-
|
Hi, I tried the following setup:
Dockerfile: FROM dtcooper/raspberrypi-os:bookworm
ENV DEBIAN_FRONTEND=noninteractive
RUN apt update && apt install -y \
python3 \
python3-picamera2 --no-install-recommends
RUN ln -s /usr/bin/python3 /usr/bin/python
WORKDIR /app
COPY . .
CMD ["python", "code/camera.py"]docker-compose.yml: name: picamera2-test
services:
picamera2-test:
build: .
stdin_open: true
tty: true
container_name: picamera2-test
volumes:
- ./code:/code
- /run/udev/:/run/udev:ro
devices:
- /dev/video0:/dev/video0
privileged: true./code/camera.py: from picamera2 import Picamera2
picam2 = Picamera2(0)
picam2.start_and_capture_file("/code/test.jpg", show_preview=False)
print("Captured /code/test.jpg")The error log is: Choosing camera 1 with the lower resolution works fine: I have also tried the solution #954 from @hyzhak replacing his streaming server with my Python code. But this gives the same result. I also took notice from ch. 8.3 of the picamera2-manual but I am currently not able to translate this Docker directives. Any help would be highly appreciated. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 7 replies
-
|
Have you read this: https://docs.docker.com/engine/containers/resource_constraints/ ? |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your hint. The problem here seems to be that an application within the container cannot allocate enough memory provided by the container. The picamera2-manual states in chapter 8.3 that cameras use CMA memory and Therefore, my suspect is that memory blocks with contiguous physical addresses cannot be allocated from containerized applications, unless the container is enabled to use the Contiguous Memory Allocator (CMA). However, I am not sure whether this can be done and how Docker could be instructed to do so. |
Beta Was this translation helpful? Give feedback.
-
|
In fact Pi 5s don't use CMA because they have an I/O MMU (possibly the manual needs an update if it suggests otherwise). But unfortunately I don't know anything about Docker so can't really offer any helpful suggestions. |
Beta Was this translation helpful? Give feedback.
-
|
@davidplowman, thanks for clarification. I tried to analyze this a little bit more. Looking at the available cma on my Pi 5, Since, according to the picamera2-manual ch. 8.3, the 12MP Model 3 image requires 48 MB, this will probably be the bottleneck. I increased cma by replacing with in Now, after reboot, I have: Now, the Doesn't this indicate that also on a Pi 5 CMA is used? |
Beta Was this translation helpful? Give feedback.
-
|
Might be worth posting the output of So there's a Linux CMA heap and a "system" heap. Camera buffers come from "vidbuf_cached", which on a Pi 5 should point to the system heap. On other Pis, it would point to the CMA heap. There was at one point a bug on the Pi 5 which meant that vidbuf_cached was pointing to the wrong one, but that was fixed quite some time ago. |
Beta Was this translation helpful? Give feedback.
-
|
On my Pi 5 it shows |
Beta Was this translation helpful? Give feedback.
-
|
Hi again. Not sure what to suggest there. Your system looks OK, so it's got absolutely no business trying to allocate buffers out of CMA. Whether putting it in a Docker container has some strange effect on this... I'm afraid I couldn't say. So I'm a bit at a loss, I'm afraid! |
Beta Was this translation helpful? Give feedback.
-
|
Hi, |
Beta Was this translation helpful? Give feedback.
@davidplowman, thanks for clarification.
I tried to analyze this a little bit more.
Looking at the available cma on my Pi 5,
cat /proc/meminfoshows:Since, according to the picamera2-manual ch. 8.3, the 12MP Model 3 image requires 48 MB, this will probably be the bottleneck.
I increased cma by replacing
with
in
/boot/firmware/config.txtNow, after reboot, I have:
Now, the
Cannot allocate memoryerror is gone when running the docker container with the Model 3 camera and a 12MP image is created.Doesn't this indicate tha…