Thanks for your flow chart, I have a deeper understanding about how phoenix updating data table and its related index table.
And we have known that it is hbase-indexer but phoenix that causes the data inconsistency between hbase and solr.

If you are interested with this problem, please follow the issue we just open on hbase-indexer github:
https://github.com/NGDATA/hbase-indexer/issues/78

We finllay configure the indexer read mode to be RowReadMode.NEVER to solve this problem.

2016-06-02 15:12 GMT+08:00 ChengLei(营销事业部_技术部) <chenglei@sogou-inc.com>:
 
I think the problem is psql actually contruct upsert sql to import csv data  into hbase, so the regionserver ‘s indexer(org.apache.phoenix.hbase.index.Indexer) coprocessor will encode index data into WAL.
just as this: http://www.slideshare.net/jesse_yates/phoenix-secondary-indexing-la-hug-sept-9th-2013