@Ciantic, they key error is probably:
[Build] main /bin/sh: Invalid ELF image for this architecture
Which usually indicates a CPU architecture mismatch, such as Intel vs ARM, 32 bit vs 64 bit. It is happening at Dockerfile step 5/13, "npm install --unsafe-perm --production".
Given the following output:
[Info] Building for aarch64/raspberrypi4-64
[Info] Installing qemu for aarch64 emulation...
[Debug] Resolving services with [raspberrypi4-64|aarch64]
I gather that you project is for the RPi 4, 64 bits. The base image is however for the RPi 3, 32 bits:
[Build] main Step 1/13 : FROM balenalib/raspberrypi3-node:8-build AS base
I don’t know if this image mismatch is necessarily a problem – I actually seem to remember that the RPi3 image works fine on the RPi4. However, it might be worth trying an RPi 4 base image, balenalib/raspberrypi4-64-node.
Also: balena deploy, as used, is building the image using your laptop (dev machine) Docker installation. To narrow down the issue, try also building without emulation on balenaCloud builders (we have ARM servers) with balena push, and/or building on a real RPi4 device on your desk. To test the latter, flash a development image and place the device in local mode, then run something like:
balena deploy myApp --logs --source . --dockerPort 2375 --dockerHost 192.168.1.2
replacing 192.168.1.2 with your device’s IP address.
If building without emulation works, the problem may be that QEMU is simply not up to the task; it is known to fail on some compilation tasks or other tasks, which npm install may be triggering. Try also running some other binary executables (other than npm) to rule this out – starting with something like /bin/sh -c 'echo hello world'.