feat(mem-analysis): Adding Dockerfile_with_heaptrack

[Ivansete-status]

Issue

#1662

Description

Adding a Dockerfile that allows to generate a Docker image with both waku and heaptrack. The generated docker containers will run a wakunode2 instance underneath heaptrack.

Once the container is running, it will generate periodic mem-allocation stats in the container's / folder with the name heaptrack.wakunode.1.gz. If the container is restarted, then the previous allocation file is lost so it is important to copy out the file before the container restarts or crashes.

Heaptrack is the tool used and it has two operational modes. In this case, the 'attach' mode makes the underlying program to crash (SYGSEGV). I tried this with a simple C program and it happens the same so it seems to be a Docker-related issue. On the other hand, when running wakunode throughout heaptrack, the stats are properly written periodically.

This docker file also enforces nimbus-build-system to use the nim compiler that points at the heaptrack_support branch. This is achieved by passing NIM_COMMIT=heaptrack_support to the make wakunode2 call.

How to use it

0. Go to the nwaku root folder.
1. Create a docker image with a command like the next:
   sudo make docker-image DOCKER_IMAGE_NAME=docker_repo:docker_tag HEAPTRACKER=1
2. Run the docker container as usual, giving the needed wakunode2 params.
3. Wait for some minutes until the mem-stats file gets populated.
4. copy out the stats file: sudo docker cp 768e7de52d3c:/heaptrack.wakunode.1.gz . (replace 768.. with your docker container id.).
5. Analyse the file with: bin/heaptrack heaptrack.wakunode.1.gz (Heaptrack should be built locally beforehand.).

Additional hints

#1682 (comment)

[LNSD]

I think this line can be removed now :)

[Ivansete-status]

I've applied @vpavlin's recommendation and now we only install libunwind (apk add libunwind). Thanks!

[vpavlin]

Suggested a change, but otherwise LGTM

[vpavlin]

It would be good to only call apk once

Suggested change

	RUN apk add --no-cache gdb
	RUN apk add --no-cache gdb libunwind

[Ivansete-status]

Good point! I'll do that thanks !

[vpavlin]

Suggested change

RUN apk add libunwind

[jm-clius]

Thanks for this! No strong opinion, but since we'll still have to add -d:heaptrack or -d:heaptrack -d:heaptrack_inject to the [make command] I think we at least need another flag/ARG for HEAPTRACK or DEBUG. Since this could pollute the Dockerfile, wouldn't an alternative be to create a new Dockerfile specifically for these debug builds under say /tools/nwaku-heaptrack ? cc @vpavlin here for advice - not sure what standard/best practice is here re number of Dockerfiles: opting for separate files for different use cases or a single Dockerfile, but with flags/ARGs for the different build scenarios?

[vpavlin]

If the Dockerfiles are essentially copies of each other, then that feels like a bad practice because we may forget to update one or another and then we potentially get different images, hence different behaviour of waku node. For such case it would make sense to "polute" the Dockerfile with more ARGs and instruct Makefile.

If one Dockerfile is just an exetension of another (i.e. building on top of another image build) or is separate then it makes sense to have separate Dockerfiles to keep the main one as clean as possible.

I was not aware we need to add -d:heaptrack to the wakunode build based on @Ivansete-status comments

[jm-clius]

If the Dockerfiles are essentially copies of each other, then that feels like a bad practice

Good point and thanks. I think proper use of ARGs here in the same Dockerfile is then the way to go.

[Ivansete-status]

Actually I was preparing the next commit by adding a combination of new 'heaptrack' args to the Dockerfile plus Makefile.
I'll commit once I confirm it works.

[vpavlin]

I don't think the ARG shoul have the suffix ARG:)

[Ivansete-status]

ah yes, I change it now. thanks :)

[vpavlin]

Would this result in NIM_COMMIT having a value NIM_COMMIT=heaptrack_support - i.e. trying to checkout branch NIM_COMMIT=heaptrack_support, rather than heaptrack_support?

[Ivansete-status]

ah that may be happening. For that reason I used the NIM_COMMIT_ARG previously

[Ivansete-status]

Ok, I've applied your suggestions and I think it now looks much better. Thanks @vpavlin !

[vpavlin]

LGTM!