running on a snapshot of those tables. They pass when running
identical logic against the live table. Unsalted tables give the same, correct
result whether running against a live table or a snapshot.
The symptom on the salted snapshots is that the row count is way too high (a
factor of about 7x), but the exact amount appears non-deterministic.
My working theory is that somewhere the snapshot MapReduce integration for
Phoenix sets up the scans improperly for salted tables.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
