Grid for Jobs that Run in Remote Mode

Master service process

When you enable a Data Integration Service that runs on a grid, one service process starts on each node with the service role in the grid. The Data Integration Service designates one service process as the master service process. The master service process manages application deployments, logging, job requests, and the dispatch of mappings to worker service processes for optimization and compilation. The master service process also acts as a worker service process and can optimize and compile mappings.

Worker service processes

The Data Integration Service designates the remaining service processes as worker service processes. When a worker service process starts, it registers itself with the master service process so that the master is aware of the worker. A worker service process optimizes and compiles mappings, and then generates a grid task. A grid task is a job request sent by the worker service process to the Service Manager on the master compute node.

Service Manager on the master compute node

When you enable a Data Integration Service that runs on a grid, the Data Integration Service designates one node with the compute role as the master compute node.

The Service Manager on the master compute node performs the following functions to determine the optimal worker compute node to run the mapping:

• - Communicates with the Resource Manager Service to manage the grid of available compute nodes. When the Service Manager on a node with the compute role starts, the Service Manager registers the node with the Resource Manager Service.
• - Orchestrates worker service process requests and dispatches mappings to worker compute nodes.

The master compute node also acts as a worker compute node and can run mappings.

DTM processes on worker compute nodes

The Data Integration Service designates the remaining nodes with the compute role as worker compute nodes. The Service Manager on a worker compute node runs mappings in separate DTM processes started within containers.

Supported Node Roles

Service role

A Data Integration Service process runs on each node with the service role. Service components within the Data Integration Service process run workflows and profiles, and perform mapping optimization and compilation.

Compute role

DTM processes run on each node with the compute role. The DTM processes run deployed mappings, mappings run by Mapping tasks within a workflow, and mappings converted from a profile.

Both service and compute roles

A Data Integration Service process and DTM processes run on each node with both the service and compute roles. At least one node with both service and compute roles is required to run ad hoc jobs, with the exception of profiles. Ad hoc jobs include mappings run from the Developer tool or previews, scorecards, or drill downs on profile results run from the Developer tool or Analyst tool. The Data Integration Service runs these job types in separate DTM processes on the local node.

In addition, nodes with both roles can complete all of the tasks that a node with the service role only or a node with the compute role only can complete. For example, a workflow can run on a node with the service role only or on a node with both the service and compute roles. A deployed mapping can run on a node with the compute role only or on a node with both the service and compute roles.

Job Types

Workflows

When you run a workflow instance, the master service process runs the workflow instance and non-mapping tasks. The master service process uses round robin to dispatch each mapping within a Mapping task to a worker service process. The LDTM component of the worker service process optimizes and compiles the mapping. The worker service process then communicates with the master compute node to dispatch the compiled mapping to a separate DTM process running on a worker compute node.

Deployed mappings

When you run a deployed mapping, the master service process uses round robin to dispatch each mapping to a worker service process. The LDTM component of the worker service process optimizes and compiles the mapping. The worker service process then communicates with the master compute node to dispatch the compiled mapping to a separate DTM process running on a worker compute node.

Profiles

When you run a profile, the master service process converts the profiling job into multiple mapping jobs based on the advanced profiling properties of the Data Integration Service. The master service process then distributes the mappings across the worker service processes. The LDTM component of the worker service process optimizes and compiles the mapping. The worker service process then communicates with the master compute node to dispatch the compiled mapping to a separate DTM process running on a worker compute node.

Ad hoc jobs, with the exception of profiles

When you run an ad hoc job, with the exception of profiles, the Data Integration Service uses round robin to dispatch the first request directly to a worker service process that runs on a node with both the service and compute roles. The worker service process runs the job in a separate DTM process on the local node. To ensure faster throughput, the Data Integration Service bypasses the master service process. When you run additional ad hoc jobs from the same login, the Data Integration Service dispatches the requests to the same worker service process.

Example Grid that Runs Jobs in Remote Mode

• •On Node1, the master service process runs the workflow instance and non-mapping tasks. The master service process dispatches a mapping included in a Mapping task from workflow1 to the worker service process on Node2. The master service process also acts as a worker service process and can optimize and compile mappings. Profile jobs can also run on Node1.
• •On Node2, the worker service process optimizes and compiles the mapping. The worker service process then communicates with the master compute node on Node3 to dispatch the compiled mapping to a worker compute node. The Data Integration Service dispatches a preview request directly to the worker service process on Node2. The service process creates a DTM instance within a separate DTM process on Node2 to run the preview job. Node2 also serves as a worker compute node and can run compiled mappings.
• •On Node3, the Service Manager on the master compute node orchestrates requests to run mappings. The master compute node also acts as a worker compute node and runs the mapping from workflow1 in a separate DTM process started within a container.

Rules and Guidelines for Grids that Run Jobs in Remote Mode

• •The grid must contain at least one node with both the service and compute roles to run an ad hoc job, with the exception of profiles. The Data Integration Service runs these job types in a separate DTM process on the local node. Add additional nodes with both the service and compute roles so that these job types can be distributed to service processes running on other nodes in the grid.
• •To support failover for the Data Integration Service, the grid must contain at least two nodes that have the service role.
• •If you associate a Content Management Service with the Data Integration Service to run mappings that read reference data, each node in the grid must have both the service and compute roles.
• •The grid cannot include two nodes that are defined on the same host machine.
• •Informatica does not recommend assigning multiple Data Integration Services to the same grid nor assigning one node to multiple Data Integration Service grids.
If a worker compute node is shared across multiple grids, mappings dispatched to the node might fail due to an over allocation of the node's resources. If a master compute node is shared across multiple grids, the log events for the master compute node are also shared and might become difficult to troubleshoot.

Recycle the Service When Jobs Run in Remote Mode

• •Override compute node attributes for a node assigned to the grid.
• •Add or remove a node from the grid.
• •Shut down or restart a node assigned to the grid.

Configuring a Grid that Runs Jobs in Remote Mode

1. 1. Update the roles for the nodes in the grid.
2. 2. Create a grid for mappings, profiles, and workflows that run in separate remote processes.
3. 3. Assign the Data Integration Service to the grid.
4. 4. Configure the Data Integration Service to run jobs in separate remote processes.
5. 5. Enable the Resource Manager Service.
6. 6. Configure a shared log directory.
7. 7. Optionally, configure properties for each Data Integration Service process that runs on a node with the service role.
8. 8. Optionally, configure compute properties for each DTM instance that can run on a node with the compute role.
9. 9. Recycle the Data Integration Service.

Step 1. Update Node Roles

1. In the Administrator tool, click the Manage tab > Services and Nodes view.
2. In the Domain Navigator, select a node that you plan to add to the grid.
3. In the Properties view, click Edit for the general properties.
The Edit General Properties dialog box appears.
4. Select or clear the service and compute roles to update the node role.
5. Click OK.
6. If you disabled the compute role, the Disable Compute Role dialog box appears. Perform the following steps:
1. a. Select one of the following modes to disable the compute role:
• ▪ Complete. Allows jobs to run to completion before disabling the role.
• ▪ Stop. Stops all jobs and then disables the role.
• ▪ Abort. Tries to stop all jobs before aborting them and disabling the role.
2. b. Click OK.
7. Repeat the steps to update the node role for each node that you plan to add to the grid.

Step 2. Create a Grid

• •Any number of nodes with the service role only.
• •Any number of nodes with the compute role only.
• •At least one node with both the service and compute roles to run previews and to run ad hoc jobs, with the exception of profiles.

1. In the Administrator tool, click the Manage tab.
2. Click the Services and Nodes view.
3. In the Domain Navigator, select the domain.

4. On the Navigator Actions menu, click New > Grid.
The Create Grid dialog box appears.
5. Enter the following properties:

Property	Description
Name	Name of the grid. The name is not case sensitive and must be unique within the domain. It cannot exceed 128 characters or begin with @. It also cannot contain spaces or the following special characters: ` ~ % ^ * + = { } \ ; : ' " / ? . , < > | ! ( ) ] [
Description	Description of the grid. The description cannot exceed 765 characters.
Nodes	Select nodes to assign to the grid.
Path	Location in the Navigator, such as: DomainName/ProductionGrids

6. Click OK.

Step 3. Assign the Data Integration Service to the Grid

1. On the Services and Nodes view, select the Data Integration Service in the Domain Navigator.
2. Select the Properties tab.
3. In the General Properties section, click Edit.
The Edit General Properties dialog box appears.
4. Next to Assign, select Grid.
5. Select the grid to assign to the Data Integration Service.
6. Click OK.

Step 4. Run Jobs in Separate Remote Processes

1. On the Services and Nodes view, select the Data Integration Service in the Domain Navigator.
2. Select the Properties tab.
3. In the Execution Options section, click Edit.
The Edit Execution Options dialog box appears.
4. For the Launch Job Options property, select In separate remote processes.
5. Click OK.

Step 5. Enable the Resource Manager Service

1. On the Services and Nodes view, expand the System_Services folder.
2. Select the Resource Manager Service in the Domain Navigator, and click Recycle the Service.

Step 6. Configure a Shared Log Directory

1. On the Services and Nodes view, select the Data Integration Service in the Domain Navigator.
2. Select the Processes tab.
3. Select a node to configure the shared log directory for that node.
4. In the Logging Options section, click Edit.
The Edit Logging Options dialog box appears.
5. Enter the location to the shared log directory.
6. Click OK.
7. Repeat the steps for each node listed in the Processes tab to configure each service process with identical absolute paths to the shared directories.

Related Topics:
• Log Directory

Step 7. Optionally Configure Process Properties

Related Topics:
• Data Integration Service Process Properties

Step 8. Optionally Configure Compute Properties

Related Topics:
• Data Integration Service Compute Properties

Step 9. Recycle the Data Integration Service

Logs for Jobs that Run in Remote Mode

LDTM messages written by the worker service process on the service node

The first section of the mapping log contains LDTM messages about mapping optimization and compilation and about generating the grid task written by the worker service process on the service node.

The grid task messages include the following message that indicates the location of the log file written by the DTM process on the worker compute node:

INFO: [GCL_5] The grid task [gtid-1443479776986-1-79777626-99] cluster logs can be found at [./1443479776986/taskletlogs/gtid-1443479776986-1-79777626-99].

The listed directory is a subdirectory of the following default log directory configured for the worker compute node:

<Informatica installation directory>/logs/<node name>/dtmLogs/

DTM messages written by the DTM process on the compute node

The second section of the mapping log contains messages about mapping execution written by the DTM process on the worker compute node.

The DTM section of the log begins with the following lines which indicate the name of the worker compute node that ran the mapping:

###
### <MyWorkerComputeNodeName>
###

### Start Grid Task [gtid-1443479776986-1-79777626-99] Segment [s0] Tasklet [t-0] Attempt [1]

The DTM section of the log concludes with the following line:

### End Grid Task [gtid-1443479776986-1-79777626-99] Segment [s0] Tasklet [t-0] Attempt [1]

Override Compute Node Attributes to Increase Concurrent Jobs

• •You run long-running jobs on the grid.
• •The Data Integration Service cannot reuse DTM processes because you run jobs from different deployed applications.
• •Job concurrency is more important than the job execution time.