Cloud Functions for Firebase killed due to memory limit exceeded
FirebaseMemory ManagementGoogle Cloud-FunctionsFirebase Problem Overview
I keep getting a sporadic error from Cloud Functions for Firebase when converting a relatively small image (2mb). When successful, the function only takes about 2000ms or less to finish, and according to Image Magick documentation should I should not see any problems.
I tried increasing the buffer size for the command, which isn't allows from within Firebase, and I tried to find alternatives to .spawn()
as that could be overloaded with garbage and slow things down. Nothing works.
Firebase Solutions
Solution 1 - Firebase
You can set this from within your Cloud Function file on Firebase.
const runtimeOpts = {
timeoutSeconds: 300,
memory: '1GB'
}
exports.myStorageFunction = functions
.runWith(runtimeOpts)
.storage
.object()
.onFinalize((object) = > {
// do some complicated things that take a lot of memory and time
});
Taken from the docs here: https://firebase.google.com/docs/functions/manage-functions#set_timeout_and_memory_allocation
Don't forget to then run firebase deploy
from your terminal.
Solution 2 - Firebase
I was lost in the UI, couldn't find any option to change the memory, but finally found it:
- Go to the Google Cloud Platform Console (not the Firebase console)
- Select Cloud Functions in the menu
- Now you see your firebase function in here if it's correct. Otherwise check if you selected the right project.
- Ignore all checkboxes, buttons and menu items, just click on the name of the function.
- Click on edit (top menu) and only change the allocated memory and click save.
Solution 3 - Firebase
[update] As one commenter suggested, this should no longer be an issue, as firebase functions now maintain their settings on re-deploy. Thanks firebase!
Turns out, and this is not obvious or documented, you can increase the memory allocation to your functions in the Google Functions Console. You can also increase the timeout for long-running functions. It solved the problem with memory overload and everything is working great now.
Edit: Note that Firebase will reset your default values on deploy, so you should remember to login to the console and update them right away. I am still looking around for a way to update these settings via CLI, will update when I find it.
Solution 4 - Firebase
The latest firebase deploy command does overwrite the memory allocation to default 256MB and timeout up to 60s.
Alternatively , to specify the desired memory allocation and maximum timeout , I use gcloud command such as:
gcloud beta functions deploy YourFunctionName --memory=2048MB --timeout=540s
Other options, please refer to:
https://cloud.google.com/sdk/gcloud/reference/beta/functions/deploy
Solution 5 - Firebase
Solution 6 - Firebase
Figuring out from UI is a bit tricky so here are some guided screenshots:
Go to url https://console.cloud.google.com/functions/list
You can also increase default timeout of 60 sec
Solution 7 - Firebase
Update: It looks that they now preserve settings on re-deploy so you can safely change memory allocation in cloud console!
Solution 8 - Firebase
you can add the configurations in your firebase functions definitions something like:
functions.runWith({memory: '2GB', timeoutSeconds: '360'})
Solution 9 - Firebase
It seems the default ImageMagick resource config in Firebase Cloud Functions doesn't match the actual memory allocated to the function.
Running identify -list resource
in the context of a Firebase Cloud Function yields:
File Area Memory Map Disk Thread Throttle Time
--------------------------------------------------------------------------------
18750 4.295GB 2GiB 4GiB unlimited 8 0 unlimited
The default memory allocated to a FCF is 256MB - the default ImageMagick instance thinks it has 2GB and therefore doesn't allocate buffer from disk and can easily try to over allocate memory causing the function to fail on Error: memory limit exceeded. Function killed.
One way is to increase required memory as suggested above - although there's still risk IM will try to over allocate depending on your use case and outliers.
Safer yet would be to set the correct memory limit to IM as part of the image manipulation process using -limit memory [your limit]
. You can figure out your approx memory usage by running your IM logic with `-debug Cache' - it will show you all the buffers allocated, their sizes and if they were memory or disk.
If IM hits the memory limit it will start allocating buffers on disk (memory mapped and then regular disk buffers.You'll have to consider your specific balance between I/O performance vs memory cost. Price of every additional byte of memory you allocate to your FCF is multiplied by 100ms of usage - so that can grow quickly.
Solution 10 - Firebase
Another option here would be to avoid using .spawn()
altogether.
There is a great image processing package for node called Sharp that uses the low-memory footprint library libvips. You can check out the Cloud Function sample on Github.
Alternately, there is a Node wrapper for ImageMagick (and GraphicsMagick) called gm. It even supports the -limit option to report your resource limitations to IM.