Automate Data Operations¶

CumulusCI offers a suite of tasks to help you to manage data as part of your project automation. Within your repository, you can define one or several datasets, collections of data you use for specific purposes. CumulusCI tasks support extracting defined datasets from scratch orgs or persistent orgs, storing those snapshots within the repository, and automating the load of datasets into orgs. Data operations are executed via the Bulk and REST APIs.

A dataset consists of

a definition file, written in YAML, which specifies the sObjects and fields contained in the dataset and the order in which they are loaded or extracted from an org.
a storage location, which may take the form of a SQL database (typically, a SQLite file stored within the repository, although external databases are supported) or a SQL script file.

Datasets are stored in the datasets/ folder within a repository by default. Projects created with a recent version of CumulusCI ship with this directory in place.

The Lifecycle of a Dataset¶

A dataset starts with a definition: which objects, and which fields, are to be captured, persisted, and loaded into orgs? (The details of definition file format are covered below).

With a definition available, the dataset may be captured from an org into the repository. A captured dataset may be stored under version control and incorporated into project automation, loaded as part of flows during org builds or at need. As the project’s needs evolve, datasets may be re-captured from orgs and versioned alongside the project metadata.

Projects may define one or many datasets. Datasets can contain an arbitrary amount of data.

Defining Datasets¶

A dataset is defined in YAML as a series of steps. Each step registers a specific sObject as part of the dataset, and defines the relevant fields on that sObject as well as its relationships to other sObjects that are included in the data set.

Note: this section discusses how to define a dataset and the format of the definition file. In many cases, it’s easier to use the generate_dataset_mapping task than to create this definition by hand. See below for more details.

A simple dataset definition looks like this:

Accounts:
    sf_object: Account
    fields:
        - Name
        - Description
        - RecordTypeId
    lookups:
        ParentId:
            table: Account
            after: Accounts
Contacts:
    sf_object: Contact
    fields:
        - FirstName
        - LastName
        - Email
    lookups:
        AccountId:
            table: Account

This example defines two steps: Accounts and Contacts. (The names of steps are arbitrary). Each step governs the extraction or load of records in the sObject denoted in its sf_object property.

Relationships are defined in the lookups section. Each key within lookups is the API name of the relationship field. Beneath, the table key defines the stored table to which this relationship refers.

CumulusCI loads steps in order. However, sObjects earlier in the sequence of steps may include lookups to sObjects loaded later, or to themselves. For these cases, the after key may be included in a lookup definition, with a value set to the name of the step after which the referenced record is expected to be available. CumulusCI will defer populating the lookup field until the referenced step has been completed. In the example above, an after definition is used to support the ParentId self-lookup on Account.

API Selection¶

By default, CumulusCI will determine the data volume of the specified object and select an API for you: for under 2,000 records, the REST Collections API is used; for more, the Bulk API is used. The Bulk API is also used for delete operations where the hard delete operation is requested, as this is available only in the Bulk API. Smart API selection helps increase speed for low- and moderate-volume data loads.

To prefer a specific API, set the api key within any mapping step; allowed values are "rest", "bulk", and "smart", the default.

CumulusCI defaults to using the Bulk API in Parallel mode. If required to avoid row locks, specify the key bulk_mode: Serial in each step requiring the use of serial mode.

For all API modes, you can specify a batch size using the batch_size key. Allowed values are between 1 and 200 for the REST API and 1 and 10,000 for the Bulk API.

Note that the semantics of batch sizes differ somewhat between the REST API and the Bulk API. In the REST API, the batch size is the size of upload batches and also the actual size of individual transactions. In the Bulk API, the batch size is the maximum record count in a Bulk API upload batch, which is subject to its own limits, including restrictions on total processing time. Bulk API batches are automatically chunked further into transactions by the platform, and the transaction size cannot be controlled.

Database Mapping¶

CumulusCI’s definition format includes considerable flexibility for use cases where datasets are stored in SQL databases whose structure is not identical to the Salesforce database. Salesforce objects may be assigned to arbitrary database tables, and Salesforce field names mapped to arbitrary columns.

For new mappings, it’s recommended to allow CumulusCI to use sensible defaults by specifying only the Salesforce entities. Legacy datasets are likely to include explicit database mappings, which would look like this for the same data model as above:

Accounts:
    sf_object: Account
    table: Account
    fields:
        Name: Name
        Description: Description
        RecordTypeId: RecordTypeId
    lookups:
        ParentId:
            table: Account
            after: Accounts
Contacts:
    sf_object: Contact
    table: Contact
    fields:
        FirstName: FirstName
        LastName: LastName
        Email: Email
    lookups:
        AccountId:
            table: Account

Note that in this version, fields are specified as a colon-separated mapping, not a list. Each pair in the field map is structured as Salesforce API Name: Database Column Name. Additionally, each object has a table key to specify the underlying database table.

New mappings that do not connect to an external SQL database (that is, mappings which simply extract and load data between Salesforce orgs) should not need to use this feature, and new mappings that are generated by CumulusCI use the simpler version shown above. Existing mappings may be converted to this streamlined style in most cases by loading the existing dataset, modifying the mapping file, and then extracting a fresh copy of the data. Note however that datasets which make use of older and deprecated CumulusCI features, such as the record_type key, may need to continue using explicit database mapping.

Record Types¶

CumulusCI supports automatic mapping of Record Types between orgs, keyed upon the Developer Name. To take advantage of this support, simply include the RecordTypeId field in any step. CumulusCI will transparently extract Record Type information during dataset capture and map Record Types by Developer Name into target orgs during loads.

Older dataset definitions may also use a record_type key:

Accounts:
    sf_object: Account
    fields:
        - Name
    record_type: Organization

This feature limits extraction to records possessing that specific Record Type, and assigns the same Record Type upon load.

It’s recommended that new datasets use Record Type mapping by including the RecordTypeId field. Using record_type will result in CumulusCI issuing a warning.

Relative Dates¶

CumulusCI supports maintaining relative dates, helping to keep the dataset relevant by ensuring that date and date-time fields are updated when loaded.

Relative dates are enabled by defining an anchor date, which is specified in each mapping step with the anchor_date key, whose value is a date in the format 2020-07-01.

When you specify a relative date, CumulusCI modifies all date and date-time fields on the object such that when loaded, they have the same relationship to today as they did to the anchor date. Hence, given a stored date of 2020-07-10 and an anchor date of 2020-07-01, if you perform a load on 2020-09-10, the date field will be rendered as 2020-09-19 - nine days ahead of today’s date, as it was nine days ahead of the anchor date.

Relative dates are also adjusted upon extract so that they remain stable. Extracting the same data mentioned above would result in CumulusCI adjusting the date back to 2020-07-10 for storage, keeping it relative to the anchor date.

Relative dating is applied to all date and date-time fields on any mapping step that contains the anchor_date clause. If orgs are configured to permit setting audit fields upon record creation and the appropriate user permission is enabled, CumulusCI can apply relative dating to audit fields, such as CreatedDate. For more about how to automate that setup, review the create_bulk_data_permission_set task below.

For example, this mapping step:

Contacts:
    sf_object: Contact
    fields:
        - FirstName
        - LastName
        - Birthdate
    anchor_date: 1990-07-01

would adjust the Birthdate field on both load and extract around the anchor date of July 1, 1990. Note that date and datetime fields not mapped, as well as fields on other steps, are unaffected.

Person Accounts¶

CumulusCI supports extracting and loading person account data. In your dataset definition, map person account fields like LastName, PersonBirthdate, or CustomContactField__pc to Account steps (i.e. where sf_object equals Account).

Account:
  sf_object: Account
  table: Account
  fields:
    # Business Account Fields
    - Name
    - AccountNumber
    - BillingStreet
    - BillingCity

    # Person Account Fields
    - FirstName
    - LastName
    - PersonEmail
    - CustomContactField__pc

    # Optional (though recommended) Record Type
    - RecordTypeId

Record Types¶

It’s recommended, though not required, to extract Account Record Types to support datasets with person accounts so there is consistency in the Account record types loaded. If Account RecordTypeId is not extracted, the default business account Record Type and default person account Record Type will be applied to business and person account records respectively.

Extract¶

During dataset extraction, if the org has person accounts enabled, the IsPersonAccount field is extracted for Account and Contact records so CumulusCI can properly load these records later. Additionally, Account.Name is not createable for person account Account records, so Account.Name is not extracted for person account Account records.

Load¶

Before loading, CumulusCI checks if the dataset contains any person account records (i.e. any Account or Contact records with IsPersonAccount as true). If the dataset does contain any person account records, CumulusCI validates the org has person accounts enabled.

You can enable person accounts for scratch orgs by including the PersonAccounts feature in your scratch org definition.

Advanced Features¶

CumulusCI supports two additional keys within each step

The filters key encompasses filters applied to the SQL data store when loading data. Use of filters can support use cases where only a subset of stored data should be loaded.

filters:
    - 'SQL string'

Note that filters uses SQL syntax, not SOQL. Filters do not perform filtration or data subsetting upon extraction; they only impact loading. This is an advanced feature.

The static key allows individual fields to be populated with a fixed, static value.

static:
    CustomCheckbox__c: True
    CustomDateField__c: 2019-01-01

The soql_filter key allows to specify a WHERE clause that should be used when extracting data from your Salesforce org:

Account:
sf_object: Account table: Account fields:

Name

Industry

Type

soql_filter: “Industry = ‘Higher Education’ OR Type = ‘Higher Education’”

Note that trying to load data that is extracted using soql_filter may cause “invalid cross reference id” errors if related object records are filtered on extract. Use this feature only if you fully understand how CumulusCI load data task resolves references to related records when loading data to a Salesforce org.

Primary Keys¶

CumulusCI offers two modes of managing Salesforce Ids and primary keys within the stored database.

If the fields list for an sObject contains a mapping:

Id: sf_id

CumulusCI will extract the Salesforce Id for each record and use that Id as the primary key in the stored database.

If no such mapping is provided, CumulusCI will remove the Salesforce Id from extracted data and replace it with an autoincrementing integer primary key.

Use of integer primary keys may help yield more readable text diffs when storing data in SQL script format. However, it comes at some performance penalty when extracting data. It’s recommended that most mappings do not map the Id field and allow CumulusCI to utilize the automatic primary key.

Handling Namespaces¶

All CumulusCI bulk data tasks support automatic namespace injection or removal. In other words, the same mapping file will work for namespaced and unnamespaced orgs, as well as orgs with the package installed managed or unmanaged. If a mapping element has no namespace prefix and adding the project’s namespace prefix is required to match a name in the org, CumulusCI will add one. Similarly, if removing a namespace is necessary, CumulusCI will do so.

In the extremely rare circumstance that an org contains the same mapped schema element in both namespaced and non-namespaced form, CumulusCI does not perform namespace injection or removal for that element.

Namespace injection can be deactivated by setting the inject_namespaces option to False.

The generate_dataset_mapping generates mapping files with no namespace and this is the most common pattern in CumulusCI projects.

Namespace Handing with Multiple Mapping Files¶

It’s also possible, and common in older managed package products, to use multiple mapping files to achieve loading the same data set in both namespaced and non-namespaced contexts. This is no longer recommended practice.

A mapping file that is converted to use explicit namespacing might look like this:

Original version:

Destinations:
    sf_object: Destination__c
    fields:
        Name: Name
        Target__c: Target__c
    lookups:
        Supplier__c:
            table: Supplier__c

Namespaced version:

Destinations:
    sf_object: MyNS__Destination__c
    table: Destination__c
    fields:
        MyNS__Name: Name
        MyNS__Target__c: Target__c
    lookups:
        MyNS__Supplier__c:
            key_field: Supplier__c
            table: Supplier__c

Note that each of the definition elements that refer to local storage remains un-namespaced, while those elements referring to the Salesforce schema acquire the namespace prefix.

For each lookup, an additional key_field declaration is required, whose value is the original storage location in local storage for that field’s data. In most cases, this is simply the version of the field name in the original definition file.

Adapting an originally-namespaced definition to load into a non-namespaced org follows the same pattern, but in reverse.

Note that mappings which use the flat list style of field specification must use mapping style to convert between namespaced and non-namespaced deployment.

It’s recommended that all new mappings use flat list field specifications and allow CumulusCI to manage namespace injection. This capability typically results in significant simplication in automation.

Optional Data Elements¶

Some projects need to build datasets that include optional data elements - fields and objects that are loaded into some of the project’s orgs, but not others. This can cover both optional managed packages and features that are included in some, but not all, orgs. For example, a managed package A that does not require another managed package B but is designed to work with it may wish to include data for managed package B in its data sets, but load that data if and only if B is installed. Likewise, a package might wish to include data supporting a particular org feature, but not load that data in an org where the feature is turned off (and its associated fields and objects are for that reason unavailable).

To support this use case, the load_dataset and extract_dataset tasks offer a drop_missing_schema option. When enabled, this option results in CumulusCI ignoring any mapped fields, sObjects, or lookups that correspond to schema that is not present in the org.

Projects that require this type of conditional behavior can build their datasets in an org that contains managed package B, capture it, and then load it safely in orgs that both do and do not contain B. However, it’s important to always capture from an org with B present, or B data will not be preserved in the dataset.

Custom Settings¶

Datasets don’t support Custom Settings. However, a separate task is supplied to deploy Custom Settings (both list and hierarchy) into an org: load_custom_settings. The data for this task is defined in a YAML text file

Each top-level YAML key should be the API name of a Custom Setting. List Custom Settings should contain a nested map of names to values. Hierarchy Custom settings should contain a list, each of which contains a data key and a location key. The location key may contain either profile: <profile name>, user: name: <username>, user: email: <email>, or org.

Example:

List__c:
    Test:
        MyField__c: 1
    Test 2:
        MyField__c: 2
Hierarchy__c:
    -
        location: org
        data:
            MyField__c: 1
    -
        location:
            user:
                name: test@example.com
        data:
            MyField__c: 2"""

CumulusCI will automatically resolve the location specified for Hierarchy Custom Settings to a SetupOwnerId. Any Custom Settings existing in the target org with the specified name (List) or setup owner (Hierarchy) will be updated with the given data.

Dataset Tasks¶

`create_bulk_data_permission_set`¶

Create and assign a Permission Set that enables key features used in Bulk Data tasks (Hard Delete and Set Audit Fields) for the current user. The Permission Set will be called CumulusCI Bulk Data.

Note that prior to running this task you must ensure that your org is configured to allow the use of Set Audit Fields. You can do so by manually updating the required setting in the User Interface section of Saleforce Setup, or by updating your scratch org configuration to include

"securitySettings": {
  "enableAuditFieldsInactiveOwner": true
}

For more information about the Set Audit Fields feature, review this Knowledge article.

After this task runs, you’ll be able to run the delete_data task with the hardDelete option, and you’ll be able to map audit fields like CreatedDate.

`extract_dataset`¶

Extract the data for a dataset from an org and persist it to disk.

Options¶

mapping: the path to the YAML definition file for this dataset.
sql_path: the path to a SQL script storage location for this dataset.
database_url: the URL for the database storage location for this dataset.

mapping and either sql_path or database_url must be supplied.

Example:

cci task run extract_dataset -o mapping datasets/qa/mapping.yml -o sql_path datasets/qa/data.sql --org qa

`load_dataset`¶

Load the data for a dataset into an org. If the storage is a database, persist new Salesforce Ids to storage.

Options¶

mapping: the path to the YAML definition file for this dataset.
sql_path: the path to a SQL script storage location for this dataset.
database_url: the URL for the database storage location for this dataset.
start_step: the name of the step to start the load with (skipping all prior steps).
ignore_row_errors: If True, allow the load to continue even if individual rows fail to load. By default, the load stops if any errors occur.

mapping and either sql_path or database_url must be supplied.

Example:

cci task run load_dataset -o mapping datasets/qa/mapping.yml -o sql_path datasets/qa/data.sql --org qa

`generate_dataset_mapping`¶

Inspect an org and generate a dataset definition for the schema found there.

This task is intended to streamline the process of creating a dataset definition. To use it, first build an org (scratch or persistent) containing all of the schema needed for the dataset.

Then, execute generate_dataset_mapping. The task inspects the target org and creates a dataset definition encompassing the project’s schema, attempting to be minimal in its inclusion outside that schema. Specifically, the definition will include:

Any custom object without a namespace
Any custom object with the project’s namespace
Any object with a custom field matching the same namespace criteria
Any object that’s the target of a master-detail relationship, or a custom lookup relationship, from another included object.

On those sObjects, the definition will include

Any custom field (including those defined by other packages)
Any required field
Any relationship field targeting another included object
The Id, FirstName, LastName, and Name fields, if present

Certain fields will always be omitted, including

Lookups to the User object
Binary-blob (base64) fields
Compound fields
Non-createable fields

The resulting definition file is intended to be a viable starting point for a project’s dataset. However, some additional editing is typically required to ensure the definition fully suits the project’s use case. In particular, any fields required on standard objects that aren’t automatically included must be added manually.

Reference Cycles¶

Dataset definition files must execute in a sequence, one sObject after another. However, Salesforce schemas often include reference cycles: situations in which Object A refers to Object B, which also refers to Object A, or in which Object A refers to itself.

CumulusCI will detect these reference cycles during mapping generation and ask the user for assistance resolving them into a linear sequence of load and extract operations. In most cases, selecting the schema’s most core object (often a standard object like Account) will successfully resolve reference cycles. CumulusCI will automatically tag affected relationship fields with after directives to ensure they’re populated after their target records become available.

Options¶

path: Location to write the mapping file. Default: datasets/mapping.yml
ignore: Object API names, or fields in Object.Field format, to ignore
namespace_prefix: The namespace prefix to treat as belonging to the project, if any

Example:

cci task run generate_dataset_mapping --org qa -o namespace_prefix my_ns

`load_custom_settings`¶

Load custom settings stored in YAML into an org.

Options¶

settings_path: Location of the YAML settings file.

`delete_data`¶

You can also delete records using CumulusCI. You can either delete every record of a particular object, certain records based on a where clause or every record of multiple objects. Because where clauses seldom make logical sense when applied to multiple objects, you cannot use a where clause when specifying multiple objects.

Details are available with cci org info delete_data and in the task reference.

Examples¶

cci task run delete_data -o objects Opportunity,Contact,Account --org qa

cci task run delete_data -o objects Opportunity -o where "StageName = 'Active' "

cci task run delete_data -o objects Account -o ignore_row_errors True

cci task run delete_data -o objects Account -o hardDelete True

Generate Fake Data¶

It is possible to use CumulusCI to generate arbitrary amounts of synthetic data using the snowfakery task. That task is built on the Snowfakery language. CumulusCI ships with Snowfakery embedded, so you do not need to install it.

To start, you will need a Snowfakery recipe. You can learn about writing them in the Snowfakery docs.

Once you have it, you can fill an org with data like this:

$ cci task run snowfakery --recipe datasets/some_snowfakery_recipe.yml

If you would like to execute the recipe multiple times to generate more data, you do so like this:

$ cci task run generate_and_load_from_yaml --run-until-recipe-repeated 400

Which will repeat the recipe 400 times.

There are two other ways to control how many times the recipe is repeated: –run-until-records-loaded and –run-until-records-in-org.

Generated Record Counts¶

Consider this example:

$ cci task run snowfakery --run-until-records-loaded 1000:Account

This would say to run the recipe until the task has loaded 1000 new Accounts. In the process, it might also load Contacts, Opportunities, custom objects oor whatever else is in the recipe. But it finishes when it has loaded 400 Accounts.

The counting works like this:

Snowfakery always executes a complete recipe. It never stops halfway through. If your recipe creates more records than you need, you might overshoot. Usually the amount of overshoot is just a few records, but it depends on the details of your recipe.

At the end of executing a recipe, it checks whether it has created enough of the object type mentioned by the –run-until-records-loaded parameter.

If so, it finishes. If not, it runs the recipe again.

So if your recipe creates 10 Accounts, 5 Contacts and 15 Opportunities, then when you run the command above it will run the recipe 100 times (100*10=1000) which will generate 1000 Accounts, 500 Contacts and 1500 Opportunities.

–run-until-records-in-org works similarly, but it determines how many times to run the recipe based on how many records are in the org at the start. For example, if the org already has 300 Accounts in it then:

$ cci task run snowfakery --run-until-records-in-org 1000:Account

Would be equivalent to –run-until-records-loaded 700:Account because one needs to add 700 Accounts to the 300 resdent ones to get to 1000.

Controlling the Loading Process¶

CumulusCI’s data loader has many knobs and switches that you might want to adjust during your load. It supports a “.load.yml” file format which allows you to manipulate these load settings. The simplest way to use this file format is to make a file in the same directory as your recipe with a filename that is derived from the recipe’s by replacing everything after the first “.” with “.load.yml”. For example, if your recipe is called “babka.recipe.yml” then your load file would be “babka.load.yml”.

Inside of that file you put a list of declarations in the following format:

- sf_object: Account
  api: bulk
  bulk_mode: parallel

Which would specifically load accounts using the bulk API’s parallel mode.

The specific keys that you can associate with an object are:

api: “smart”, “rest” or “bulk”
batch_size: a number
bulk_mode: “serial” or “parallel”
load_after: the name of another sobject to wait for before loading

“api”, “batch_size” and “bulk_mode” have the same meanings that they do in mapping.yml as described in API Selection.

For example, one could force Accounts and Opportunities to load after Contacts:

- sf_object: Account
  load_after: Contact

- sf_object: Opportunity
  load_after: Contact

If you wish to share a loading file between multiple recipes, you can refer to it with the --loading_rules option. That will override the default filename (<recipename>.load.yml). If you want both, or any combination of multiple files, you can do that by listing them with commas between the filenames.

Batch Sizes¶

You can also control batch sizes with the -o batch_size BATCHSIZE parameter. This is not the Salesforce bulk API batch size. No matter what batch size you select, CumulusCI will properly split your data into batches for the bulk API.

You need to understand the loading process to understand why you might want to set the batch_size.

If you haven’t set the batch_size then Snowfakery generates all of the records for your load job at once.

So the first reason why you might want to set the batch_size is because you don’t have enough local disk space for the number of records you are generating (across all tables).

This isn’t usually a problem though.

The more common problem arises from the fact that Salesforce bulk uploads are always done in batches of records a particular SObject. So in the case above, it would upload 1000 Accounts, then 500 Contacts, then 1500 Opportunities. (remember that our scenario involves a recipe that generates 10 Accounts, 5 Contacts and 15 Opportunities).

Imagine if the numbers were more like 1M, 500K and 1.5M. And further, imagine if your network crashed after 1M Accounts and 499K Contacts were uploaded. You would not have a single “complete set” of 10/5/15. Instead you would have 1M “partial sets”.

If, by contrast, you had set your batch size to 100_000, your network might die more around the 250,000 Account mark, but you would have 200,000/20 1 =10K complete sets plus some “extra” Accounts which you might ignore or delete. You can restart your load with a smaller goal (800K Accounts) and finish the job.

1: remember that our sets have 20 Accounts each

Another reason you might choose smaller batch sizes is to minimize the risk of row locking errors when you have triggers enabled. Turning off triggers is generally preferable, and CumulusCI has a task for doing for TDTM trigger handlers, but sometimes you cannot avoid them. Using smaller batch sizes may be preferable to switching to serial mode. If every SObject in a batch uploads less than 10,000 rows then you are defacto in serial mode (because only one “bulk mode batch” at a time is being processed).

In general, bigger batch sizes achieve higher throughput. No batching at all is the fastest.

Smaller batch sizes reduce the risk of something going wrong. You may need to experiment to find the best batch size for your use case.

Automate Data Operations¶

The Lifecycle of a Dataset¶

Defining Datasets¶

API Selection¶

Database Mapping¶

Record Types¶

Relative Dates¶

Person Accounts¶

Record Types¶

Extract¶

Load¶

Advanced Features¶

Primary Keys¶

Handling Namespaces¶

Namespace Handing with Multiple Mapping Files¶

Optional Data Elements¶

Custom Settings¶

Dataset Tasks¶

create_bulk_data_permission_set¶

extract_dataset¶

Options¶

load_dataset¶

Options¶

generate_dataset_mapping¶

Reference Cycles¶

Options¶

load_custom_settings¶

Options¶

delete_data¶

Examples¶

Generate Fake Data¶

Generated Record Counts¶

Controlling the Loading Process¶

Batch Sizes¶

`create_bulk_data_permission_set`¶

`extract_dataset`¶

`load_dataset`¶

`generate_dataset_mapping`¶

`load_custom_settings`¶

`delete_data`¶