pg17

A mirror of the upstream pgvector/pgvector:pg17 image, republished to ghcr.io/vm0-ai/pg17 so CI does not depend on anonymous Docker Hub pulls.

Why

vm0-ai/vm0 runs the API test shards against a pgvector/pgvector:pg17 service container. Anonymous Docker Hub pulls from GitHub-hosted runners fail intermittently, and because the failure happens during Initialize containers the job dies before a single test runs. Two examples on 2026-07-25 alone:

• run 30157827913 - test-api (4), three registry-1.docker.io timeouts
• run 30162893832 - test-api (3), same failure, ejected a PR from the merge queue

The toolchain images are already served from GHCR and pull reliably in the same jobs, so mirroring removes the last Docker Hub dependency from the critical path.

What gets published

.github/workflows/mirror.yml resolves the upstream tag to a digest, copies the whole multi-arch index (linux/amd64 and linux/arm64) with docker buildx imagetools create, then verifies the mirrored digest matches upstream. Nothing is rebuilt, so the bits are identical to what Docker Hub serves.

Tags published:

• ghcr.io/vm0-ai/pg17:pg17 - tracks upstream pg17
• ghcr.io/vm0-ai/pg17:pg17-<YYYYMMDD> - immutable snapshot of each mirror run

The workflow runs weekly and on manual dispatch. Use the source_tag input to mirror a different upstream tag.

Usage

services:
  postgres:
    image: ghcr.io/vm0-ai/pg17:pg17