fix: build-push-docker-manifest polling

[erikburt]

This adds polling to manifest, before trying to get the digest from ECR.

Changes

• Reverts: 321b3dc (feat: build-push-docker-manifest debug logs #1564)
• Reverts: 8fc19e0 (feat: manifest debug for reusable-docker-build-publish #1566)
• Adds fault-tolerant polling step for fetching the manifest digest

Notes

Seeing exit code 255 during manifest push step. Example: https://github.com/smartcontractkit/chainlink/actions/runs/26590320161/job/78349080233?pr=22679

[77](https://github.com/smartcontractkit/chainlink/actions/runs/26590320161/job/78349080233?pr=22679#step:4:494)
#1 0.000 copying sha256:2531e76391220f8e6ab3d252722f30143ea2f220cf53472633372303bff1aed6 from ***.dkr.ecr.us-west-2.amazonaws.com/chainlink@sha256:db57e3d3a3ecc538a0cbb810da9860179bb1c6e1abfd1476b1bba7a8404ca9e0 to ***.dkr.ecr.us-west-2.amazonaws.com/chainlink
#1 1.306 pushing sha256:3ee95908cced393678e9214d8325c66e40e7ee08d5506cc9f4ee055d42648726 to ***.dkr.ecr.us-west-2.amazonaws.com/chainlink:pr-22679-b2b70d0-plugins-testing
#1 DONE 1.7s
Error: Process completed with exit code 255.

• When we push the arch specific images, and before we push the manifest index, we poll for the images to ensure that the manifest is valid.

  • .github/actions/build-push-docker-manifest/action.yml

    Lines 259 to 316 in 644aaa0

    	- name: Wait for source images to be available
    	shell: bash
    	env:
    	DOCKER_MANIFEST_NAME: ${{ steps.manifest-name.outputs.name }}
    	DOCKER_IMAGE_NAME_DIGESTS: ${{ inputs.docker-image-name-digests }}
    	DOCKER_IMAGE_AVAILABILITY_CHECK:
    	${{ inputs.docker-image-availability-check }}
    	run: |
    	# Convert comma-separated list into array
    	IFS=',' read -ra DIGESTS <<< "$DOCKER_IMAGE_NAME_DIGESTS"
    	if [[ "${DOCKER_IMAGE_AVAILABILITY_CHECK}" == "true" ]]; then
    	echo "Checking image availability before creating manifest..."
    	MAX_RETRIES=5
    	RETRY_DELAY=10
    	ALL_AVAILABLE=false
    	for attempt in $(seq 1 $MAX_RETRIES); do
    	echo "Attempt ${attempt}/${MAX_RETRIES}: Checking image availability..."
    	MISSING_IMAGES=()
    	for digest in "${DIGESTS[@]}"; do
    	# Trim whitespace from digest
    	digest=$(echo "$digest" | xargs)
    	IMAGE_WITH_DIGEST="${DOCKER_MANIFEST_NAME}@${digest}"
    	if docker buildx imagetools inspect "${IMAGE_WITH_DIGEST}" >/dev/null 2>&1; then
    	echo " ✓ ${digest} is available"
    	else
    	echo " ⏳ ${digest} not yet available"
    	MISSING_IMAGES+=("${digest}")
    	fi
    	done
    	if [[ ${#MISSING_IMAGES[@]} -eq 0 ]]; then
    	echo "✅ All images are available!"
    	ALL_AVAILABLE=true
    	break
    	else
    	echo "❌ ${#MISSING_IMAGES[@]} image(s) still missing: ${MISSING_IMAGES[*]}"
    	if [[ $attempt -lt $MAX_RETRIES ]]; then
    	echo "Waiting ${RETRY_DELAY} seconds before next attempt..."
    	sleep $RETRY_DELAY
    	fi
    	fi
    	done
    	if [[ "$ALL_AVAILABLE" != "true" ]]; then
    	echo "::error::Failed to verify image availability after ${MAX_RETRIES} attempts"
    	echo "::error::Missing images: ${MISSING_IMAGES[*]}"
    	exit 1
    	fi
    	else
    	echo "Image availability check disabled, using fixed delay..."
    	echo "Waiting 30 seconds for source images to be fully available..."
    	sleep 30
    	fi

• But when we pushed the manifest, we immediately queried the ECR for the digest which I believe is why we were seeing errors

• .github/actions/build-push-docker-manifest/action.yml

  Lines 370 to 374 in 644aaa0

  	# Execute the command
  	docker buildx imagetools create "${CMD_ARGS[@]}"
  	# Get manifest digest (format: sha256:hash)
  	MANIFEST_DIGEST=$(docker buildx imagetools inspect "${DOCKER_MANIFEST_NAME_WITH_TAG}" | grep -m1 'Digest:' | awk '{print $2}')

• This fixes that issue by similarly polling for the manifest, ensuring it's existence, and safely extracting the digest

Testing

• Created branch pointing at this one: https://github.com/smartcontractkit/.github/compare/main...test%2Fdocker-manifest-polling?assignees=erikburt&body=&expand=1
• Update CL workflows: test: manifest polling chainlink#22681
  • https://github.com/smartcontractkit/chainlink/actions/runs/26596265155/job/78369339402?pr=22681#step:4:533

Attempt 1/5: Inspecting manifest (***.dkr.ecr.us-west-2.amazonaws.com/chainlink:pr-22681-8bb010f) to retrieve digest...
Successfully retrieved manifest digest on attempt 1: sha256:6008273de81214df86c3a3e809167a32c60fd00827be1b08bacfd970c133ad41
manifest-digest=sha256:6008273de81214df86c3a3e809167a32c60fd00827be1b08bacfd970c133ad41
manifest-name=***.dkr.ecr.us-west-2.amazonaws.com/chainlink
manifest-name-with-digest=***.dkr.ecr.us-west-2.amazonaws.com/chainlink@sha256:6008273de81214df86c3a3e809167a32c60fd00827be1b08bacfd970c133ad41
manifest-name-with-tag=***.dkr.ecr.us-west-2.amazonaws.com/chainlink:pr-22681-8bb010f

^ This at least proves there's no regression. We will see if it fixes the issue that actually caused the 255 exits.