Ongoing Tasks: RavenDB ETL

• Loading Documents
• Alternative Syntax
• Documents Identifiers
• Filtering
• Loading Data from Other Documents
• Accessing Metadata
• Creating Multiple Documents from a Single Document

Loading Documents

• To load data to the destination database you must call the loadTo<CollectionName>() method and pass a JS object.

• Indicating the collection name in the load method is a convention.

• The objects passed to the loadTo method will be sent to the target database in the last stage - Load.

• All results created in a single ETL run will be sent in a single batch and processed transactionally in the destination.

  • For example, if you want to write data to the Employees collection you need to call the following method in the script body:

    loadToEmployees({ ... });

• The method parameter must be a JS object. You can create it as follows:

  loadToEmployees({
          Name: this.FirstName + " " + this.LastName
      });

• Or simply transform the current document object and pass it:

  this.Name = this.FirstName + " " + this.LastName;
  
      delete this.Address;
      delete this.FirstName;
      delete this.LastName;
  
      loadToEmployees(this);

Example: loadTo Method

The following is an example of a RavenDB ETL script processing documents from the Employees collection:

var managerName = null;

    if (this.ReportsTo !== null)
    {
        var manager = load(this.ReportsTo);
        managerName = manager.FirstName + " " + manager.LastName;
    }

    // load documents to `Employees` collection in the destination
    loadToEmployees({
        // the loaded documents will have these fields:
        Name: this.FirstName + " " + this.LastName,
        Title: this.Title,
        BornOn: new Date(this.Birthday).getFullYear(),
        Manager: managerName
    });

Documents Identifiers

The documents generated in the destination database are given an ID according to the collection name specified in the loadTo method.

If the specified destination collection is the same as the source then the document is loaded to the same collection and the original identifier is preserved.

• For example, the following ETL script defined in the Employees collection will keep the same identifiers in the target database:
  

  // original ID will be preserved
      loadToEmployees({ ... });

If the 'loadTo' method indicates a different destination collection, e.g. People,
then the Employees documents will get new identifiers that combine the original ID and the new collection name in the destination database.

This forces us to load new documents with incremented IDs instead of overwriting the fields in existing documents.

By default, RavenDB deletes the old document version in the destination.
This can be changed by changing the deletions behavior.

RavenDB has to create a new, updated document in the destination with an incremented server-made identity.

• For example, if the source collection is Employees while the destination collection is People:

  // a new document with a new, incremented identifier will be generated in destination 
      // by default, the old version will be deleted
      loadToPeople({ ... });

• In addition, ETL appends the symbol / to the requested id so that the target database will generate identifiers on its side. As a result, documents in the People collection in the target database will have identifiers such as: employees/1-A/people/00000000000000000024-A.

Filtering

Documents can be filtered from the ETL by calling the loadTo method only for documents that match some condition:

if (this.Active) {
        // load only active users
        loadToEmployees({ ... });
    }

Loading Data from Other Documents

The load method loads a document with the specified ID into the script context so it can be transformed.

// this.ReportsTo has some document ID
    var manager = load(this.ReportsTo);

Accessing Metadata

The metadata can be accessed in the following way:

var value = this['@metadata']['custom-metadata-key'];

Creating Multiple Documents from a Single Document

The loadTo method can be called multiple times in a single script.
That allows you to split a single source document into multiple documents on the destination database:

// documents will be created in `Addresses` collection
    loadToAddresses({
        City: this.Address.City ,
        Country: this.Address.Country ,
        Address: this.Address.Line1
    });

    delete this.Address;

    // documents will be created in the `Employees` collection
    loadToEmployees(this);

• An ETL task can be created with an empty script.
• The documents will be transferred without any modifications to the same collection as the source document.
  

• Attachments are sent automatically when you send a full collection to the destination using an empty script.

• If you use a script you can indicate that an attachment should also be sent by using dedicated functions:

  • loadAttachment(name) returns a reference to an attachment that is meant be passed to addAttachment()
  • <doc>.addAttachment([name,] attachmentRef) adds an attachment to a document that will be sent in the process, <doc> is a reference returned by loadTo<CollectionName>()

• Sending attachments together with documents
• Changing attachment name
• Loading non-existent attachment
• Accessing attachments from metadata

Sending attachments together with documents

• Attachment is sent along with a transformed document if it's explicitly defined in the script by using addAttachment() method. By default, the attachment name is preserved.
• The script below sends all attachments of a current document by taking advantage of getAttachments() function, loads each of them during transformation, and adds them to a document that will be sent to the 'Users' collection on the destination database.

  var doc = loadToUsers(this);
  
      var attachments = getAttachments();
  
      for (var i = 0; i < attachments.length; i++) {
          doc.addAttachment(loadAttachment(attachments[i].Name));
      }

Changing attachment name

• If addAttachment() is called with two arguments, the first one can indicate a new name for an attachment. In the below example attachment photo will be sent and stored under the picture name.
• To check the existence of an attachment hasAttachment() function is used

  var employee = loadToEmployees({
          Name: this.FirstName + " " + this.LastName
      });
  
      if (hasAttachment('photo')) {
        employee.addAttachment('picture', loadAttachment('photo'));
      }

Loading non-existent attachment

Function loadAttachment() returns null if a document doesn't have an attachment with a given name. Passing such reference to addAttachment() will be no-op and no error will be thrown.

Accessing attachments from metadata

The collection of attachments of the currently transformed document can be accessed either by getAttachments() helper function or directly from document metadata:

var attachments = this['@metadata']['@attachments'];

• Counters are sent automatically when you send a full collection to the destination using an empty script.
• If a script is defined RavenDB doesn't send counters by default.
• To indicate that a counter should also be sent, the behavior function (e.g. by increment operation). If the relevant function doesn't exist, a counter isn't loaded.
• The reason that counters require special functions is that incrementing a counter doesn't modify the change vector of a related document so the document isn't processed by ETL on a change in the counter.
• Another option of sending a counter is to explicitly add it in a script to a loaded document.

• Counter behavior function
• Adding counter explicitly in a script

Counter behavior function

• Every time a counter of a document from a collection that ETL script is defined on is modified then the behavior function is called to check if the counter should be loaded to a destination database.

The function is defined in the script and should have the following signature:

function loadCountersOf<CollectionName>Behavior(docId, counterName) {
       return [true | false]; 
    }

Example: Modifying a Counter Named "downloads"

• The following script is defined on the Products collection:

  if (this.Category == 'software') {
          loadToProducts({
              ProductName: this.Name
          });
      }
  
      function loadCountersOfProductsBehavior(docId, counterName) {
         var doc = load(docId);
  
         if (doc.Category == 'software' && counterName = 'downloads')
              return true;
      }

Adding counter explicitly in a script

Counter behavior functions typically handle counters of documents that are loaded to the same collection. If a transformation script for Employees collection specifies that they are loaded to the People collection in a target database, then due to document ID generation strategy by ETL process (see Documents Identifiers), the counters won't be sent because the final ID of a loaded document isn't known on the source side.

You can use special functions in the script code to deal with counters on documents that are loaded into different collections:

var person = loadToPeople({ Name: this.Name + ' ' + this.LastName });

    person.addCounter(loadCounter('likes'));

• The above example indicates that the likes counter will be sent together with a document. It uses the following functions to accomplish that:
  • loadCounter(name) returns a reference to a counter that is meant be passed to addCounter()
  • <doc>.addCounter(counterRef) adds a counter to a document that will be sent in the process, <doc> is a reference returned by loadTo<CollectionName>()

• If the transformation script is empty, time series are transferred along with their documents by default.
• When the script is not empty, ETL can be set for time series via:
  • Time Series Load Behavior Function
  • Adding Time Series to Documents

Time Series Load Behavior Function

• The time-series behavior function is defined in the script to set the conditions under which time-series data is loaded.
• The load behavior function evaluates each time-series segment and decides whether to load it to the destination database. ETL only updates the data that has changed: if only one time-series entry is modified, only the segment that entry belongs to is evaluated.
• Changes to time-series trigger ETL on both the time-series itself and on the document it extends.
• The function returns either a boolean or an object with two Date values that specify the range of time-series entries to load.
• The time-series behavior function can only be applied to time-series whose source collection and target collection have the same name. Loading a time-series from an Employees collection on the server-side to a Users collection at the target database is not possible using the load behavior function.
• The function should be defined with the following signature:
  

  function loadTimeSeriesOf<collection name>Behavior(docId, timeSeriesName) {
         return [ true | false | <span of time> ];
      }
      //"span of time" refers to this type: { string?: from, string?: to }

Example

The following script is defined in the Companies collection. The behavior function loads each document in the collection into the script context using load(docId), then filters by the document's Address.Country property as well as the time series' name. This sends only stock price data for French companies.

loadToCompanies(this);

    function loadTimeSeriesOfCompaniesBehavior(docId, timeSeriesName) {
       var company = load(docId);

       if (company.Address.Country == 'France' && timeSeriesName = 'StockPrices')
            return true;
    }

Adding Time Series to Documents

• Time series can be loaded into the script context using loadTimeSeries().
• Once a time series is loaded into the script, it can be added to a document using AddTimeSeries().
  

  var employee = loadToEmployees({
          Name: this.Name + ' ' + this.LastName
      });
  
      employee.addTimeSeries(loadTimeSeries('StockPrices'));

Filtering by start and end date

Both the behavior function and loadTimeSeries() accept a start and end date as second and third parameters. If these are set, only time-series data within this time span is loaded to the destination database.

company.addTimeSeries(loadTimeSeries('StockPrices', new Date(2020, 3, 26), new Date(2020, 3, 28)));

function loadTimeSeriesOfUsersBehavior(doc, ts)
{
    return {
        from: new Date(2020, 3, 26),
        to: new Date(2020, 3, 28)
    };
};

Revisions are not sent by the ETL process.

But, if revisions are configured on the destination database, then when the target document is overwritten by the ETL process a revision will be created as expected.

• If the destination collection is different than the source, the identifiers created for the sent documents are different from the original source documents identifiers.
  The source isn't aware of the new IDs created so documents deletion requires a special approach.

• In order to remove the matching documents on the destination side, the deletion of a document sends a command to remove documents that have an ID with a well-known prefix.

• For example:

  1. Document employees/1-A is processed by ETL and put into People collection with ID:
     employees/1-A/people/0000000000000000001-A.
  2. Deletion of the employees/1-A document on the source side triggers sending a command that deletes documents having the following prefix in their ID: employees/1-A/people/.

• If you output multiple documents from a single document then multiple delete commands will be sent, one for each prefix containing the destination collection name.

• When documents are sent to the same collection and IDs don't change then deletion on the source results in sending a single delete command for a given ID.

• Deletions can be filtered by defining deletion behavior functions in the script.

  • Collection specific function deletion handling
  • Generic function for deletion handling

Collection specific function deletion handling

In order to define deletion handling for a specific collection use the following signature:

function deleteDocumentsOf<CollectionName>Behavior(docId) {
   return [true | false]; 
}

• <CollectionName> needs to be substituted by a real collection name that ETL script is working on (same convention as for loadTo method)
• The first parameter is the identifier of a deleted document.

Generic function for deletion handling

Another option is the usage of a generic function for deletion handling:

function deleteDocumentsBehavior(docId, collection) {
   return [true | false]; 
}

• The first parameter is the identifier of a deleted document.

• The second parameter is the name of a collection.

• A document deletion is propagated to a destination only if the function returns true.

Deletions: Filtering deletions in the destination database

You can further specify the desired deletion behavior by adding filters.

By the time an ETL process runs a delete behavior function, the original document is already deleted from the source. It is no longer available. You may want the ETL to set up an archive of documents that were deleted from the source, or save a part of deleted documents in a separate document for later use.

Following are three examples of ways to save documents for later use when they are deleted from the source database:

Example - preventing deletions in the destination database

loadToUsers(this);

function deleteDocumentsOfUsersBehavior(docId) {
    return false;
}

Example - storing deletion info in additional document

When you delete a document you can store a deletion marker document that will prevent propagating the deletion by ETL.

• In the below example if LocalOnlyDeletions/{docId} exists then we skip this deletion during ETL.
• You can add @expires tag to the metadata when storing the marker document, so it would be automatically cleaned up after a certain time by the expiration extension.

  loadToUsers(this);
  
      function deleteDocumentsOfUsersBehavior(docId) {
          var localOnlyDeletion = load('LocalOnlyDeletions/' + docId);
  
          return !localOnlyDeletion;
      }

Example - filtering deletions using the generic function

If you define ETL for all documents, regardless a collection they belong to, then the generic function can be used to filter out deletions by specifying a collection name

function deleteDocumentsBehavior(docId, collection) {
    return 'Users' != collection;
}