Status
Current state: [One of "Under Discussion", "Accepted", "Rejected"]
Discussion thread: here (<- link to https://mail-archives.apache.org/mod_mbox/flink-dev/)
JIRA: here (<- link to https://issues.apache.org/jira/browse/FLINK-XXXX)
Released: <Flink Version>
Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).
Motivation
In FLIP-218 & FLIP-305, Flink support CREATE TABLE AS SELECT
statement which allows users to create new tables based on existing tables or query result. It's convenient for data analysts and data scientists to manage their data. However, Flink does not currently support the REPLACE TABLE AS SELECT
statment which enables users to replace an existing table with new data. With REPLACE TABLE AS SELECT
, they won't need to drop the table firstly, and use CREATE TABLE AS SELECT
then. Only one single REPLACE TABLE AS SELCT
statement can meet their needs.
So, this FLIP is aimed to support REPLACE TABLE AS SELECT
statement in Flink.
Note: this FLIP is much similiar to FLIP-218 & FLIP-305, you may need to read these two FLIP to get more context.
Public Interfaces
Syntax
We propose add the following syntax for REPLACE TABLE AS SELECT
statement:
REPLACE TABLE table_identifier [( [ < physical_column_definition >, ... ] [ < table_constraint >, ... ])] [ COMMENT table_comment ] [ PARTITIONED BY ( col_name1, col_name3, ...) ] [ WITH ( key1=val1, key2=val2, ... ) ] AS <table subquery> <physical_column_definition>: column_name column_type [ <column_constraint> ] [COMMENT column_comment] <column_constraint>: [CONSTRAINT constraint_name] PRIMARY KEY NOT ENFORCED <table_constraint>: [CONSTRAINT constraint_name] PRIMARY KEY (column_name, ...) NOT ENFORCED
Note: The column definition should be physical column definition, otherwise, it should throw exception.
Also, we would like to propose to CREATE OR REPLACE TABLE AS
to wrap CREATE TABLE AS SELECT
and REPLACE TABLE AS SELECT
which will create table if the table to be replaced doesn't exist.
CREATE OR REPLACE TABLE table_identifier [( [ < physical_column_definition >, ... ] [ < table_constraint >, ... ])] [ COMMENT table_comment ] [ PARTITIONED BY ( col_name1, col_name3, ...) ] [ WITH ( key1=val1, key2=val2, ... ) ] AS <table subquery> <physical_column_definition>: column_name column_type [ <column_constraint> ] [COMMENT column_comment] <column_constraint>: [CONSTRAINT constraint_name] PRIMARY KEY NOT ENFORCED <table_constraint>: [CONSTRAINT constraint_name] PRIMARY KEY (column_name, ...) NOT ENFORCED
Public Interfaces Change
To support atomic, we propose to add the following part to the interface SupportsStaging
proposed in FLIP-305
/** * Enables different staged operations to ensure atomicity in a {@link DynamicTableSink}. * * <p>By default, if this interface is not implemented, indicating that atomic operations are not * supported, then a non-atomic implementation is used. */ @PublicEvolving public interface SupportsStaging { //.... emit the parts proposed in FLIP-305 enum StagingPurpose { CREATE_TABLE_AS, CREATE_TABLE_AS_IF_NOT_EXISTS, // the following is what to add in this FLIP REPLACE_TABLE_AS, CREATE_OR_REPLACE_TABLE_AS } }
Also, we propose to modify the name of the option "table.ctas.atomicity-enabled" proposed in FLIP-305:
@PublicEvolving public class TableConfigOptions { @Documentation.TableOption(execMode = Documentation.ExecMode.BATCH_STREAMING) public static final ConfigOption<Boolean> TABLE_RTAS_CTAS_ATOMICITY_ENABLED = key("table.rtas-ctas.atomicity-enabled") .booleanType() .defaultValue(false) .withDescription( "Specifies if the create table/replace table/create or replace table as select operation is executed atomically. " + "By default, the operation is non-atomic. The target table is created/replaced in Client side, and it won't rollback even though the job fails or is cancelled. " + "If set this option to true and DynamicTableSink implements the SupportsStaging interface, the create table/replace table/create or replace table as select operation is expected to be executed atomically, " + "the behavior of which depends on the actual DynamicTableSink."); }
Proposed Changes
For REPLACE TABLE AS
statement:
1: Construct the table to be created
If the column definition is specificed, it will use the defined column to create the new table. Otherwise, it'll retrieve the column definition from the sub-query.
2: Check the table exists or not. If the table doesn't exist, throw TableException(String.format("The table %s to be replaced doesn't exist. You may want to use CREATE TABLE AS statement or CREATE OR REPLACE TABLE AS statement.",tableIdentifier)).
3:Check the atomicity is enabled, it requires both the options table.rtas-ctas.atomicity-enabled is set to false and the corresponding table sink implementement
SupportsStaging.
a: if atomic is enabled, it expects the atomicity to be guaranteed by external connector implementation. The Flink will generate an insert job according to the table subquery in REPLACE TABLE AS
statment, and call method StagedTable#begain
before the insert job start, call method StagedTable#commit
after the job finish, call method StagedTable#abort
if the job fail or canceled.
b: if not, then the atomicity can not be guaranted. Flink will do the operations for Replace Table one by one without atomicity guarantee. More exactly, it will drop the old table, create the new table, insert data into the new tables.
For CREATE OR REPLACE TABLE AS
stament, when the table exists, it'll consider it as REPLACE TABLE AS
statement. Otherwise, it'll consider it as CREATE TABLE AS
statement.
Note: Again, the propose changes much depend on FLIP-305. For more detail, please see the proposed chagne part in FLIP-305.
Compatibility, Deprecation, and Migration Plan
No any compatibility problem.
Test Plan
UT & IT
Rejected Alternatives
If there are alternative ways of accomplishing the same thing, what were they? The purpose of this section is to motivate why the design is the way it is and not some other way.