Alibaba Cloud Integration#

Alibaba Cloud (Aliyun) integration in cb-tumblebug operates on two layers:

1. cb-spider adapter — the standard path used for most VM lifecycle operations.
2. Direct ECS SDK layer (src/core/csp/alibaba/) — bypasses cb-spider for spec discovery, pricing, availability checks, VM status polling, and image resolution. This layer was introduced because the cb-spider path for Alibaba has known gaps in these areas (e.g., pricing does a costly disk-category sweep; image IDs deprecate rapidly).

The direct layer is organized into four files under src/core/csp/alibaba/:

All four files register themselves into the cb-tumblebug provider registry via init() hooks and call the shared getAlibabaCreds(ctx) helper to retrieve credentials from OpenBao.

Credentials & Connection Config#

Alibaba credentials use two fields mapped in src/core/csp/csp.go:

At runtime, all direct-SDK code calls getAlibabaCreds(ctx) which reads the secret from OpenBao via csp.BuildSecretPath(ctx, "alibaba"). The context may carry a CredentialHolder (set from connConfig.CredentialHolder) that scopes the secret path to a specific tenant 1.

Connection configs reference Alibaba via providerName = "alibaba" and carry RegionDetail.RegionName and RegionZoneInfo.AssignedZone, both of which are passed directly into the ECS SDK client on every direct API call.

Supported Regions#

The canonical multi-region reference is init/templates/infra-alibaba-global.json, which defines a "1 VM per region" test infra across 12 regions:

Spec ID format: alibaba+{region}+{cspSpecName}, e.g. alibaba+ap-northeast-1+ecs.e-c4m1.large 2.

Note that cn-* China Mainland regions and cn-*-lr Local Regions may require separate Alibaba account activation and may have different instance type availability from global regions.

Direct API Integration (Bypassing CB-Spider)#

Spec Discovery#

For Alibaba connections, spec.go calls alibabaPricing.FetchAvailableSpecListByRegion(ctx, region, zone) instead of the standard LookupSpecList (cb-spider) path 3. This function (defined in price.go:356-455):

1. Calls DescribeAvailableResource(DestinationResource="InstanceType", Status=Available) to get the zone-specific available instance type list.
2. Calls DescribeInstanceTypes (paginated, 100 per page) to enrich each available type with vCPU, memory, local disk, and GPU details.
3. Returns a SpiderSpecList sorted by name.

Pricing#

FetchNodePricesByRegion (and its filtered variant FetchNodePricesByRegionFiltered) fetch on-demand ECS prices per instance type:

• Calls DescribeInstanceTypes to enumerate types, then dispatches concurrent workers (default 5) each calling DescribePrice per type.
• Extracts only the instanceType component from DetailInfos to get cost/hour; falls back to TradePrice if detail is absent 4.
• When a region requires an explicit disk category (error InvalidSystemDiskCategory.ValueNotSupported), retries with cloud_essd then cloud_efficiency 5.
• Currency defaults to CNY when the API response omits it 6.

Tuning env vars 7:

Availability Checking#

availabilityChecker registers via init() and implements the csp.AvailabilityChecker interface. CheckInstance calls DescribeAvailableResource(DestinationResource="SystemDisk") to build a per-zone availability matrix showing which system-disk categories have stock 8.

A zone is considered available when its status is "Available" and it has at least one supported disk category. This proactively catches the most common Alibaba VM-creation failure: No AvailableSystemDisk for instance type in region (a zone+disk+instanceType stock issue that spec catalog lookups cannot detect) 9.

VM Status Polling#

BatchDescribeInstanceStatuses registered via init() 10:

• Accepts up to N instance IDs and batches them into groups of 100 for DescribeInstances.
• Maps Alibaba ECS states to TB status strings 11:

Image Family Resolution#

Alibaba date-stamped image IDs (e.g. ubuntu_22_04_x64_20G_alibase_20260615.vhd) are deprecated rapidly, causing stored image assets to break at VM-creation time. The solution: each Alibaba public image carries a stable ImageFamily tag (e.g. acs:ubuntu_22_04_x64).

ResolveLatestIdByFamily(ctx, region, family) in image.go:

• Calls DescribeImageFromFamily to get the current latest image ID for a family in a region.
• Caches results for 10 minutes 12 with singleflight deduplication to prevent API burst during parallel VM creation 13.
• A "no image in family" result is a soft miss (returns empty ID, caller falls back to original) 14.

At VM-creation time, resolveLatestAlibabaImage in src/core/resource/image.go extracts the ImageFamily from stored imageInfo.Details and calls this resolver 15.

Infrastructure Template#

init/templates/infra-alibaba-global.json is the reference multi-region deployment template. It provisions one ECS instance per region across all 12 supported Alibaba regions 16, using:

• Image: ubuntu_22_04_x64_20G_alibase_20260615.vhd (Ubuntu 22.04; at runtime the image family resolver replaces this with the latest build)
• Spec: region-appropriate ECS types, e.g. ecs.e-c4m1.large for most global regions, ecs.t6-c4m1.large for some China/ME regions, and ecs.c9i.large for newer regions like cn-wuhan-lr and na-south-1

Use this template as a starting point for global Alibaba test deployments or as a reference for spec/image IDs per region.