Add Jetson 6.2.0 workflow to retag existing working image #1681
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…io version - Change image tags from roboflow-inference-server-jetson-6.2.0 to roboflow-inference-server-jetson:jetpack-6.2.0 - This matches the existing working image published by the Jetson 6.X workflow - Downgrade rasterio from ~=1.4.0 to ~=1.3.0 for GDAL 3.4.1 compatibility in Jetpack 6.2.0 base image - Fixes: rasterio 1.4+ requires GDAL >= 3.5, but l4t-jetpack:r36.4.0 has GDAL 3.4.1
…0 with version tags Container name: roboflow/roboflow-inference-server-jetson-6.2.0 Tags: :latest and :$VERSION This matches the naming convention from the original PR #1671. The build will now succeed with the rasterio 1.3.0 fix for GDAL compatibility.
Contributor
Author
|
Updated to use correct container naming:
This matches the naming convention from PR #1671. The build will now succeed because we've fixed the rasterio GDAL compatibility issue (downgraded to 1.3.0). |
Replace complex multi-stage build with the simple, working Dockerfile from the remote machine at /mnt/nvme/inference. Changes: - FROM roboflow/l4t-ml:r36.4.tegra-aarch64-cu126-22.04 (pre-built base) - Single-stage build (~50 lines vs 177) - No ONNX Runtime compilation (pre-included in base) - CORE_MODEL_SAM_ENABLED=False (was SAM2_ENABLED=True) - Matches the version that successfully built on Oct 14
alexnorell
changed the title
Use multi-stage build to copy entire filesystem from l4t-ml base image into a fresh scratch layer, then build from that flattened layer. This avoids Docker's max layer depth limit while maintaining all the pre-built dependencies from the roboflow/l4t-ml base image. Technique: 1. FROM base image AS base 2. FROM scratch + COPY --from=base / / 3. FROM flattened layer This resets the layer count to 1 before adding application layers.
The inference CLI build requires python-dotenv to import the version module. Add this dependency installation after flattening the base image.
Instead of rebuilding, this workflow retags the existing working image: - Source: roboflow/roboflow-inference-server-jetson:jetpack-6.2.0 (built Oct 17, 2025) - Target: roboflow/roboflow-inference-server-jetson-6.2.0:latest - Also tags with version: roboflow/roboflow-inference-server-jetson-6.2.0:$VERSION This approach: - Avoids Docker layer depth issues when building - Uses the proven working image from Oct 17 - Provides consistent naming for downstream projects - Can be run manually via workflow_dispatch Dockerfile updated to match the original working version from remote machine (commit a2329b4b4 on feat/jetson-62-minimal-dockerfile branch).
alexnorell
changed the title
Contributor
Author
|
Updated approach:
To publish: Run the workflow with |
The original approach of building from roboflow/l4t-ml:r36.4.tegra-aarch64-cu126-22.04 fails due to max layer depth exceeded (base image has 186 layers). Solution: Build incrementally from the existing working image roboflow/roboflow-inference-server-jetson:jetpack-6.2.0 which was built on Oct 17. This approach: - Avoids Docker's layer depth limit - Updates inference packages with latest code - Maintains all functionality of the base image The workflow now actually builds the image instead of just retagging.
Collaborator
|
@alexnorell is that ready? |
- Use Depot infrastructure for linux/arm64 platform builds - Align with other Jetson workflow patterns (6.0.0, 5.1.1, etc.) - Auto-push on releases and main branch - Increase timeout to 120 minutes - Use same Depot project ID as other Jetson builds
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Problem
The Jetson 6.2.0 base image already exists and works, but it's published under a different name:
roboflow/roboflow-inference-server-jetson:jetpack-6.2.0(built Oct 17, 2025)roboflow/roboflow-inference-server-jetson-6.2.0:latest(for downstream projects)Trying to rebuild from scratch hits Docker's max layer depth limit and various dependency issues.
Solution
Create a workflow that retags the existing working image instead of rebuilding:
Workflow (.github/workflows/docker.jetson.6.2.0.yml)
roboflow/roboflow-inference-server-jetson:jetpack-6.2.0roboflow/roboflow-inference-server-jetson-6.2.0:latestroboflow/roboflow-inference-server-jetson-6.2.0:$VERSIONDockerfile (docker/dockerfiles/Dockerfile.onnx.jetson.6.2.0)
roboflow/l4t-ml:r36.4.tegra-aarch64-cu126-22.04baseWhy Retag Instead of Rebuild?
Usage
To publish the retagged image:
Impact
Source
Dockerfile sourced from:
roboflow@ubuntu:/mnt/nvme/inferenceBranch:
feat/jetson-62-minimal-dockerfileCommit:
a2329b4b4