Excel Batch Create, Update, and Upsert 06-13-2022 18:35 PM - last edited 06-20-2022 13:48 PM

[takolota]

Update & Create Excel Records 50-100x Faster

I was able to develop an Office Script to update rows and an Office Scripts to create rows from Power Automate array data. So instead of a flow creating a new action API call for each individual row update or creation, this flow can just send an array of new data and the Office Scripts will match up primary key values, update each row it finds, then create the rows it doesn't find.
And these Scripts do not require manually entering or changing any column names in the Script code.

• In testing for batches of 1000 updates or creates, it's doing ~2500 row updates or creates per minute, 50x faster than the standard Excel create row or update row actions at max 50 concurrency. And it accomplished all the creates or updates with less than 25 actions or only 2.5% of the standard 1000 action API calls.

• The Run Script code for processing data has 2 modes, the Mode 2 batch method that saves & updates a new instance of the table before posting batches of table ranges back to Excel & the Mode 1 row by row update calling on the Excel table.
The Mode 2 script batch processing method will activate for creates & updates on tables less than 1 million cells. It does encounter more errors with larger tables because it is loading & working with the entire table in memory.
Shoutout to Sudhi Ramamurthy for this great batch processing addition to the template!
Code Write-Up: https://docs.microsoft.com/en-us/office/dev/scripts/resources/samples/write-large-dataset
Video: https://youtu.be/BP9Kp0Ltj7U

The Mode 1 script row by row method will activate for Excel tables with more than 1 million cells. But it is still limited by batch file size so updates & creates on larger tables will need to run with smaller cloud flow batch sizes of less than 1000 in a Do until loop. 
The Mode 1 row by row method is also used when the ForceMode1Processing field is set to Yes.

Be aware that some characters in column names, like \ / - _ . : ; ( ) & $ may cause errors when processing the data. Also backslashes \ in the data, which are usually used to escape characters in strings, may cause errors when processing the JSON.

Office Script Code
(Also included in a Compose action at the top of the template flow)

Batch Update Script Code: https://drive.google.com/file/d/1kfzd2NX9nr9K8hBcxy60ipryAN4koStw/view?usp=sharing
Batch Create Script Code: https://drive.google.com/file/d/13OeFdl7em8IkXsti45ZK9hqDGE420wE9/view?usp=sharing

You can download the Version 5 of this template attached to this post, copy the Office Script codes into an online Excel instance, & try it out for yourself.
-Open an online Excel workbook, go the the automate tab, select New Script, then copy & paste the Office Script code into the code editor. Do this for both the Batch Update and the Batch Create script code. You may want to name them BatchUpdateV6 & BatchCreateV5 appropriately.
-Once you get the template flow into your environment, follow the notes in the flow to change the settings to your datasources, data, & office scripts.

If you need just a batch update, then you can remove the batch create scope.
If you need just a batch create, then you can replace the Run script Batch update rows action with the Run script Batch create rows action, delete the update script action, and remove the remaining batch create scope below it. Then any update data sent to the 1st Select GenerateUpdateData action will just be created, it won't check for rows to update.

(ExcelBatchUpsertV5 is the core piece, ExcelBatchUpsertV5b includes a Do until loop set-up if you plan on updating and/or creating more than 1000 rows on large tables.)

Anyone facing issues with the standard zip file import package method can check this post for an alternative method of importing the flow: Re: Excel Batch Create, Update, and Upsert - Page 33 - Power Platform Community (microsoft.com)

Or this one: https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Excel-Batch-Create-Update-and-Upsert/m-p...

Thanks for any feedback, & please subscribe to my YouTube channel (https://youtube.com/@tylerkolota?si=uEGKko1U8D29CJ86)

watch?v=HiEU34Ix5gA

[takolota]

Excel Batch Delete Template is Listed Here:

Excel Batch Delete - Power Platform Community (microsoft.com)

And if you want a work-around trigger for When an Excel row is created, modified, or deleted, check here:

https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/When-an-Excel-row-is-created-modified-or...

[takolota]

Looks like I forgot to error handle in the Office Script when there is a primary key in the source data that isn't in the destination Excel workbook.

Check & copy over the new Office Script code in the original post that now sends a list of primary keys not found in the Excel table to the Run script output log.

EDIT: Any primary keys not found are now listed in an array in the “results” piece of the Run script outputs. This can then be referenced in any later actions.

[Yutao]

Thanks for sharing the solution @takolota! Performing 10x faster is very impressive!

I'm wondering if tweaking the Office Scripts a little bit would make the solution run even faster -

So instead of reading, finding, and setting table data all inside the nested loops, maybe you can try reading the complete table data into an array in one shot through:

    let tableRange = table.getRange();

    let tableData = tableRange.getValues();

Then you can do searching and updating all against this array tableData. Hopefully you won't need to call any workbook APIs and can use vanilla string and array methods of JavaScript/TypeScript to search and update content within an array.

And at the end, you can put the array data back into the table in just one shot:

    tableRange.setValues(tableData);

The idea here is to avoid accessing workbook data (getColumn, getRange, getRowIndex, getCell, find, etc.) in a loop, because those methods can trigger network calls behind the scenes, which will slow things down especially when dealing with large volume of data. But once you have read the data into the array, all the operations will happen inside memory, which can be a lot faster.

Please refer to this article for more details about Office Scripts performance optimizations: https://docs.microsoft.com/en-us/office/dev/scripts/develop/web-client-performance

Something for you to consider! 🙂

[takolota]

@Yutao 

THANK YOU!

I was reviewing the code and making some minor improvements as you sent that. However I didn't know enough about JavaScript & Office Scripts to know exactly what required a call without looking into it more.

The Run script action just updated 5,000 rows in less than 20 seconds.
Of course, when I tried 50,000 rows the...

TableRange.setValues(TableData)

broke because of a 413, file too large error.

So it seems the next part of this may be optimizing the batch size & chunking of updates.

@Yutao 

[Yutao]

@takolota 

That's AWESOME! Looking forward to seeing your video!

[takolota]

@Yutao 

I would like this to still work regardless of how large people's tables are in Excel. So the TableRange.setValues(TableData) method of uploading the entire new table from memory will have to wait until I have a better way of batching & chunking those updates like this person does in their video:

(2) Office Scripts: Update large Excel range in performant way - YouTube
I also like that there is less chance of errors with the faster table posting method if someone happens to be working in the table when it gets updated. So this will all have to make it into some V2 later on.

For now, I will leave this version of the Office Script with the TableRange.setValues(TableData) for future reference, and I will be changing some parts of it back.

function main(workbook: ExcelScript.Workbook,
	TableName: string,
	PrimaryKeyColumnName: string,
	UpdatedData: updateddata[],
) 
{
	let table = workbook.getTable(TableName);
	let RowNum: number;
	let TableRange = table.getRange()
	let TableData = TableRange.getValues()
	let ArrayPK = table.getColumn(PrimaryKeyColumnName).getRange().getValues().join("#|#").split("#|#")
	let PKValue: string
	let ArrayPKErrors = new Array("")
	let InitialTableRowNum = TableRange.getRowIndex()

//Iterate through each object item in the array from the flow
	for (let i = 0; i < UpdatedData.length; i++) 
	{
		//If the record's Primary Key value is found continue, else post to error log
		if (ArrayPK.indexOf(UpdatedData[i].PK) > 0)
		{
		//Get the row number for the line to update by matching the foreign key from the other datasource to the primary key in Excel
		RowNum = ArrayPK.indexOf(UpdatedData[i].PK)

		//Iterate through each item or line of the current object
		for (let j = 0; j < Object.keys(UpdatedData[i]).length - 1; j++) 
			{
			//Update each value for each item or column given
			TableData[RowNum][Number(Object.keys(UpdatedData[i])[j])] = UpdatedData[i][Object.keys(UpdatedData[i])[j]]
			}
	}
		//Post PK not found value to ArrayPKErrors
		else {ArrayPKErrors.push(UpdatedData[i].PK)}
	}
	//Post table in memory to the Excel table
	TableRange.setValues(TableData);
	console.log("Primary key values not found are listed in the result array")
	//Post ArrayPKErrors to flow results
	ArrayPKErrors.shift()
	return ArrayPKErrors
}
	interface updateddata {
		'0': string,
		'1': string,
		'2': string,
		'3': string,
		'4': string,
		'5': string,
		'6': string,
		'7': string,
		'8': string,
		'9': string,
		'10': string,
		'PK': string
	}

Thankfully, it is still pretty fast if I do a mix of the original version and of the last version. I can have several operations done completely in memory, then just call on the actual Excel table to update each row. That way it's not referencing the entire table at any point and won't error if it is a large table. Unfortunately that final call to update each row has a size limitation as well when referencing the array sent from Power automate. So the size of the array the flow passes should be set to about 1000 or less. The flow is still able to run each batch of 1000 in less than 20 seconds and we can just use a Do until loop to keep sending batches of 1000. I have updated the original post with this version of the Office Script as it will work for many more use-cases even if it is a little slower.

[takolota]

@Yutao 

Do you know if many calculations in a Select action are computationally intensive on Microsoft resources? I know it won't affect us customers & our daily action limits or licenses, but I'm just wondering because I could probably rework a bit of this Reformat update for script action to check which columns exist using the 1st Update data JSON object in a compose above the action, then use much simpler expressions in the Select action that will calculate for each update record.
Although, I could probably also adjust this to check for the difference between '' and null values if people want to use a formula in the update data to either explicitly set the updated cell to empty or to not update or change it with the '' value.

[Yutao]

@takolota 

Sorry I'm not super familiar with the Select operation. Could you please explain a bit what you are trying to achieve with it? It does feel quite overwhelming seeing this 🙀:

Also wondering if this can be done with a script as well.

[takolota]

@Yutao 

Yeah, I've cleaned that up a little bit by shifting part of the formulas to another Select action that doesn't have to query a larger array. The special characters are a work-around the JSON requirement for different key label names because I need to be able to not have unused columns in the update data sent to the script. It helps keep the data package small & accomplishes a lot of pre-filtering so the Office Script is more efficient & has more capacity. Even with the work-around it would have been nice to have a Rept() function for strings.
If you have any ideas on something I can set a formula to so it doesn't start a key value pair, that would be great. I could probably combine the entire key & value into a single formula and enter a blank in the alternative mapping pane for the JSON if the column isn't used, then reformat each item into proper JSON. But that really hurts readability. All these ~| &^ &’ ^< special character placeholder labels are replaced & removed in the following action.