Add entrypoint container images #819

rhatdan · 2025-02-14T15:13:41Z

Install podman-remote and ramalama so we can use ramalama
from within a container.

$ podman run --env XDG_RUNTIME_DIR -v $HOME:$HOME -v /run/user:/run/user --userns=keep-id -ti --privileged quay.io/ramalama/ramalama ramalama run granite

Signed-off-by: Daniel J Walsh dwalsh@redhat.com

sourcery-ai · 2025-02-14T15:13:46Z

Reviewer's Guide by Sourcery

This pull request introduces functionality to add entrypoints to container images for whisper-server and llama-server. It adds the add_entrypoint and add_entrypoints functions to handle the creation of temporary Containerfiles and image building. The changes integrate these functions into the existing build process for both single and multi-architecture images.

Sequence diagram for adding entrypoints to container images

sequenceDiagram
  participant CB as container_build.sh
  participant A as add_entrypoints()
  participant AE as add_entrypoint()
  participant P as podman/conman

  CB->>A: build()
  A->>AE: add_entrypoint "whisper-server"
  AE->>P: Build image with entrypoint
  P-->>AE: Image built
  A->>AE: add_entrypoint "llama-server"
  AE->>P: Build image with entrypoint
  P-->>AE: Image built
  AE-->>A: Entrypoints added
  A-->>CB: Complete

File-Level Changes

Change	Details	Files
Introduces a function to add entrypoints to container images.	Introduces `add_entrypoint` function to create a temporary Containerfile, define an entrypoint, build the image, and then remove the temporary file. Introduces `add_entrypoints` function to add whisper-server and llama-server entrypoints.	`container_build.sh`
Integrates the new entrypoint functionality into the image building process.	Calls `add_entrypoints` after building a single-architecture container image. Calls `add_entrypoints` after building a multi-architecture container image.	`container_build.sh`

Tips and commands

Interacting with Sourcery

Trigger a new review: Comment @sourcery-ai review on the pull request.
Continue discussions: Reply directly to Sourcery's review comments.
Generate a GitHub issue from a review comment: Ask Sourcery to create an
issue from a review comment by replying to it. You can also reply to a
review comment with @sourcery-ai issue to create an issue from it.
Generate a pull request title: Write @sourcery-ai anywhere in the pull
request title to generate a title at any time. You can also comment
@sourcery-ai title on the pull request to (re-)generate the title at any time.
Generate a pull request summary: Write @sourcery-ai summary anywhere in
the pull request body to generate a PR summary at any time exactly where you
want it. You can also comment @sourcery-ai summary on the pull request to
(re-)generate the summary at any time.
Generate reviewer's guide: Comment @sourcery-ai guide on the pull
request to (re-)generate the reviewer's guide at any time.
Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
pull request to resolve all Sourcery comments. Useful if you've already
addressed all the comments and don't want to see them anymore.
Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
request to dismiss all existing Sourcery reviews. Especially useful if you
want to start fresh with a new review - don't forget to comment
@sourcery-ai review to trigger a new review!
Generate a plan of action for an issue: Comment @sourcery-ai plan on
an issue to generate a plan of action for it.

Customizing Your Experience

Access your dashboard to:

Enable or disable review features such as the Sourcery-generated pull request
summary, the reviewer's guide, and others.
Change the review language.
Add, remove or edit custom review instructions.
Adjust other review settings.

Getting Help

Contact our support team for questions or feedback.
Visit our documentation for detailed guides and information.
Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

rhatdan · 2025-02-14T15:13:54Z

@benoitf PTAL

sourcery-ai

Hey @rhatdan - I've reviewed your changes - here's some feedback:

Overall Comments:

Consider using a single add_entrypoint call with a loop inside the function to reduce code duplication.
It might be better to pass the entrypoint script name as an argument to add_entrypoints.

Here's what I looked at during the review

🟡 General issues: 1 issue found
🟢 Security: all looks good
🟢 Testing: all looks good
🟢 Complexity: all looks good
🟢 Documentation: all looks good

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

sourcery-ai · 2025-02-14T15:14:57Z

container_build.sh

+add_entrypoint() {
+    containerfile=$(mktemp)
+    cat > ${containerfile} <<EOF
+FROM $2
+ENTRYPOINT [ "/usr/bin/$3.sh" ]
+EOF
+    eval $1 --no-cache -t "$2-$3" -f ${containerfile} .
+    rm ${containerfile}


suggestion (bug_risk): Consider ensuring cleanup of temporary files on errors.

If the eval or the build command fails, the temporary file created by mktemp may remain. Using a trap to remove the file on error would improve the robustness of this function.

benoitf · 2025-02-14T15:47:03Z

container_build.sh

+FROM $2
+ENTRYPOINT [ "/usr/bin/$3.sh" ]
+EOF
+    eval $1 --no-cache -t "$2-$3" -f ${containerfile} .


on my end it's failing on --no-cache parameter not found in podman CLI

it seems $1 $2 $3 are not the expected parameters
1 = podman
2 = build
3 = whisper-server

benoitf · 2025-02-14T16:54:18Z

so changing the order of the arguments in the script, I was able to build an image and start it using

podman run --rm -it --env=HOME=/tmp --cap-drop=all --security-opt=no-new-privileges --init -v $HOME/.local/share/containers/podman-desktop/extensions-storage/redhat.ai-lab/models:/tmp/models -p 9000:9000 --device /dev/dri -e PORT=9000 -e MODEL_PATH=/tmp/models/hf.bartowski.granite-3.1-8b-instruct-GGUF/granite-3.1-8b-instruct-Q4_K_M.gguf ramalama-llama-server:latest

and was able to use curl to query the model and it worked 👍

 curl --location 'http://localhost:9000/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data '{
  "messages": [
    {
      "content": "You are a helpful assistant.",
      "role": "system"
    },
    {
      "content": "What is the capital of France?",
      "role": "user"
    }
  ]
}'

{"choices":[{"finish_reason":"stop","index":0,"message":{"content":"The capital of France is Paris.\n\....

Install podman-remote and ramalama so we can use ramalama from within a container. $ podman run --env XDG_RUNTIME_DIR -v $HOME:$HOME -v /run/user:/run/user --userns=keep-id -ti --privileged quay.io/ramalama/ramalama ramalama run granite Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>

rhatdan · 2025-02-14T19:35:03Z

@benoitf This ramalama image now contains the upstream version of RamaLama along with podman-remote, so we should be able to run RamaLama in this container.

With a command like:

$ podman run --env XDG_RUNTIME_DIR -v $HOME:$HOME -v /run/user:/run/user --userns=keep-id -ti --privileged quay.io/ramalama/ramalama ramalama run granite

benoitf

tried the dedicated images

I think there is an issue in the script (not related to this PR)

I provided -e MODEL_CHAT_FORMAT="llama-2" and got a core dumped about a missing --jinja

podman run --rm -it --env=HOME=/tmp --cap-drop=all --security-opt=no-new-privileges --init --pull=newer -v $HOME/.local/share/containers/podman-desktop/extensions-storage/redhat.ai-lab/models:/tmp/models -p 9000:9000 --device /dev/dri -e PORT=9000 -e HOST=0.0.0.0 -e MODEL_PATH=/tmp/models/hf.bartowski.granite-3.1-8b-instruct-GGUF/granite-3.1-8b-instruct-Q4_K_M.gguf -e MODEL_CHAT_FORMAT="llama-2" quay.io/ramalama/ramalama-llama-server:latest
terminate called after throwing an instance of 'std::runtime_error'
  what():  error: the supplied chat template is not supported: llama-2
note: llama.cpp was started without --jinja, we only support commonly used templates

/usr/bin/llama-server.sh: line 14:     3 Aborted                 (core dumped) llama-server --model ${MODEL_PATH} --host ${HOST:=0.0.0.0} --port ${PORT:=8001} --gpu_layers ${GPU_LAYERS:=0} ${CHAT_FORMAT}

it seems that llama.cpp server accept "llama2" and crash with "llama-2"

but llamacpp python is using llama-2
https://github.com/abetlen/llama-cpp-python?tab=readme-ov-file#chat-completion

I'm wondering if we could for example replace "llama-2" by "llama2" in the llama-sever script to have a better backward compatibility

ericcurtin · 2025-02-14T22:23:29Z

container-images/scripts/build_llama_and_whisper.sh

+  # link podman-remote to podman for use by RamaLama
+  ln -sf /usr/bin/podman-remote /usr/bin/podman
+  git clone https://github.com/containers/ramalama
+  cd ramalama


How about a pip install? At least it wouldn't be installing from main branch, a more unstable target...

could it just copy the files from where we are.
Like if I change some scripts I would assume it's embedding the files from my current working directory (can be my changes, the tag, etc) rather than the latest stable or main branch

Also if I checkout 2 months after it would use the same files (the one stored in the current working copy)

We have a chicken and egg situation:

Currently we are building the images at the same time as the release. Juggling the two at the same time is going to be difficult.

When I build the 0.7.0 image, I usually do this before cutting the release. Installing the older version say 0.6.0 into the 0.7.0 image seems wrong. I don't think that is the proper solution.

Building a users home checked out version is also very risky, because that version may never be released to github period. We could change the scripts to allow it to install the current version if this is really necessary, but I don't think it is, and users could hack their scripts in themselves.

I've thought about it over the week-end and my conclusion would be that there should be a dedicated image with only the CLI (no any inference implementation like llama.cpp).

ramalama-cli image would be the image for the python code of RamaLama that will the run containers (so with podman-remote packaged inside this image)

And I would not expect to have podman-remote or any python code/dependency in any of the ramalama image that provides llama.cpp

And then the RamaLama cli should also be a smaller image (and the other as well)

ramalama-cli (python code + podman-remote)--> start containers like llama.cpp or whisper.cpp or vLLM images (that only contain the runner)

Building a users home checked out version is also very risky, because that version may never be released to github period

if I do 'make images' I expect that it builds a specific tag like :next or :dev or another one from my local copy.
Then I can test the images locally.

The CI job would publish only for tags or from main branch code so I don't see any issues

I've thought about it over the week-end and my conclusion would be that there should be a dedicated image with only the CLI (no any inference implementation like llama.cpp).

ramalama-cli image would be the image for the python code of RamaLama that will the run containers (so with podman-remote packaged inside this image)

And I would not expect to have podman-remote or any python code/dependency in any of the ramalama image that provides llama.cpp

And then the RamaLama cli should also be a smaller image (and the other as well)

ramalama-cli (python code + podman-remote)--> start containers like llama.cpp or whisper.cpp or vLLM images (that only contain the runner)

This almost works, except for on Windows and now we have to package and maintain python on Windows, not good.

Building a users home checked out version is also very risky, because that version may never be released to github period

if I do 'make images' I expect that it builds a specific tag like :next or :dev or another one from my local copy. Then I can test the images locally.

The CI job would publish only for tags or from main branch code so I don't see any issues

We have a chicken and egg situation:

I think we solve the chicken and egg situation, if we do something like:

COPY ../../* /ramalama/

(maybe even a create an .rpm/tarball during the build script, copy it, install it).

git clone main branch, could be in any sort of random state.

Currently we are building the images at the same time as the release. Juggling the two at the same time is going to be difficult.

When I build the 0.7.0 image, I usually do this before cutting the release. Installing the older version say 0.6.0 into the 0.7.0 image seems wrong. I don't think that is the proper solution.

Building a users home checked out version is also very risky, because that version may never be released to github period. We could change the scripts to allow it to install the current version if this is really necessary, but I don't think it is, and users could hack their scripts in themselves.

rhatdan · 2025-02-17T11:03:55Z

I don't think it is worth managing another Image for adding RamaLama to the base image. Their is a maintainence issue and a pain for every user who wants to use RamaLama within a container. Lets get this merged and then we can debate ways to get the current RamaLama into the container at Runtime.

Adding a -v .:/run/ramalama might be enough. But I would like to get this in for this weeks release.

rhatdan requested review from ericcurtin, bmahabirbu, maxamillion, swarajpande5, jhjaggars, cgruver, slp and engelmi as code owners February 14, 2025 15:13

sourcery-ai bot reviewed Feb 14, 2025

View reviewed changes

benoitf reviewed Feb 14, 2025

View reviewed changes

rhatdan force-pushed the entrypoint branch 2 times, most recently from 314ce8c to 245f6db Compare February 14, 2025 19:26

rhatdan force-pushed the entrypoint branch from 245f6db to 4d209b9 Compare February 14, 2025 19:33

benoitf approved these changes Feb 14, 2025

View reviewed changes

ericcurtin reviewed Feb 14, 2025

View reviewed changes

rhatdan merged commit c6a91af into containers:main Feb 17, 2025
17 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add entrypoint container images #819

Add entrypoint container images #819

rhatdan commented Feb 14, 2025 •

edited

Loading

sourcery-ai bot commented Feb 14, 2025 •

edited

Loading

Interacting with Sourcery

Customizing Your Experience

Getting Help

rhatdan commented Feb 14, 2025

sourcery-ai bot left a comment

sourcery-ai bot Feb 14, 2025

benoitf Feb 14, 2025

benoitf Feb 14, 2025

benoitf commented Feb 14, 2025

rhatdan commented Feb 14, 2025

benoitf left a comment •

edited

Loading

ericcurtin Feb 14, 2025

benoitf Feb 14, 2025

rhatdan Feb 15, 2025

benoitf Feb 16, 2025

ericcurtin Feb 16, 2025 •

edited

Loading

ericcurtin Feb 16, 2025

rhatdan commented Feb 17, 2025

Add entrypoint container images #819

Add entrypoint container images #819

Conversation

rhatdan commented Feb 14, 2025 • edited Loading

sourcery-ai bot commented Feb 14, 2025 • edited Loading

Reviewer's Guide by Sourcery

Sequence diagram for adding entrypoints to container images

File-Level Changes

Interacting with Sourcery

Customizing Your Experience

Getting Help

rhatdan commented Feb 14, 2025

sourcery-ai bot left a comment

Choose a reason for hiding this comment

sourcery-ai bot Feb 14, 2025

Choose a reason for hiding this comment

benoitf Feb 14, 2025

Choose a reason for hiding this comment

benoitf Feb 14, 2025

Choose a reason for hiding this comment

benoitf commented Feb 14, 2025

rhatdan commented Feb 14, 2025

benoitf left a comment • edited Loading

Choose a reason for hiding this comment

ericcurtin Feb 14, 2025

Choose a reason for hiding this comment

benoitf Feb 14, 2025

Choose a reason for hiding this comment

rhatdan Feb 15, 2025

Choose a reason for hiding this comment

benoitf Feb 16, 2025

Choose a reason for hiding this comment

ericcurtin Feb 16, 2025 • edited Loading

Choose a reason for hiding this comment

ericcurtin Feb 16, 2025

Choose a reason for hiding this comment

rhatdan commented Feb 17, 2025

rhatdan commented Feb 14, 2025 •

edited

Loading

sourcery-ai bot commented Feb 14, 2025 •

edited

Loading

benoitf left a comment •

edited

Loading

ericcurtin Feb 16, 2025 •

edited

Loading