Cost Optimisation

Scope. Canonical home for the cost-optimisation outcomes delivered by the platform: scale-to-zero compute, automated storage lifecycle, cost-allocation visibility, tier-configurable services, and the quantified business case. The FinOps practices and implementation mechanics are canonical in practices/finops.md.

What this repo uniquely brings to cost optimisation

1. Scale-to-zero compute (cross-ref)

Setting min_instance_count = 0 eliminates compute costs when no traffic is being served. Cloud Run bills per request, per second — idle applications cost nothing. GKE Autopilot bills per pod resource request rather than per provisioned node, with Vertical Pod Autoscaling continuously right-sizing requests to eliminate waste.

Runtime mechanics canonical in capabilities/serverless.md §1–3.

2. Automated storage lifecycle (canonical)

Storage cost creep is automated away by default:

• Cloud Run revision pruning — modules/App_CloudRun/scripts/prune-old-revisions.sh retains only max_revisions_to_retain revisions (default 7) after every deploy; traffic-serving revisions are always preserved.
• Artifact Registry cleanup — max_images_to_retain, delete_untagged_images, image_retention_days prevent unbounded image accumulation.
• GCS bucket lifecycle — per-bucket lifecycle rules automate transition to cheaper storage classes (Nearline, Coldline, Archive).

Full implementation in practices/finops.md §1.

3. CDN cost offload (cross-ref)

enable_cdn = true routes cacheable responses to Google edge points of presence, offloading them from compute. BUSINESS_CASE.md projects 30–50% compute and egress savings on read-heavy applications. CDN implementation in capabilities/networking.md.

4. Cost-allocation visibility (canonical)

The app<name><tenant><id> resource-naming convention flows directly into Cloud Billing labels, enabling per-tenant chargeback reports via simple BigQuery views over the billing export — no manual tagging required. modules/Services_GCP/gke_metering.tf adds GKE pod-level cost visibility on Autopilot clusters.

Canonical in practices/finops.md §2 and capabilities/multitenancy_saas.md §1.

5. Tier-configurable shared services

Every expensive shared resource exposes cost/performance trade-off variables:

Service	Cost control variables
Cloud SQL	db-custom-* tiers; HA and PITR optional
Memorystore Redis	BASIC vs STANDARD_HA
Filestore NFS	BASIC_HDD / BASIC_SSD / ZONAL; enable_nfs = false to skip entirely
Cloud Run	cpu_limit, memory_limit, min_instance_count, max_instance_count, concurrency

Full tier reference in capabilities/data_and_databases.md.

6. Cost vs performance profiles

AGENTS.md /performance documents three explicit deployment profiles:

Profile	Key settings
Low Cost	min_instance_count = 0, scale-to-zero, smallest viable DB instance
Low Latency	min_instance_count = 1, larger resources, connection pooling, Redis caching
Balanced	min_instance_count = 0, right-sized resources, CDN and Redis enabled

7. Quantified business case

Per BUSINESS_CASE.md §3.C and IAC_AUTOMATION_BUSINESS_CASE.md:

Metric	Value
Provisioning time reduction	~95% (3–5 days → <2 hours)
Cost per new application setup	$200 vs $3,200 manually
Maintenance effort reduction (10-app portfolio)	~95% (40 h → 2 h per cycle)
Maintenance cost reduction (10-app portfolio)	$4,000 → $200 per cycle
Projected annual savings (mid-size portfolio)	>$100,000
Compute/egress savings on read-heavy apps	30–50% via CDN offload

Cross-references

• practices/finops.md — implementation mechanics (scale-to-zero, lifecycle policies, cost-allocation labels, tier trade-offs)
• capabilities/serverless.md — Cloud Run / GKE Autopilot per-second billing mechanics (§1–3)
• capabilities/data_and_databases.md — tier-configurable backing services
• capabilities/networking.md — CDN configuration and cost offload
• capabilities/multitenancy_saas.md — per-tenant chargeback via resource naming (§1)
• outcomes/developer_productivity.md — quantified provisioning and maintenance savings (§4)