multiple transforms

[jviseur]

I have a template with 11 transforms of which, at present has 7 transform that are basically the same except that the output is different for each of them. The questions I have are:

1. Is there a limitation to the number of transforms that can be added?

2. Is there a way to call the transform with a parameter so like { datafield | transformname("option") }

Thank you for your help
John

[jreyesr]

Hello again.

1. No, I haven't placed any limits, and I'm not aware of N8N itself placing limits on the number of connected "AI Tools", which is what these transforms are, in the end
2. Yes, it's possible to pass parameters to transforms. See for example this sample document (you have a sample workflow that uses that document in this link, you can copy all that JSON and paste it into N8N's editor, or use the Import from URL feature in the top right menu in N8N (the three dots where you also find the WF's settings). The | sumby transform takes a string parameter:

If you want to write a Code transform that takes parameters, configure it like this:

• Enable the Specify Input Schema switch, choose Generate from JSON example on Schema Type, and write a JSON document like this in JSON Example:

  {
      "input": "",
      "args": [""]
  }

  This indicates that the data passed to the node will be like this: input will be a string, and args will be an array, whose single element will be also a string.

• Then, in the Code block, write your usual Javascript. You'll have access to the transform's input (whatever came from before the |, datafield in your example) in query.input, and the args in the query.args array, in the order that they were passed. In your example, query.args[0] will be the string "option"

• Keep in mind that, if your datafield isn't a single string but a more complex object, your transform will receive it as a string anyways. This is a limitation inherited from N8N's core, because these round sub-nodes that connect below a big node and we use here for transforms are almost exclusively used for AI/LLM nodes, which naturally only handle text. If you need to pass objects or arrays or other complex objects in datafield, include JSON.parse(query.input) in your JS code so it converts back into the actual object, and then use the result of that in your operations instead of using query.input directly. You don't need to do this if datafield is already a string

[jviseur]

Thank you for responding to my questions

[jviseur]

I have a template that has 14 pages 11 of which are the same each of those 11 pages hits one of the transforms 44 times. With 7 pages it runs fine but with all 11 pages i get Multi Error and in the log I get Out of Memory (see details below.

What should the n8n Server be set too especially if we are going to process templates using N8N?

Is there a better, more efficient way, to process a data table (11 rows [max], and 7 columns). Repeated over 11 pages?

I have also enclosed the code used in the transformer that gets hit

const docname = query.input?.trim();
const outType = query.args[0];

if (!docname) return "";

const variants = {
  type: "PCS",
  weight: "N/A",
  words: "CARGO IS:  PAPER-VESSEL, WORDS"
};

return variants[outType] || "";

Details

Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 <--- Last few GCs ---> Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 [1:0x2409b000] 1188984 ms: Mark-Compact 245.2 (258.6) -> 244.3 (258.6) MB, pooled: 1 MB, 157.54 / 0.01 ms (average mu = 0.885, current mu = 0.109) allocation failure; scavenge might not succeed Jul 07 22:37:10 [1:0x2409b000] 1189246 ms: Mark-Compact 245.5 (258.8) -> 244.8 (259.1) MB, pooled: 0 MB, 237.40 / 0.01 ms (average mu = 0.766, current mu = 0.095) allocation failure; GC in old space requested Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 <--- JS stacktrace ---> Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory Jul 07 22:37:10 ----- Native stack trace ----- Jul 07 22:37:10 2025-07-07T10:37:10Z Jul 07 22:37:10 1: 0xe13fde node::OOMErrorHandler(char const*, v8::OOMDetails const&) [node] Jul 07 22:37:10 2: 0x11d5070 v8::Utils::ReportOOMFailure(v8::internal::Isolate*, char const*, v8::OOMDetails const&) [node] Jul 07 22:37:10 3: 0x11d5347 v8::internal::V8::FatalProcessOutOfMemory(v8::internal::Isolate*, char const*, v8::OOMDetails const&) [node] Jul 07 22:37:10 4: 0x1402c05 [node]

Thank you

[jreyesr]

The transformer code looks fine to me, it doesn't look like it accidentally eats too much memory.

For those "JavaScript heap out of memory" errors, N8N recommends increasing old memory, which can be done by setting the NODE_OPTIONS environment var to --max-old-space-size=2048 (2GB) or higher if necessary. In theory, Node.js already sets its limit to 2GB if you have over 2.8GB of RAM in the machine, I believe, but maybe there's an issue there or N8N sets it lower by default or something.

Are you running N8N on a personal computer or on a server? If on a computer, you could also check that it's indeed consuming memory using the OS's task manager (Task Manager on Windows, the System Monitor on Ubuntu, I don't know how it's called on MacOS). See if you can spot a large increase in memory usage over time as the workflow runs.

[jviseur]

No it is being run on a server.

Would using {{rows}} {{/rows}} be a more efficient way, to process a data table (11 rows [max], and 7 columns on each of the 11 pages?

Like {{rows}} {{att1}} {{att2}} {{att3}} {{att4}} {{att5}} {{att6}} {{att7}} {{/rows}} on one line with the appropriate data structure to populate the lines

Regards,
John

[jreyesr]

No it is being run on a server.

Do you use n8n start or npx n8n or something along those lines, or Docker/Docker Compose? Have you tried to set the --max-old-space-size config? I've seen it recommended in a few forum posts, see here

Would using {{rows}} {{/rows}} be a more efficient way

Possibly. How are you filling the table now, if not with loops? Directly specifying all the values for the 77 cells?

If you're able to share the report, or perhaps something with the same structure (same number of pages, same tables) and a sample JSON document, that'd help me see if I can reproduce the issue.

Also, how large is the template? N8N also has issues if you try to handle Binary files that are somewhat large, in the range of tens of megabytes. There's another environment var for that, you could also try setting N8N_DEFAULT_BINARY_DATA_MODE to filesystem and see if that helps

[jviseur]

Each row has the following data

Coll1
{Qty1}

Coll 2
{DocName1|docname(“type”)}

Coll 3
{Qty1}

Coll 4
{DocName1|docname(“type”)}

Coll 5
{DocName1|docname(“weight”)}

Coll 6
{DocName1|docname(“words”)}

Coll 7
{DocName1}

The above is repeated for each row and only the number at the end of Qty and DocName changes for each row. The whole process is repeated for each of the 11 pages.

[jviseur]

Regarding your question

Do you use n8n start or npx n8n or something along those lines, or Docker/Docker Compose? Have you tried to set the --max-old-space-size config

TBH I have no idea as I am not the administrator. The info I do have is

App Title and Version n8n 1.97.1
Package Version [io.n8n.cloudronapp@3.89.0]
Installed At 04/26/2025
Last Updated 07/02/2025

I found this in the log
Initializing n8n process
Jul 08 15:36:42 Puppeteer node: Container detected via .dockerenv file

[jreyesr]

Ah, it's on Cloudron. I haven't used it, but in theory you have a file /app/data/env.sh that you can edit: https://docs.cloudron.io/packages/n8n/#custom-env. You should be able to add the following two lines to it:

export N8N_DEFAULT_BINARY_DATA_MODE=filesystem
export NODE_OPTIONS="--max-old-space-size=4096"

Then, reboot N8N (if you can) and try again.

If necessary, modify the number 4096 to fit your installation. 4096 is 4GB (4*1024) would be the appropriate value if N8N has about 5GB of RAM available of RAM, but from what I read the amount of memory available can be configured in Cloudron (or, at least, the administrator can).

In fact, if I'm reading things correctly, by default the N8N installs on Cloudron are limited to 512MB of RAM. From my experience using N8N, that's actually quite low (my local N8N instance, doing practically nothing on a just-rebooted machine, is at about 465MB already). You may want to check that with the administrator of your N8N instance (whoever manages the Cloudron server), ask them if they've modified the default memory limit, and ask them to them bump the limit up to at least 1GB, 2GB would be much better.

[jviseur]

A question on tables in Word
I have a pages that have a grid of 11 by 7. It is a preformatted page.
I the first row I have
{#bol}{Att1}, {Att2}, {Att3}, {Att4}, {Att5}, {Att6}, {Att7}{/bol}

It runs nicely and add the desired row. However It adds the rows not overwriting the existing row. Is there an option to stop it adding or do I need to create empty data elements?

[jreyesr]

I think that the trick here is to only add a single row, not all the rows:

That's also what the docs seem to do: https://docxtemplater.com/docs/tag-types/#creating-multiple-table-rows, and their invoice example:

Apparently, the single row in which you have the tags gets "multiplied" for each item in bol, and they "push down" the empty rows that you've also included in the table. It's not like it overwrites what you have in your table.

[jviseur]

Thank you for the answers to my question. I have initially bumped the ram up to 1.25GiB and process runs like a dream. Also what I have found is that using the {row} {/row} processes faster than having replacements and as a bonus the template looks cleaner.

[jviseur]

This is the completed flow

[jviseur]

Are you able to tell me how DocxTemplater handles inline Carriage Returns in a specific field?

[jreyesr]

You mean, like when the JSON goes like this: {"field": "First paragraph\nSecond paragraph\nThird paragraph"}?

By default, DocxTemplater ignores newlines. However, there's a setting (linebreaks: true in DocxTemplater's options) to convert them to linebreaks (what you'd insert in Word via Ctrl+Enter, as opposed to a normal Enter).

This N8N node uses that setting (treating newline characters as line breaks) as the default, since I considered it more natural than having your templates silently ignore newlines. So, in theory, newlines in JSON strings should appear in the output document (if they don't, that's a bug and it shall be fixed). If you'd like to revert to DocxTemplater's default setting of completely ignoring newlines, there's an advanced option for that under the Add Option button, called Render Newlines. When you just add it, it'll be turned on, since that's the default setting:

[jviseur]

Thank you for the info. I managed to get it to function by adding \n were the soft return is supposed to be in the code node. Now I need to replace that node by generating it from our WordPress host and hopefully that will function. It functioned when we used mailmerge so it should in here.

[jviseur]

Another question:

I am attempting to place a bulleted list in the document and I am using

1. some text already in the document.{#AppendBulletItems}
2. {name}
3. {/AppendBulletItems}

It produces
2. Ecclesiastical Deed Poll
3.
4. Addendum to Ecclesiastical Deed Poll
5.
6. Mandate of Non-Consent
7.
8. Writ of Habeas Corpus
9.
10. Writ of Prohibito
11.
12. courtesy advisory
13.

What I like to see is this
2. Ecclesiastical Deed Poll

3. Addendum to Ecclesiastical Deed Poll

4. Mandate of Non-Consent

5. Writ of Habeas Corpus

6. Writ of Prohibito

7. courtesy advisory

/

const bulletItems = [];
let counter = 2;

if (json.EDP) {
  bulletItems.push({ name: "Ecclesiastical Deed Poll" });
  bulletItems.push({ name: "Addendum to Ecclesiastical Deed Poll" });
}

if (json.MONCWrits) {
  bulletItems.push({ name: "Mandate of Non-Consent" });
  bulletItems.push({ name: "Writ of Habeas Corpus" });
  bulletItems.push({ name: "Writ of Prohibito" });
}

if (json.CourtesyAdvice) {
  bulletItems.push({ name: "courtesy advisory" });
}

json.AppendBulletItems = bulletItems;

Where am I going wrong?

Regards,
John

[jreyesr]

So, that was slightly annoying to get working, but I've got it, I think. You need to use the dash syntax feature (demo here). Then, write the list like this:

1. This is always present
2. {-w:p options}{.}{/options}
3. This is the final element

For example, this template:

renders to this document when using this JSON: {"name": "John", "options": ["Option A", "Option B", "Option C"]}:

(you can, of course, just skip the final fixed element as desired)

From what I understand, the trick there is that the {-w:p ...} tag instructs DocxTemplater to create a new paragraph for each item (it's a variant of {# ...}, but it explicitly states to end and restart the whole paragraph for each iteration). And since the line is inside a numbered list, creating a paragraph for each of course creates new numbered items. And, since the entire tag (from {-w:p ...} to {# ...}) is on a single line, you don't get the empty numbered items between each line that you got on your attempt (those come from item 3. in the template, the {/AppendBulletItems} line, because technically the thing that repeats is {name} and also a newline)

[jviseur]

@jreyesr thank yo so much for your help and assistance. I /we really appreciate it