Using work items

What is a Work Item?

Steps in the same process can share data during their execution, thanks to the concept of a Work Item. Each step execution receives work items as inputs from the previous step and creates work items as output to the next one. During the execution, each step can freely read and update the data contained in an item. It's also possible to trigger processes with custom work items via an API call, and the contents can be used in the first step of the process.

A work item contains a payload of JSON-based data and optionally attached files.

Step inputs and outputs

There are two main ways to approach building processes with work items:

1. One work item for each process run is passed between steps and optionally modified at each stage.
2. Each step of the process creates output items, with the option of creating multiple items at once.

The first option is the current default and is easier for composing simple sequences of steps. Every trigger of the process ensures all steps run one after the other if no errors are encountered. The second option is more powerful and enables different patterns such as producer-consumer.

One example use-case of the latter would be a step that receives a large Excel file of orders and splits it into work items per row (the producer). The second step would be to process each work item individually (the consumer). With this pattern, the producer can wait to handle incoming requests, and the consumer can run in parallel and potentially process a large volume of data.

Another advantage of splitting work into individual items is the separation of concern. If one row of the input data is wrong, it can be manually fixed and re-run without affecting all other rows.

Example: Web store order processor with multiple work items

Dependencies

The primary way to interact with work items is through the RPA.Robocorp.WorkItems library. We start by importing it in our .robot file:

*** Settings ***
Library           RPA.Robocorp.WorkItems

Make sure you have an up-to-date version of the rpaframework library in your conda.yaml file:

channels:
  - conda-forge
dependencies:
  - python=3.7.5
  - pip=20.1
  - pip:
      - rpaframework==11.4.0 # 11.4.0 or newer supports work items including looping.

By default, the RPA.Robocorp.WorkItems library uses Control Room to store the work items and pass them between steps. It also supports local work items for easier development. More on that later.

The producer: Creating output work items

In the example linked above, the first task defined in producer.robot looks like this:

*** Tasks ***
Split orders file
    [Documentation]    Read orders file from input item and split into outputs
    Get work item file    ${ORDER_FILE_NAME}
    Open workbook    ${ORDER_FILE_NAME}
    ${table}=    Read worksheet as table    header=True
    ${groups}=    Group table by column    ${table}    Name
    FOR    ${products}    IN    @{groups}
        Create Output Work Item
        ${rows}=    Export table    ${products}
        @{items}=    Create List
        FOR    ${row}    IN    @{rows}
            ${name}=    Set Variable     ${row}[Name]
            ${zip}=    Set Variable     ${row}[Zip]
            Append To List    ${items}   ${row}[Item]
        END
        Set work item variable    Name    ${name}
        Set work item variable    Zip    ${zip}
        Set work item variable    Items    ${items}
        Save Work Item
    END

The sections we are interested in, for now, are the FOR loops. For every row in the table, we call the following keywords:

• Create output work item creates a new empty work item.
• Set work item variable sets a new variable name and value into the work item.
• Save work item saves the item and all contained data/files to the next step.

A keen reader might have seen an example of reading files in work items in the first task! We used the Get work item file keyword to receive the initial Excel file of our orders. How the first step receives this item is described in the following sections.

All changes to work items are stored locally until the keyword Save work item is called.

The consumer: Reading work item data

In the consumer.robot, we want to log in to the storefront, but we want only to do that once as it can be the slowest part of the process. After login, we want to loop through the incoming work items making orders for each:

*** Tasks ***
Load and Process All Orders
    [Documentation]    Order all products in input item queue
    Open Swag Labs
    Wait until keyword succeeds    3x    1s    Login
    For Each Input Work Item    Load and Process Order
    [Teardown]    Close browser

The For each input work item keyword loops through the work items in the input queue, passing each to the Load and Process Order keyword.

*** Keywords ***
Load and Process Order
    [Documentation]    Order all products in one work item products list
    ${payload}=    Get Work Item Payload
    ${name}=    Set Variable    ${payload}[Name]
    ${zip}=    Set Variable    ${payload}[Zip]
    ${items}=    Set Variable    ${payload}[Items]
    ${passed}   Run Keyword And Return Status    Process order    ${name}    ${zip}    ${items}
    IF     ${passed}
        Release input work item    DONE
    ELSE
        Log    Order prosessing failed for: ${name} zip: ${zip} items: ${items}   level=ERROR
        Release input work item    FAILED
    END

The Load and Process Order keyword handles incoming work items one by one. It uses the Get Work Item Payload keyword to get the input work item data and passes them to the Process order keyword. Process order contains the logic that adds the Items to the shopping cart and then performs checkout using Name and Zip.

If the Process order keyword fails:

• we log failure details, which end up in log.html and are visible in the execution output
• we call Release input work item FAILED, which marks that specific work item as failed.

In the example, one of the items cannot be found in the storefront:

• When running in Control Room, the item will be shown as failed.
• In Control Room, the failed work items can be checked, fixed, re-run, or marked as completed, so you only have to handle the failed items.

If the Process order keyword would generate more output work items for other process steps, those output work items would be released for processing when Release input work item DONE is called, signaling a successfully handled input work item.

Configuring Control Room

To run the full process in Control Room, we create a new process under the Workforce section. The process will have the two steps described earlier:

Because we want to control how output items are created, we enable the Explicit Output option:

Without this option, the initial work item that triggered the process is automatically copied from one step to the next.

Another feature of Control Room that we can use with our current process is concurrency, the ability to run multiple work items in parallel in the same step. Concurrency can be changed under the Advanced section of a process step:

In this example, we use a concurrency of 5. If there are more than our set limit of work items, they will wait until previous items have been processed.

Triggering the process

This robot expects to receive an Excel file of a certain format, named orders.xlsx. There is an example of such input in the root folder of the robot.

You can accomplish this in two ways:

• Triggering a process manually with a custom work item, using the Control Room UI.
• Triggering a process via a Control Room API call, passing the work item content.

Manipulating the whole item payload

As we have seen, the Get Work Item Variable and Set Work Item Variable keywords allow you to manipulate the values in your work item conveniently.

There are cases, for example, when receiving a webhook from an external system, when you might want or have to structure the data differently. In these cases, you can take advantage of the Get Work Item Payload and Set Work Item Payload keywords to work with the whole item payload.

Assuming you have a work item structured like this:

[
  {
    "payload": ["Hello", "World"]
  }
]

Note how the item payload is no longer a map of key-value pairs.

Because you do not have a typical dictionary as the payload, the Get Work Item Variable, Set Work Item Variable, and Get Work Item Variables keywords are not helpful in this case. Instead, you will need to access the full payload item and then target the data elements you need by directly manipulating the data.

An example of manipulating raw payloads:

*** Settings ***
Documentation     A simple robot to demonstrate manipulating the item payload directly.
Library           RPA.Robocorp.WorkItems
Library           Collections

*** Tasks ***
Demonstrate how to read and write the full payload in the work item
    ${payload}=    Get Work Item Payload
    FOR    ${value}    IN    @{payload}
        Log    Received value: ${value}
    END

    Append to list    ${payload}    And    Friends
    Set Work Item Payload    ${payload}
    Save Work Item

• In the *** Settings *** section, we have added the standard Robot Framework Collections library, which allows us to manipulate data structures like lists and dictionaries.
• We are using the Get Work Item Payload keyword to get the whole payload into a variable that we call ${payload}.
• We can loop over the raw payload using a FOR loop because it is a list instead of a typical dictionary.
• To add new values to the payload, we can use the Append to list keyword that we got from the Collections library.
• We can then set the work item payload using the Set Work Item Payload, and finally make the change permanent using the Save Work Item keyword.

Using this method, you can work on payloads with arbitrary structures at the price of adding a bit of complexity: for this reason, if you can create your payload using the default structure, we recommend you do so.

Working with files in the work item

In addition to structured data, the work item can also be used to store and share files. The RPA.Robocorp.WorkItems library provides keywords to add, retrieve, list, and delete work item files. Like all other data stored in the work item, the files will be available to all robot steps added to the same process. Files added to the work item are stored by Control Room and can be inspected using its user interface.

Control Room enforces a size limit of 50 MB per file added to the work item.

Developing with work items locally

When executing a robot in a cloud environment like Control Room, the RPA.Robocorp.WorkItems library will store the work item in the cloud environment, sharing its contents between steps defined in the same process, without any configuration needed.

When developing our robot and running it locally, we want to simulate the Control Room behavior as much as possible using local files.

The robot in the example has detailed guides on using work items in local testing with different tools, so we highly recommend checking the description of that robot.

In your robot folder, the local testing focuses in the devdata folder.

By convention, you should have two folders in the devdata folder: work-items-in and work-items-out. This helps you store test input data in version control but leave the dynamically changing output out.

By convention, each folder under devdata/work-items-in is considered a "test set".

• work-items.json is a reserved file, and it contains one or more work item definitions.
• We recommend placing files linked to these work items in the same folder with the work-items.json file.

Inside work-items.json, the defined structure is a list of work items, each of which can contain payload and/or files sections.

• The content of the payload section is JSON.
• The content of the files section expects a key-value list where the key is the name used in the Robot implementation, and the value is the relative path to the file.
• This is the same as the work items you see in Control Room.

Example work items

The best way to get started with work items is to start with the example robot and test it out, both in local development and Control Room.

Work item with only a file

{
  "payload": {},
  "files": {
    "orders.xlsx": "orders.xlsx"
  }
}

In Control Room, the work item looks like this:

Work item with payload

[
  {
    "payload": {
      "Name": "Zoya Roche",
      "Zip": 3013,
      "Items": ["Sauce Labs Bolt T-Shirt", "Sauce Labs Fleece Jacket", "Sauce Labs Onesie"]
    },
    "files": {}
  }
]

In Control Room, the work item looks like this: