[A3S Cloud/H0] Export workload-identity metrics for autoscaling and operations

[ZhiXiao-Lin]

Cloud gate

H0 — measured multi-node placement, operations, and autoscaling evidence.

Planning reference: https://github.com/A3S-Lab/Cloud/blob/e2a05b7f1f63486b2fda866b22aa84eea89aadb4/docs/development-plan.md

Problem

Cloud cannot make truthful operational or scaling decisions from node-wide aggregates alone. Observer needs stable workload identity and explicit freshness semantics for per-replica resource signals.

Scope

• Attach stable workload, deployment, revision, replica, provider-unit, and node identity to collected signals.
• Collect bounded CPU, memory, network, process, restart, and availability signals at the supported workload boundary.
• Export the contract through OTLP and Prometheus-compatible metrics with controlled cardinality.
• Report sample timestamp, collection interval, and explicit fresh, stale, unavailable, or unknown state.
• Define identity behavior during rolling replacement, restart, adoption, and cleanup.
• Add Linux integration fixtures with multiple revisions and replicas on one node.

Acceptance criteria

• Cloud can attribute every accepted sample to one immutable workload revision and replica identity.
• A missing or stale sample is never represented as zero usage.
• Exported labels have documented cardinality limits and contain no tenant secrets or raw user input.
• Restart and adoption preserve logical identity while replacement creates the expected new identity.
• OTLP and Prometheus exports agree on values, timestamps, and freshness state for the same fixture.

Non-goals

• Autoscaling, scheduling, or placement decisions.
• A control-plane metrics database.
• Application-level tracing instrumentation owned by workloads.