Page History

...

Terminology

Configuration : It refers to a class  in sqoop annotated with @ConfigurationClass and represents a grouping of related configs that the corresponding configurable that owns this configuration class exposes

Here is the code example. 

@ConfigurationClass
public class FromJobConfiguration {
  @Config public FromJobConfig fromJobConfig;
  @Config public IncrementalExtractConfig incrementalExtractConfig;
  public FromJobConfiguration() {
    fromJobConfig = new FromJobConfig();
    incrementalExtractConfig = new IncrementalExtractConfig();
  }
}

Config or ConfigInputs  : They both mean the same in this wiki and are used interchangeably, i.e they represent a class in sqoop that are annotated with the @ConfigClass annotation and used in the @ConfigurationClass annotated class with @Config annotation

@ConfigClass
public class IncrementalExtractConfig {
  @Input(size = 50, editable = InputEditable.USER_ONLY)
  public String key;
  // Optional tells the type of the key, if it s long, int, date, timestamp etc
  @Input
  public String dataType;
  @Input(size = 50, editable = InputEditable.USER_ONLY, overrides = "lastReadValue")
  public String value;
  @Input(editable = InputEditable.CONNECTOR_ONLY)
  public long lastReadValue;
}

Inputs  : They are represented by the @Input annotation on a variable inside a class annotated with @ConfigClass. 

  @Input(size = 50, editable = InputEditable.USER_ONLY, overrides = "lastReadValue")
  public String value;
  @Input(editable = InputEditable.CONNECTOR_ONLY)
  public long lastReadValue;

Requirements

...

• Configs should be treated as a top level entity as they become more sophisticated with the SQOOP-1804 and in future with the implementation of SQOOP-1643. Currently editing and reading config/ config inputs can only happen via a job or link. This proposal aims at making the config objects as first class citizens, so they can be read and edited by their unique name.
• Config objects per MConfigType (i,e JOB and LINK)  are lists. So ability to edit per "CONFIG" name  is easier than having to edit per "CONFIG LIST" associated with the type. Users do see the names when they list the configs per connector, we can also have a shell command/ rest-API to list all the configs per configurable to make it easier for users to reference configs by name
• Read  the Read and Update the Config Inputs by Type/SubType and By Job /Submission ( since SQOOP-2025 we may be able to have configs by submissionId)Support
• this in both shell command and Rest-API.Update the Config Inputs by Type/SubType for the latest/last submission in the job. We should not allow editing previous submissions and it should be read only
• Only the "inputs" with attribute "USER-ONLY" or "ANY" as per SQOOP-1804 will be editable. Make sure to validate the condition of editable and adhere to cascading changes depending on the overrides attribute on each input.
• Once the input values are edited, the new values will be used in the next job run, unless we maintain history as per SQOOP-2025.

NOTE : Support all the above in both shell command and Rest-API.

 

Non Goals

• Supporting CD ( create / delete ) of Config Inputs via the REST or command line. It is only allowed via the configurable code and supported annotations on the classes today and it should remain so.
• Editing submission history 

Design and Implementation Details

Shell Commands

A new Enum will be added to list the config types - 

 from, to, link, driver

Read Config By Type and Job or Submission 

Here are some details I considered when coming up with the subType and why it made sense.

 

At one point, I thought why not have direction as a parameter for type, JOB, but direction is not relevant to all configurables. i,e if for the driver configs, "direction" has no meaning. Similarly for the type "LINK" there is no concept of direction.

Hence I went with the subType, where subType is a second level hierarchy for distinguishing the types of configs that are supported in sqoop

Alternatives are possible, but we have to bear in mind that config/ config inputs are not associated with jobs and links,  They are associated with connectors/driver ( i,e configurables )

The config input values are associated with jobs and links rather, so when reading/ editing the config input values, either we can use JOB/ LINK entities  or rely more on the "CONFIGTYPE" enum, if we want to treat config as a first class citizen

SubType for MConfigType

SubTypes have been added to the MConfigType Enum to indicate the sub types

/**
 * Represents the various config types supported by the system.
 */
@InterfaceAudience.Private
@InterfaceStability.Unstable
public enum MConfigType {
  /** Unknown config type */
  OTHER,
  @Deprecated
  // NOTE: only exists to support the connector data upgrade path
  CONNECTION,
  /** link config type */
  LINK("link"),
  /** Job config type */
  JOB("from", "to", "driver");
  
  private List<String> subType;
  
  MConfigType(String... aliases){
    this.subType = Arrays.asList(aliases);
  }
  
  List<String> getSubTypes(MConfigType type) {
    return type.subType;
  } 
}

Shell Commands

NOTE: all the job config inputs values are for the last job run only since we do not store the config values for each submission yet

// Supported type values ( from, to, link, driver )



1. show config foo --type JOB --subType from --jid 1 - will provide the config value for the last job run
//2id 1 or show config 1  --type JOB --subType from --id 1 ( NOTE: we can use either id or name )
  //*. show config "foo" --type JOB --subType "from" --sid 1 -( SINCE we are not doing SQOOP-2025, this will not be in the current patch.) 
 
2. show config foo --type JOB --subType to --jidid 11 
 
3. show config foo --type JOB --subType driver --jidid 1
 
4. show config foo --type LINK --subType link --lidid 1   
 
//  planned, we need to addSQOOP-2046 as a namesub forticket the config paramof SQOOP-1516
5. show input --namefoo --config "toJobConfig"bar --type fromLINK --jidid 1 //note a config is a list
 
 id here refers to the link id    

Edit Config By Type and Job

NOTE:  Prev ( prev submissions cannot be edited, hence we restrict editing to the last job run only)

1. edit config foo --type JOB --subType from --jid 1 - will provide the config value for the last job run
// we can only edit the last job run values

...

id 1 // id here refers to the job id 
 
// planned, SQOOP-2046 as a sub ticket of SQOOP-1516
2. edit input foo --config bar --type LINK --id 1 // id here refers to the link id
 

ALTERNATIVELY, It was suggested that for command line it is NOT good not to introduce config as a top level ( It was a preference from Gwen Shapira and Qian Xu).  Hence the command are

edit job --jid 1 --config foo --type from 
show job --jid 1 --config foo --type from 
edit job --jid 1 --config foo --type to
edit job --jid 1 --config foo --type driver
 
edit link --lid 1 --config foo --type link 



 

Rest API changes

The resource returned from the APIs is a config object  MConfig. 

Read Config By Type and Job or Submission

1. GET
 v1/config?jId/job?name=?&Id=?&type=   
 
or   
 
2. GET v1/config?type=JOB&name=?&Id=?&subType=

Edit Config By Type and Job

1. POST
 v1/config/link?jIdname=?&id=?&type=  
 
 or 
 
2. POST v1/config?type=LINK&name=?&id=?&subType=

We chose #1 in both the cases

Repository API changes

Add new API to get config inputs by

...

jobId and type

...

, name

Add new API to

...

edit/post config inputs by jobId and type

...

, name

  public abstract MConfig findFromJobConfig(long jobId, String name);
  public abstract MConfig findToJobConfig(long jobId, String name);
  public abstract MConfig findDriverJobConfig(long jobId, String name);
  public abstract MConfig findLinkConfig(long linkId, String name);
// the type exists to check who is editing the configs, the connector code via upgrade or the user ( from rest / command line )
  public abstract void updateJobConfig(long jobId, MConfig config, MConfigUpdateEntity type);
  public abstract void updateLinkConfig(long linkId, MConfig config, MConfigUpdateEntity type));

...

Testing

Unit test are almost non existent for the shell code. Hence we will rely on the basic manual testing

Rest APIs can be tested via integration tests and that will be part of the work proposed