[Feature] Dockerfile Optimization for OPEA v1.3

[joshuayao]

Priority

P1-Stopper

OS type

Ubuntu

Hardware type

Xeon-GNR

Running nodes

Single Node

Description

Streamline Docker image builds for GenAIExamples and GenAIComps(stretch goal, targeting v1.4) by introducing hardware-aware base images and update CI/CD pipelines to fit these changes.

Base Images

We need to create specialized base images for each hardware platform to ensure optimal performance and consistency across our containers. Leveraging dedicated base images simplifies image creation, guarantees uniform runtime environments, and takes full advantage of Docker’s layer caching for faster builds and smaller image sizes. We will focus on CPU/common components in v1.3. The required base images are:

• CPU Image for examples:

• Add Dockerfile for comps-base image GenAIComps#1127 - owner: @eero-t

• Analysis of Comps container's image dependencies and potential base images for them - owner: @eero-t - stretch goal, targeting v1.4?

Standardizing Dockerfiles in GenAIExamples and GenAIComps(stretch goal, targeting v1.4) with Base Image Inheritance and Multi-Stage Builds

Update all Dockerfiles in the GenAIExamples repositories to extend pre-built base images and leverage multi-stage builds. This initiative will promote consistency, enhance readability, reduce final image sizes, accelerate image builds.

• GenAIExamples:

• Use GenAIComp base image to simplify Dockerfiles & reduce image sizes #1369, owner: @eero-t

• Use GenAIComp base image to simplify Dockerfiles - part 1 #1612, owner: @eero-t

• Use GenAIComp base image to simplify Dockerfiles - part 2 #1638, owner: @eero-t

• Use GenAIComp base image to simplify Dockerfiles - part 3/4 #1671, owner: @eero-t

• GenAIComps (moved to v1.4)

• [Feature] Use base image to simplify Dockerfiles & reduce image sizes GenAIComps#1326

CI/CD Changes

• [CI/CD] Enabling building base image in CI GenAIComps#1314, owner: @chensuyue
• Enable base image build in CI/CD #1669, owner: @chensuyue

[eero-t]

@joshuayao please add link to this PR also: #1612

(Due to large number of triggered CI tests, how long they take to run, and how many of them fail for unrelated reasons, I've had to split the PR into smaller pieces.)

We need to create specialized base images for each hardware platform to ensure optimal performance and consistency across our containers.

The linked PRs just move all the duplicated content from 18 app Dockerfiles to a shared base image (to greatly simplify them, and to reduce their build time & disk usage). It's a generic GenAIComps base image, not HW specific.

In earlier phase (1.2 release), those Dockerfiles were optimized and unified, to halve their image sizes and to prove that they indeed were almost identical duplicates, see #1031.

(OIM images will be HW specific, but they are for backend services, whereas this is for frontend ones, i.e. unrelated.)

[joshuayao]

@joshuayao please add link to this PR also: #1612

(Due to large number of triggered CI tests, how long they take to run, and how many of them fail for unrelated reasons, I've had to split the PR into smaller pieces.)

We need to create specialized base images for each hardware platform to ensure optimal performance and consistency across our containers.

The linked PRs just move all the duplicated content from 18 app Dockerfiles to a shared base image (to greatly simplify them, and to reduce their build time & disk usage). It's a generic GenAIComps base image, not HW specific.

In earlier phase (1.2 release), those Dockerfiles were optimized and unified, to halve their image sizes and to prove that they indeed were almost identical duplicates, see #1031.

(OIM images will be HW specific, but they are for backend services, whereas this is for frontend ones, i.e. unrelated.)

Hi @eero-t. Thanks for your information. Added #1612.

This issue applies to both components and examples. Regarding the base image for components, while many of them are simply wrappers and not hardware-aware, some local service providers may require hardware-aware implementations. Given this, I added hardware-specific base image requirements. But for examples, I agree, no HW specific base images required. I’ve marked component support as a stretch goal. Should we consider removing it from the v1.3 release?