A Serverless Function has exceeded the unzipped maximum size

[cannonmaster]

Summary

I am using the next-flask template from vercel as the start point of my project.

The issue is on deploy stage, and the error is " A Serverless Function has exceeded the unzipped maximum size of 250 MB". It clearly indicates the dependencies size is over the limit.

First, I went to measure the size of the deps installed on both frontend and backend (flask), and none of them has abnormal size.

Then, I used vercel cli to build the project locally to list what's actually installed for each individual serverless functions during the build stage.

To my superise, vercel installed bunch of deps for my project automatically even almost all of them is not related to my project.

In addition, most of them have a fairly large size such as numpy (this is not used in the project even).

I think that this causes the oversize errors and of course currently not be able to deploy on vercel.

I have stucked on this issue whole day and tried to find solution everywhere but no luck.

Does someone could share some solution to solve it ...
For your reference, here is the deps used by my backend,

Flask==2.2.2
supabase==1.0.4
python-dotenv==1.0.0
qdrant-client==1.5.4
Flask-Limiter==3.5.0
Flask-Cors==4.0.0
langchain==0.0.304
langsmith==0.0.41

Example

No response

Steps to Reproduce

a normal build then deploy vercel gives the error

[amyegan]

Hi @cannonmaster! I just created a new project from the Next.js Flask Starter template and didn't run into this problem. Did I use the wrong template?

[BigSamu]

I am also running into the same problem in a NextJS+FastAPI application. I started getting this error when deploying in Vercel for my Pro account after many successful previous deployments:

 Error: A Serverless Function has exceeded the unzipped maximum size of 250 MB. : https://vercel.link/serverless-function-size

This problem is running me nuts, and I don't know how to debug it. A previous version of a deployment worked perfectly. Don't know what I could changed that caused this problem. I think it could be a limit or error from Vercel in terms of how many times I have run deployments. Indeed last two days I have run so many of them because I was getting familiar with the functionality of the platform.

When reviewing the size of the latest build (checking the size of my .next folder), I have 194M unzipped, which is less than 250MB.

As a reference, I attach below the list of dependencies I am using:

NextJS

 "dependencies": {
    "axios": "^1.6.0",
    "bootstrap": "^5.3.2",
    "cookie": "^0.5.0",
    "lodash": "^4.17.21",
    "macaroons.js": "^0.3.9",
    "next": "^14.0.2",
    "react": "^18",
    "react-bootstrap": "^2.9.1",
    "react-bootstrap-icons": "^1.10.3",
    "react-dom": "^18"
  },
  "devDependencies": {
    "concurrently": "^8.2.2"
  }

FastAPI

annotated-types==0.6.0
anyio==3.7.1
bcrypt==4.0.1
cffi==1.16.0
click==8.1.7
dnspython==2.4.2
email-validator==2.1.0.post1
exceptiongroup==1.1.3
fastapi==0.104.1
greenlet==3.0.1
h11==0.14.0
idna==3.4
passlib==1.7.4
psycopg2-binary==2.9.9
pycparser==2.21
pydantic==2.4.2
pydantic-settings==2.0.3
pydantic_core==2.10.1
pymacaroons==0.13.0
PyNaCl==1.5.0
python-dateutil==2.8.2
python-dotenv==1.0.0
python-multipart==0.0.6
six==1.16.0
sniffio==1.3.0
SQLAlchemy==2.0.23
starlette==0.27.0
typing_extensions==4.8.0
uvicorn==0.24.0.post1

Also here is the link to the GitHub project if someone wants to review it entirely.

I would appreciate any support!

Regards,

Samuel

[shane-cw]

Same :( Nearly the same output, leading to deploying outputs... going for a few minutes, sometimes 20 minutes, then the serverless function size too large. My next step is to comment out my functions one at a time I guess

[amyegan]

The general recommendation for this error message is to work on reducing function size, either by configuring includeFiles and excludeFiles, removing third party dependencies, or reducing the overall size/complexity of the function. More info on this can be found in the runtime bundle size limits doc: https://vercel.com/docs/functions/serverless-functions/runtimes#bundle-size-limits

With Python Serverless Functions, you'll need to be careful about Python dependencies. More info on proper setup can be found in the Python Runtime docs: https://vercel.com/docs/functions/serverless-functions/runtimes/python

You can use Vercel CLI on your local computer to get build info without starting a whole new deployment. Just run vercel build to run a local build, or add the --debug option to log additional details.

[PhilDakin]

I just encountered and fixed this issue for a project I'm working on and noticed a few odd things here.

We use the workaround for Python 3.11 support so building locally. I noticed that the size of our Vercel build artifact for our Python function is very large. On closer inspection, it seems that the entirety of .next/cache is being duplicated into the artifact for our Python serverless function:

23-12-22 15:54:46 /Users/Phil/Documents/Chew/poker/mayfair: du -sh .vercel/output/functions/api/scenario/index.func/.next
241M    .vercel/output/functions/api/scenario/index.func/.next
23-12-22 15:54:48 /Users/Phil/Documents/Chew/poker/mayfair: du -sh .vercel/output/functions/api/scenario/index.func      
275M    .vercel/output/functions/api/scenario/index.func

This seems odd for two reasons:

1. Why are any next artifacts shipped for our Python serverless function?
2. It looks like even development-only packages from the .next/cache are also duplicated into our production deployment, and are quite large.

I was able to fix this on our side by first removing .next before running our Vercel build and deploy. It greatly reduces the size of our artifact (~100M!):

rm -rf .next && vc build && vc deploy --prebuilt

23-12-22 16:02:00 /Users/Phil/Documents/Chew/poker/mayfair: du -sh .vercel/output/functions/api/scenario/index.func
153M    .vercel/output/functions/api/scenario/index.func

[egeste]

@PhilDakin I am having the same issue. Were you able to overcome this and get your builds succeed and deploy via the vercel ci/cd pipeline? I've been trying all kinds of stuff, but haven't been able to get it to work. If you have it working, could you please write up your process?

[PhilDakin]

@egeste unfortunately because we were using the 3.11 workaround, I only ever did this in the local deploy path with the solution I described previously. Can't share any insight on the same issue in the Vercel pipeline, though I would anticipate that it is a different issue than mine because there would be nothing in the .next cache in Vercel's pipeline.

[tedlittledale]

I'm getting the same issue, I tried building locally and deploying, which successfully deployed but then I get 405 errors when actually calling the function. Looking at the logs it can't find the module I've imported
e.g.
[ERROR] Runtime.ImportModuleError: Unable to import module 'vc__handler__python': No module named 'cloudinary'

[svossen]

I had this issue with a Nuxt 2 app in combination with vercel-builder.

vercel-builder suggests to use Yarn due to its autoclean functionality. I was already using Yarn, but not the autoclean functionality. After setting this up the error message disappeared and my deployments were successful again.

[chandanmaruthi]

Try Removing the local env folder "env" from your git repo
I faced this issue today and I realized I did not have gitignore for env folder which resulted in me checking in the entire env into git.
After removing the env (local environment) folder from git. this issue went away

[shredEngineer]

The general recommendation for this error message is to work on reducing function size, either by configuring includeFiles and excludeFiles, removing third party dependencies, or reducing the overall size/complexity of the function. More info on this can be found in the runtime bundle size limits doc: https://vercel.com/docs/functions/serverless-functions/runtimes#bundle-size-limits

With Python Serverless Functions, you'll need to be careful about Python dependencies. More info on proper setup can be found in the Python Runtime docs: https://vercel.com/docs/functions/serverless-functions/runtimes/python

You can use Vercel CLI on your local computer to get build info without starting a whole new deployment. Just run vercel build to run a local build, or add the --debug option to log additional details.

How do I even find out if it's a Python or JS serverless function? This is extremely frustrating.

[btahir]

Can we get an update on this? It seems this issue has been outstanding for months now.

[darisi]

Been having this error for a month now with Next.js. My pre-built .vercel folder is 38MB unzipped so not sure what would be 250MB large or which "function" this would be.

[btahir]

I see a bunch of Python process outputs in my root folder when I run vercel build so maybe its something to do with setting the build output directory?

[btahir]

Bizarre. I was able to get it to build if I deleted my public, lib, other root level folders as well as thin out the app folder to a simple page. I have no idea why this would cause the python function bundle size to be affected (I am not importing anything from my other files and even replaced the Python code with the simple Flask Hello World example and it still gave the same error).

I tried excludeFiles but did not work. Any configuration someone can share that basically excludes all files from my python function?

[shredEngineer]

In my case, looking at the build system report (set VERCEL_BUILD_SYSTEM_REPORT=1 in your environment variables), we have seen that swc may contribute to the huge serverless functions:

[22:30:15.866] • 2 files larger than 100 MB detected on disk:
[22:30:15.866]   ‣ 138 MB : node_modules/@next/swc-linux-x64-musl/next-swc.linux-x64-musl.node
[22:30:15.866]   ‣ 115 MB : node_modules/@next/swc-linux-x64-gnu/next-swc.linux-x64-gnu.node

Now I understand that swc is Vercel's transpiler, and I have chosen to override it by creating a .babelrc file. (Babel is a legacy transpiler.)

Then I get this error:

"next/font"` requires SWC although Babel is being used due to a custom babel config being present.

So I commented out the code using next/font in my project. (Thinking that maybe that package was the only one requiring these huge binaries.)

After removing the .babelrc again (so we're using swc again), now having commented out the next/font usage, it deploys successfully!

Thus, the error is connected to the next/font package, which requires swc, which does not fit into a serverless function.

This fits into the narrative of other users having had success with explicitly excluding swc from the serverless functions by editing next.config.js: https://github.com/orgs/vercel/discussions/103#discussioncomment-6356642

• I still don't know WHICH function, tho. Probably all of them?
• I have multiple JS and 1 Python serverless function(s). None of these display fonts or anything like that. Just Supabase access.

[shredEngineer]

Excluding "everything" from my Python runtime api/ folder removed the "Huge Serverless Function Error" for me, BUT now my Python API is not working anymore. However, I see this as progress...

I only have Python files in my api/ folder, so my API doesn't rely on any of the Next.js files.

Here's my understanding, even though this separation is not clearly documented:

• To exclude files from Next.js serverless functions, use outputFileTracingExcludes in next.config.js.
• To exclude files from Python serverless functions, use excludeFiles in vercel.json.

At first, I tried the next.config.js way, which successfully deployed once, and then failed again... god knows why:

The relevant part is the `experimental: { … }` config:

module.exports = {
	rewrites: async () => {
		return [
			{
				source: '/api/',
				destination:
					process.env.NODE_ENV === 'development'
						? 'http://127.0.0.1:5328/api/'
						: '/api/',
			},
		]
	},
	experimental: {
		outputFileTracingExcludes: {
			'/api/': [
				'./node_modules/@next/swc-linux-x64-musl/**/*',
				'./node_modules/@next/swc-linux-x64-gnu/**/*',
			],
		},
	},
};

When the above method did fail again, I now tried the vercel.json way, which did resolve the issue for me (at least right now, and I hope it stays that way):

{
	"functions": {
		"api/**/*": {
			"excludeFiles": "**/*"
		}
	}
}

In summary, I guess Vercel's nft library, which decides what files to include or exclude, doesn't handle the Python runtime path correctly. Also, the separate handling of next.config.js for Next.js serverless functions VS vercel.json for Python serverless functions is not documented anywhere, and was not obvious to me at all.

Right now I am just using excludeFiles and not outputFileTracingExcludes.

As I said, this removed the error for me, so it is clear that it's connected to the Python API.

However, by excluding "everything" the Python API doesn't work anymore.

I'll keep you posted about my experience with this error, which is, quite frankly, just the biggest pain in the ass.

PS: I also disabled the build cache and enabled some debugging output using these environment variables:

VERCEL_FORCE_NO_BUILD_CACHE=1
NEXT_DEBUG_FUNCTION_SIZE=1
VERCEL_BUILD_SYSTEM_REPORT=1

Moreover, I disabled the Remote Cache. Dunno if that was necessary, but I did it: https://vercel.com/docs/monorepos/remote-caching#enable-and-disable-remote-caching-for-your-team

PPS: Found another bug: The VERCEL_FORCE_NO_BUILD_CACHE variable doesn't show up after creating it in the Web UI; but it is still there, since I cannot re-add it.

[shredEngineer]

Aaaaaaaaaaand there the error is again. I just kept coding and adding Chakra UI components. No big deal. Now the "Huge Serverless Function Error" is back… argh

[shredEngineer]

Got the error removed, but now my Python API doesn't work anymore. I edited my post to include the updated answer.

[rayy40]

Oh God, I was able to finally fix the error by going through your approach. Thanks a ton, dude. Was struggling with this for hours now.
Edit: Didn't quite work, lol. Have decided to use render rather.

[btahir]

Thanks. I tried this but unfortunately did not work for me. I even tried the kitchen sink:

outputFileTracingExcludes: {
      '/*': ['./*'],
    },

but no dice. I've narrowed it down to my public folder growing past a certain size. If i take a big subfolder out the build runs otherwise I get the error. Are we sure outputFileTracingExcludes works for Python functions in api folder at the root level or just for the Next.js ones in app/api?

[shredEngineer]

Are we sure outputFileTracingExcludes works for Python functions in api folder at the root level or just for the Next.js ones in app/api?

I'd sure like to know how to find that out… @amyegan

EDIT: It seems like it does not. However, using excludeFiles in vercel.json DOES seem to work for the Python functions in api/. See my edited post above.

[btahir]

Thanks for sharing! This worked for me as well although weirdly the build keeps running without ending if I try to narrow down the folder to exclude

"excludeFiles": "./public/*"

Was hoping that by only excluding certain folders you can get the python API to work and not have the bundle error.

[housUnus]

I have the same issue (" A Serverless Function has exceeded the unzipped maximum size of 250 MB. : https://vercel.link/serverless-function-size"), I tried to remove as much as possible of non-needed packages but nothing changed, this is my installed packages from vercel:
Successfully installed Django-4.2.3 MarkupPy-1.14 MarkupSafe-2.1.3 Pillow-10.0.0 PyJWT-2.8.0 PyYAML-6.0.1 Pygments-2.15.1 Unidecode-1.3.6 Werkzeug-2.3.6 amqp-5.1.1 asgiref-3.7.2 asttokens-2.2.1 backcall-0.2.0 billiard-4.1.0 certifi-2023.5.7 cffi-1.15.1 chardet-3.0.4 charset-normalizer-3.2.0 click-8.1.6 click-didyoumean-0.3.0 click-plugins-1.1.1 click-repl-0.3.0 colorama-0.4.6 crispy-bootstrap5-0.7 cryptography-41.0.2 cssselect2-0.7.0 decorator-5.1.1 defusedxml-0.7.1 diff-match-patch-20230430 django-admin-honeypot-1.1.0 django-allauth-0.54.0 django-appconf-1.0.5 django-bulk-update-or-create-0.3.0 django-categories-1.9.2 django-ckeditor-6.6.1 django-components-0.35 django-compressor-4.4 django-constance-2.9.1 django-crispy-forms-2.0 django-debug-toolbar-4.1.0 django-decorator-include-3.0 django-etc-1.4.0 django-filer-3.0.4 django-filter-23.2 django-grip-3.4.0 django-hitcount-1.3.5 django-htmx-1.16.0 django-import-export-3.2.0 django-ipware-5.0.0 django-jazzmin-2.6.0 django-js-asset-2.1.0 django-maintenance-mode-0.18.0 django-modeltranslation-0.18.11 django-mptt-0.14.0 django-ninja-0.22.2 django-picklefield-3.1 django-polymorphic-3.1.0 django-qsstats-magic-1.1.0 django-robots-5.0 django-rosetta-0.9.9 django-select2-8.1.2 django-shortcuts-1.6 django-split-settings-1.2.0 django-sql-explorer-3.2.1 django-widget-tweaks-1.4.12 easy-thumbnails-2.8.5 et-xmlfile-1.1.0 executing-1.2.0 gripcontrol-4.2.0 gunicorn-21.2.0 h11-0.9.0 h2-3.2.0 hpack-3.0.0 hstspreload-2023.1.1 html2text-2020.1.16 httpcore-0.9.1 httpx-0.13.3 hyperframe-5.2.0 idna-2.10 install-1.3.5 ipython-8.12.2 jedi-0.18.2 kombu-5.3.1 ldif3-3.1.1 lxml-4.9.3 matplotlib-inline-0.1.6 oauthlib-3.2.2 odfpy-1.4.1 openpyxl-3.1.2 packaging-23.2 parso-0.8.3 pexpect-4.9.0 pickleshare-0.7.5 polib-1.2.0 prompt-toolkit-3.0.39 psycopg2-binary-2.9.9 ptyprocess-0.7.0 pubcontrol-3.5.0 pure-eval-0.2.2 pycparser-2.21 pydantic-1.10.11 python-card-me-0.9.3 python-dateutil-2.8.2 python-dotenv-1.0.1 python-fsutil-0.10.0 python3-openid-3.2.0 pytz-2023.3 rcssmin-1.1.1 reportlab-4.0.4 requests-2.31.0 requests-oauthlib-1.3.1 rfc3986-1.5.0 rjsmin-1.2.1 six-1.16.0 sniffio-1.3.0 sorl-thumbnail-12.9.0 sqlparse-0.4.4 stack-data-0.6.2 svglib-1.5.1 tablib-3.5.0 tinycss2-1.2.1 toposort-1.10 traitlets-5.9.0 typing_extensions-4.7.1 tzdata-2023.3 tzlocal-5.0.1 ua-parser-0.18.0 unicode-slugify-0.1.5 unicodecsv-0.14.1 urllib3-1.26.6 user-agents-2.2.0 vine-5.0.0 wcwidth-0.2.6 webencodings-0.5.1 whitenoise-6.6.0 xlrd-2.0.1 xlwt-1.3.0

I'm such at this point now, no idea how to figure it out, note that all those packages size on my local is less then even 200MB

[interactda]

Same issue here even with way less packages, cannot install boto3

[libaxuan]

我使用3.9 3.10 3.11 3.12 都不能解决这个问题
都提示
Error: A Serverless Function has exceeded the unzipped maximum size of 250 MB. : https://vercel.link/serverless-function-size

太失望了

[housUnus]

Is thing going to be solved?

[shredEngineer]

AFAIK they are in desparate need of examples to reproduce the error

[housUnus]

i can my project as public so they can try to deploy and produce the error, but where are they?!

[janibolkvadze]

Having the same issue. My Vue/Nuxt application is not even that complex.
Dependencies:

"dependencies": {
    "@nuxtjs/apollo": "4.0.1-rc.5",
    "@nuxtjs/sitemap": "2.4.0",
    "@vue/babel-helper-vue-jsx-merge-props": "1.0.0",
    "@vue/babel-preset-jsx": "1.1.2",
    "core-js": "3.15.1",
    "g": "2.0.1",
    "graphql-tag": "2.12.6",
    "nuxt": "2.15.7",
    "sass": "1.32.6",
    "sass-loader": "10",
    "vee-validate": "3.4.14",
    "vue": "2.7.14",
    "vue-class-component": "7.2.6",
    "vue-property-decorator": "8.4.2",
    "vue-template-compiler": "2.7.14"
  },
  "devDependencies": {
    "@nuxt/types": "2.15.7",
    "@nuxt/typescript-build": "2.1.0",
    "@nuxtjs/svg": "0.4.0",
    "@nuxtjs/vuetify": "1.12.3",
    "@vue/babel-helper-vue-jsx-merge-props": "1.0.0",
    "@vue/babel-preset-jsx": "1.1.2",
    "babel-helper-vue-jsx-merge-props": "2.0.3",
    "babel-plugin-syntax-jsx": "6.18.0",
    "babel-plugin-transform-vue-jsx": "3.7.0",
    "vue-class-component": "7.2.3",
    "vue-property-decorator": "8.4.2"
  }

Unable to deploy changes.

[juliussneezer04]

Same here! Using a python-only fastAPI app with the template as starting point but fails for this reason, with no clues on how to debug it. Running vercel build locally is successful, however it installs and dumps all the requirements in my local folder and resets my venv which is frustrating as well. My requirements.txt are

fastapi==0.104.0
firebase-admin==6.4.0
google-api-python-client==2.102.0
google-auth-oauthlib==1.1.0
google-cloud-scheduler==2.11.2
httpx==0.24.1
langchain==0.1.13
openai==0.28.1
pytz==2023.3
itsdangerous==2.1.0
python-dotenv==1.0.0
supabase==1.2.0
urllib3==1.26.15
uvicorn==0.27.1
gunicorn==21.2.0

a bit of tweaking requirements shows langchain is the one causing the server less functions to exceed. Does anyone know if it's a langchain issue or independent?

[taylorjdawson]

@amyegan I vercel build is also outputing to my root dir which seems very wrong. What is the resolution to this?

[w4rlock999]

@juliussneezer04 how you solve this? I think it is really a bug, I even tried to redeploy my past successful deployment and it just fails today =__=

[juliussneezer04]

@w4rlock999 I couldn't solve it, and it was taking too long for my liking to investigate deeper, so have since switched providers. I do like vercel so I'm constantly on the look for a fix :(

[w4rlock999]

what provider you use currently? yeah vercel is really comfy UNTIL this problem T.T

[paulolacombe]

same here, got a light next + flask build. Deployed fine a couple of times, the poof. Can't solve it with the CLI either because of API limits in the free version

[HTMHell]

Try adding a vercel.json file in the root of the project, with the following:

{
    "functions": {
        "api/**": {
            "excludeFiles": "{.next,*.cache,node_modules,public,app}/**"
        }
    }
}

[VladOnMyOwn]

@dchen327, try to use .vercelignore

[vcolano]

I'm facing the same error in my vercel project using NextJS functions via pages/api and Python/FastAPI functions via api. It sounds like this solution is working for others using Python Vercel functions, but I can't get it working.
I have the following project structure:

├── api
│   ├── Dockerfile
│   ├── __init__.py
│   ├── index.py  # has my FastAPI app and exposes an endpoint.
│   └── requirements.txt
├── pages
│   ├── api
│   │   └── trpc
│   │       └── [trpc].ts. # All of my tRPC endpoints.
└── vercel.json

My vercel.json looks like

{
  "functions": {
    "api/**": {
      "excludeFiles": "{.next,*.cache,node_modules,public,app}/**"
    }
  }
}

And in my build logs I get

Any help would be much appreciated!

[trevorpfiz]

@vcolano I couldn’t get it to work. Maybe my FastAPI was actually too big? I used docker to host it on Fly.io with no issues.

[lokki007]

Worked for me. Though I am using "builds" and I can't use it with "functions"

Here's my vercel.json:

{
"version": 2,
"builds": [
{
"src": "app/api//*.py",
"use": "@vercel/python",
"config": {
"excludeFiles": "/{.next,*.cache,node_modules,public,app}/**"
}
},
{
"src": "next.config.js",
"use": "@vercel/next"
}
],
"routes": [
{ "src": "/app/api/webhook", "dest": "/app/api/webhook.py" },
]
}

[peguerosdc]

This fixed it for me and my python functions (with sanic), thanks!

[trevorpfiz]

I am trying to deploy a FastAPI from a Turborepo as detailed in https://github.com/cording12/next-fast-turbo with a few extra packages, but no luck. Even tried deploying it by itself, but same error. Then tried the vercel.json from the Django/Flask examples, no luck. Not sure if actually over maximum size or running into bugs.

{
  "rewrites": [{ "source": "/(.*)", "destination": "/api/main.py" }]
}

Deploying FastAPI in a Turborepo seems even more bugged when not using builds.

[VladOnMyOwn]

Same