Single service comparison

On Google Cloud, Artifact Registry is one service covering both container images and package formats; on Huawei Cloud the same scope is split across two services with distinct control planes. CodeArts Artifact manages Maven, npm, PyPI, and other build dependencies tied to the CodeArts DevOps toolchain, while SoftWare Repository for Container (SWR) handles Docker/OCI images with Docker Registry V2 APIs. There is no single Huawei API surface or repository object that mirrors Artifact Repository's per-project multi-format model, so IAM scopes, repository lifecycle, and quota units differ per target.

Artifact Registry exposes a unified gcloud/REST API and IAM roles (roles/artifactRegistry.reader, writer, admin) over repositories scoped to a project and region, with CMEK, VPC Service Controls perimeters, and Pub/Sub change notifications. CodeArts Artifact access is governed through CodeArts project memberships and tenant-level permissions rather than standalone IAM roles, and SWR uses organization Namespace-level RBAC tied to the Huawei Cloud account. Cleanup policies, virtual/remote repository aggregation, and artifact metadata attachments in Artifact Registry have no direct one-to-one counterparts in either Huawei service and must be reconstructed with policy and pipeline tooling.

Scaling and HA behavior also diverge. Artifact Registry is a regional fully managed service with built-in replication and image prewarming for GKE. SWR stores images on OBS-backed highly reliable media (11-nines durability) and accelerates pulls with P2P, but cross-region image sync is configured manually rather than as native multi-region replication. CodeArts Artifact storage is tied to the CodeArts tenant/region and is intended for build pipelines rather than high-throughput global pull traffic. Recalculate retention, replication, and pull-latency expectations against each target rather than assuming parity.

Split the source inventory before migrating: classify each Artifact Registry repository by format. Container, Helm, and OCI artifacts map to SWR; Maven, npm, PyPI, and other build dependencies map to CodeArts Artifact. Do not treat the two Huawei services as interchangeable or expect a single one-click migration path. Confirm supported formats and regional availability for each target in the planned Huawei region, and size SWR namespaces and CodeArts Artifact quotas against current storage volume and pull rate.

For container images, use the documented SWR migration workflow: pull images from Artifact Registry (docker pull), retag to the SWR registry endpoint after creating an organization and obtaining an SWR login token, and push (docker push). For Helm and non-Docker package formats, configure build pipelines (CodeArts Pipeline or external CI) to publish to CodeArts Artifact, re-pointing repository URLs and credentials. Remote/virtual repository aggregation and cleanup policies from Artifact Registry have no native Huawei equivalent; recreate caching via CodeArts Artifact upstream configuration or pipeline-managed retention.

Re-wire all consumers to the new endpoints before cutover: GKE-style workloads move to CCE with image references changed to SWR; CI steps publishing to Artifact Registry move to CodeArts Artifact or SWR; IAM-driven access is re-mapped to CodeArts project membership and SWR Namespace RBAC. Validate signatures, vulnerability scanning, and promotion gates separately, since Artifact Analysis and CMEK-backed encryption do not map directly to SWR or CodeArts Artifact features and may require additional services such as CodeArts Check or a dedicated scanning tool.

Cost model changes: Artifact Registry bills for storage and network egress per region, with separate charges for vulnerability scanning. Huawei typically bills SWR by storage usage and pull bandwidth/requests, and CodeArts Artifact as part of CodeArts subscription tiers plus storage. Recalculate total cost of ownership using peak pull volume, retention period, cross-region sync traffic, and any added security tooling, since direct feature parity is not guaranteed and the split-target model adds operational overhead.