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

Updated on 2022-11-18 GMT+08:00

View PDF

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

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)
);

Create a global index.
CREATE INDEX TEST_TABLE_INDEX ON TEST_TABLE(ACCOUNT1,DATE,NUM,ACCOUNTDES,SEQ_NUM);
Insert data.
UPSERT INTO TEST_TABLE (DATE,NUM,SEQ_NUM,ACCOUNT1,ACCOUNTDES,FLAG,SALL) values ('20201001',30201001,13,'367392332','sffa1','','');

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

367392332

sffa888

1231243

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

Delete the old index table.

DROP INDEX TEST_TABLE_INDEX ON TEST_TABLE;
Create an index table in asynchronous mode.

CREATE INDEX TEST_TABLE_INDEX ON TEST_TABLE(ACCOUNT1,DATE,NUM,ACCOUNTDES,SEQ_NUM) ASYNC;
Recreate a index.

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