Resource-discovery parity: surface the resources cloudemu already emulates (only 7–8 types today)

[thzgajendra]

Summary

cloudemu has a two-layer coverage picture, and the two are badly out of sync:

• Emulation (API) layer — broad. ~23 service categories are implemented across AWS/Azure/GCP with real SDK-compat wire handlers (only 3 known per-cell gaps).
• Discovery (inventory) layer — very narrow. resourcediscovery walks only 5 drivers and emits only 8 resource types total. Every SDK-compat inventory API (Resource Explorer 2, Resource Graph, Cloud Asset) is a 1:1 re-label of those same 8.

So even though RDS/EKS/ELB/ElastiCache/SNS/SQS/Bedrock/SageMaker etc. are emulated, none of them appear when a client enumerates resources via RE2 / Resource Graph / Cloud Asset. The discovery layer should surface the resources cloudemu already models (and, over time, the broader provider catalog).

Evidence: resourcediscovery/walkers.go (6 walkers, 8 type constants L30-39), resourcediscovery/engine.go L102-130; server/aws/resourceexplorer2/handler.go + filter.go; server/azure/resourcegraph/kql.go (mapAzureType) + handler.go; server/gcp/cloudasset/filter.go (mapGCPAssetType).

Resource types surfaced in discovery TODAY (exhaustive)

Engine type	AWS RE2	Azure Resource Graph	GCP Cloud Asset
Instance	compute:instance	microsoft.compute/virtualmachines	compute.googleapis.com/Instance
VPC	networking:vpc	microsoft.network/virtualnetworks	compute.googleapis.com/Network
Subnet	networking:subnet	microsoft.network/subnets	compute.googleapis.com/Subnetwork
SecurityGroup	networking:securitygroup	microsoft.network/networksecuritygroups	compute.googleapis.com/Firewall
Bucket	storage:bucket	microsoft.storage/storageaccounts	storage.googleapis.com/Bucket
Table	database:table	microsoft.documentdb/databaseaccounts	firestore.googleapis.com/Database
Function	serverless:function	microsoft.web/sites	cloudfunctions.googleapis.com/Function
Workspace (Azure only)	—	microsoft.databricks/workspaces	—

Emulated-service coverage (for reference)

All 20 portable categories are implemented for all 3 providers. Specialized categories with per-cell gaps: Bedrock (AWS only), Databricks (Azure only), AI Search (Azure only). Note: statemachine/ is a generic FSM helper, not Step Functions / Workflows / Logic Apps — those are not emulated.

Remaining — prioritized backlog

Workstream A — surface what's already emulated (high-leverage, cheap)

The drivers exist; each needs a walker + an entry in the three type-mapping tables (resourceexplorer2/{handler,filter}.go, resourcegraph/kql.go mapAzureType/portableToAzureType, cloudasset/filter.go mapGCPAssetType/portableToGCPAssetType). Order:

1. Relational DB (all) — RDS/Aurora/DocumentDB/Neptune (instance+cluster+snapshot+proxy), Redshift, Cloud SQL, Azure SQL / postgres-flex / mysql-flex. Driver: relationaldb/.
2. Kubernetes (all) — EKS cluster/nodegroup, GKE cluster/nodepool, AKS managedclusters(+agentpools).
3. Load Balancer (all) — ALB/NLB/GWLB, GCP LB, Azure LB/App Gateway. Driver: loadbalancer/.
4. Cache (all) — ElastiCache, Memorystore/Redis, Azure Cache.
5. Container Registry — ECR, Artifact Registry, ACR.
6. Message Queue + Notification — SQS/PubSub/Service Bus, SNS.
7. Secrets — Secrets Manager / Secret Manager / Key Vault.
8. DNS — Route 53 zones, Cloud DNS ManagedZone, Azure DNS.
9. IAM — user/role/policy/group (note: cloudasset searchAllIamPolicies returns empty today).
10. Networking sub-types — NAT GW / IGW / EIP / VPC peering (AWS); Disk/Address/Router/ForwardingRule (GCP); disks/public-IP/NIC/App-Gateway (Azure).
11. ML / GenAI — SageMaker*, Vertex AI kinds, Azure AI (CognitiveServices + MachineLearningServices), Bedrock*.
12. Monitoring/Logging — CloudWatch alarms/dashboards/logs; Azure Insights/Log Analytics.

Also refine existing mappings: GCP compute Disk/Snapshot are collapsed; Database→Table maps Firestore only to Database (finer kinds expected).

Workstream B — net-new emulation (larger)

• AWS: EBS volume/snapshot, EIP, NAT GW, IGW, VPC peering, EFS, KMS, OpenSearch, MSK, MemoryDB, Kinesis(+analytics), DAX, EMR(+serverless), Glue, DMS, Step Functions, WAF, GuardDuty, CloudTrail, Config, ACM, SES, WorkSpaces, Transfer, Beanstalk, App Runner, Lightsail.
• GCP: BigQuery, Spanner, Bigtable, AlloyDB, Dataproc/Dataflow/Composer, Filestore, Memcache, API Gateway, App Engine, Workflows, TPU, Notebooks, compute Interconnect/NodeGroup.
• Azure: SQL servers/DB (relational, distinct from Cosmos), Container Instances, Data Factory, Synapse, HDInsight, Kusto/Data Explorer, Logic Apps, NetApp, Recovery Services, IoT Hub, API Management, Container Apps, SQL MI, Stream Analytics, and remaining network sub-types (DDoS/ExpressRoute/Front Door/Network Watcher/Traffic Manager/VPN Gateway).

Counts (types surfaced in discovery today)

• AWS — 7 surfaced.
• Azure — 8 surfaced.
• GCP — 7 surfaced.

Fastest path = Workstream A: write walkers over drivers cloudemu already has, and add matching type-mapping entries in the three inventory handlers.