How Do I Deal with the Restrictions of the Phoenix BulkLoad Tool?

Question

When the indexed field data is updated, if a batch of data exists in the user table, the BulkLoad tool cannot update the global and partial mutable indexes.

Answer

Problem Analysis

1. Create a table.

   CREATE TABLE TEST_TABLE(
   DATE varchar not null,
   NUM integer not null,
   SEQ_NUM integer not null,
   ACCOUNT1 varchar not null,
   ACCOUNTDES varchar,
   FLAG varchar,
   SALL double,
   CONSTRAINT PK PRIMARY KEY (DATE,NUM,SEQ_NUM,ACCOUNT1)
   );

2. Create a global index.

   CREATE INDEX TEST_TABLE_INDEX ON TEST_TABLE(ACCOUNT1,DATE,NUM,ACCOUNTDES,SEQ_NUM);

3. Insert data.

   UPSERT INTO TEST_TABLE (DATE,NUM,SEQ_NUM,ACCOUNT1,ACCOUNTDES,FLAG,SALL) values ('20201001',30201001,13,'367392332','sffa1','','');

4. Execute the BulkLoad task to update data.
   hbase org.apache.phoenix.mapreduce.CsvBulkLoadTool -t TEST_TABLE -i /tmp/test.csv, where the content of test.csv is as follows:

   20201001

   30201001

   13

   367392332

   sffa888

   1231243

   23

5. Symptom: The existing index data cannot be directly updated. As a result, two pieces of index data exist.

   +------------+-----------+-----------+---------------+----------------+
   | :ACCOUNT1  |   :DATE   |   :NUM    | 0:ACCOUNTDES  | :SEQ_NUM  |
   +------------+-----------+-----------+---------------+----------------+
   | 367392332  | 20201001  | 30201001  | sffa1          |  13                    |
   | 367392332  | 20201001  | 30201001  | sffa888       | 13                    |
   +------------+-----------+-----------+---------------+----------------+

Solution

1. Delete the old index table.

   DROP INDEX TEST_TABLE_INDEX ON TEST_TABLE;

2. Create an index table in asynchronous mode.

   CREATE INDEX TEST_TABLE_INDEX ON TEST_TABLE(ACCOUNT1,DATE,NUM,ACCOUNTDES,SEQ_NUM) ASYNC;

3. Recreate a index.

   hbase org.apache.phoenix.mapreduce.index.IndexTool --data-table TEST_TABLE --index-table TEST_TABLE_INDEX --output-path /user/test_table