finops（finops 效率）

FinOps CLI

Cloud cost analysis and optimization tool. Scans AWS accounts (single or Organization with multiple accounts) for cost data and resources, stores everything locally in SQLite, and generates reports with trend analysis, anomaly detection, and period comparisons.

Prerequisites

• The finops binary must be installed and available in PATH
• Valid AWS credentials configured (via environment variables, AWS profiles, or IAM roles)
• AWS permissions needed: ce:GetCostAndUsage, sts:GetCallerIdentity, organizations:DescribeOrganization, organizations:ListAccounts, and read-only access to EC2, RDS, S3, Lambda, ECS, EKS, ElastiCache, CloudFront for resource discovery

Core Workflow

The CLI follows a scan → report → analyze pattern:

1. Scan — Download cost data and discover resources from AWS into local SQLite
2. Report — Generate reports from the local database (never calls AWS APIs)
3. Analyze — Interpret JSON output to provide insights and recommendations

CRITICAL: Always use --output json --file - for report commands, and pipe through 2>/dev/null | sed '/^Report saved to:/d' to get clean JSON on stdout. The CLI prints a Report saved to: - status line to stdout after the JSON which will break JSON parsers if not stripped.

# Standard pattern for getting clean JSON from any report command:
finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

Command Reference

finops scan

Downloads costs and discovers resources from cloud providers into local SQLite.

finops scan --provider aws [flags]

Behavior:

• Auto-detects AWS Organizations — scans all member accounts if in org mode. Works identically for single-account setups (scans only the current account).
• Incremental sync: first run fetches full history, subsequent runs fetch from last sync date.
• Permission errors on individual accounts in an organization are logged as warnings; scan continues with remaining accounts.
• Idempotent: safe to run multiple times (upserts via unique constraints).
• Discovers resources only for services with actual spend (zero-cost services are skipped).
• Use -v (verbose) to see per-account details including record counts and skipped accounts.

Resource discovery services: EC2 instances, EBS volumes, RDS databases, S3 buckets, Lambda functions, ECS/EKS clusters, ElastiCache clusters, NAT Gateways, CloudFront distributions.

Services without resource discovery (cost is tracked but no resource detail): CloudTrail, CodePipeline, Glue, KMS, CloudWatch, Route 53, SES, SNS, SQS, WAF, API Gateway, OpenSearch, QuickSight, Bedrock, DynamoDB, Redshift, Athena, Step Functions, Elastic Load Balancing, and others.

finops report summary

Full cost overview with totals, top services, cost by region, cost by account, commitment utilization, and resource counts.

finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

finops report top-services

Ranked list of top N services by total cost. Returns the same JSON structure as summary but with only top_services populated — cost_by_account, monthly_spend, and cost_by_region will be null.

finops report top-services --output json --file - --limit 20 2>/dev/null | sed '/^Report saved to:/d'

finops report trend

Cost evolution over time with trend direction indicator.

finops report trend --output json --file - 2>/dev/null | sed '/^Report saved to:/d'
finops report trend --output json --file - --service "Amazon Relational Database Service" 2>/dev/null | sed '/^Report saved to:/d'

Note: If the service name does not match any data, the response will have "data_points": null (not an empty array) and "direction": "flat". Always use the exact service name from top_services output.

finops report anomalies

Detects cost spikes using z-score statistical analysis (threshold: |z| >= 2.0).

finops report anomalies --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

Severity levels: high (|z| >= 4.0), medium (|z| >= 3.0), low (|z| >= 2.0). Negative deviations indicate unexpected drops; positive indicate spikes.

finops report compare

Side-by-side comparison of two time periods with absolute and percentage changes.

Known issue (v0.1.x): The date range separator : conflicts with Unix path separator on macOS/Linux, causing parse failures. Until fixed, use the trend report to compare periods visually, or use summary data to calculate differences manually.

finops report compare --output json --file - --current "2026-01-01:2026-03-01" --previous "2025-10-01:2025-12-01" 2>/dev/null | sed '/^Report saved to:/d'

finops report resources

Lists discovered resources with spec details and associated service cost context. Supports combined filters.

finops report resources --output json --file - 2>/dev/null | sed '/^Report saved to:/d'
finops report resources --output json --file - --service "AWS Lambda" --region "eu-west-1" 2>/dev/null | sed '/^Report saved to:/d'

finops db stats

Shows database size, record counts, last sync date, and retention setting. Use this to verify data freshness before generating reports.

finops db stats

Example output:

Database Statistics

Size:                288.00 KB
Cost records:        709
Resources:           258
Sync history:        12
Last sync:           2026-03-19T21:25:48Z
Retention:           12 (default) months

finops db prune

Modify data retention and prune old records.

finops db prune --retention 6

finops version

Print the CLI version.

finops version

Global Flags

Shared Report Flags

All finops report * subcommands accept:

JSON Output Schemas

CRITICAL: Always use --output json --file - 2>/dev/null | sed '/^Report saved to:/d' to get clean, parseable JSON on stdout.

Summary JSON

{
  "report_type": "summary",
  "generated_at": "2026-03-19T21:26:02Z",
  "period": {
    "start": "2025-09-01",
    "end": "2026-03-01"
  },
  "total_spend": 85459.92,
  "currency": "USD",
  "active_services": 39,
  "resources_discovered": 258,
  "cost_by_account": [
    {
      "account_id": "557981700545",
      "total_amount": 59978.97,
      "currency": "USD",
      "percentage": 70.2,
      "resource_count": 258,
      "top_services": [
        { "service": "Amazon Relational Database Service", "amount": 10396.41 },
        { "service": "Amazon OpenSearch Service", "amount": 7700.55 }
      ]
    }
  ],
  "monthly_spend": [
    { "period": "2025-09-01", "amount": 11697.77 },
    { "period": "2025-10-01", "amount": 14529.19 }
  ],
  "top_services": [
    {
      "service": "Amazon Relational Database Service",
      "total_amount": 14572.27,
      "currency": "USD",
      "percentage": 17.1
    }
  ],
  "cost_by_region": [
    {
      "region": "eu-west-1",
      "total_amount": 52000.00,
      "currency": "USD",
      "resource_count": 195,
      "service_costs": [
        { "service": "Amazon Relational Database Service", "amount": 14000.00 }
      ],
      "resources": [
        {
          "service": "Amazon Relational Database Service",
          "resource_type": "rds:db",
          "resource_id": "arn:aws:rds:eu-west-1:557981700545:db:rds-app-pro",
          "name": "rds-app-pro",
          "account_id": "557981700545"
        }
      ]
    }
  ],
  "commitments": {
    "total_committed": 3258.00,
    "total_used": 3257.99,
    "total_savings": 0,
    "avg_utilization_pct": 100,
    "currency": "USD",
    "has_data": true,
    "spot_instance_count": 0,
    "types": [
      {
        "type": "savings_plan",
        "total_commitment": 3258.00,
        "used_commitment": 3257.99,
        "on_demand_equivalent": 6995.14,
        "net_savings": 0
      },
      {
        "type": "reserved_instance",
        "total_commitment": 0,
        "used_commitment": 0,
        "on_demand_equivalent": 70622.58,
        "net_savings": 0
      }
    ]
  }
}

Key notes:

• cost_by_region[].resources[] has a reduced field set compared to the resources report: only service, resource_type, resource_id, name, account_id (no state, spec, tags, region).
• commitments is present when commitment data exists. has_data indicates if there is any commitment information.
• top_services within cost_by_account shows the top 5 services per account.

Top-Services JSON

Uses the same schema as Summary but only top_services is populated. Fields cost_by_account, monthly_spend, and cost_by_region will be null:

{
  "report_type": "summary",
  "generated_at": "2026-03-19T21:26:04Z",
  "period": { "start": "2025-09-01", "end": "2026-03-01" },
  "total_spend": 85459.92,
  "currency": "USD",
  "active_services": 5,
  "resources_discovered": 0,
  "cost_by_account": null,
  "monthly_spend": null,
  "top_services": [
    { "service": "Tax", "total_amount": 15450.32, "currency": "USD", "percentage": 18.1 },
    { "service": "Amazon Relational Database Service", "total_amount": 14572.27, "currency": "USD", "percentage": 17.1 }
  ],
  "cost_by_region": null
}

Trend JSON

{
  "report_type": "trend",
  "generated_at": "2026-03-19T21:26:04Z",
  "service": "Amazon Relational Database Service",
  "direction": "down",
  "data_points": [
    { "period": "2025-09-01", "amount": 2000.71 },
    { "period": "2025-10-01", "amount": 2307.75 },
    { "period": "2025-11-01", "amount": 2403.52 },
    { "period": "2025-12-01", "amount": 3125.51 },
    { "period": "2026-01-01", "amount": 2581.33 },
    { "period": "2026-02-01", "amount": 2153.44 }
  ]
}

• direction: "up" (>5% increase last month), "down" (>5% decrease), "flat" (within 5%).
• service: empty string "" when no --service filter is applied (shows all combined).
• data_points: null (not []) when the service name doesn't match any data.

Anomalies JSON

{
  "report_type": "anomalies",
  "generated_at": "2026-03-19T21:26:03Z",
  "anomalies": [
    {
      "period": "2026-02-01",
      "service": "AWS Lambda",
      "expected": 1.35,
      "actual": 5.44,
      "deviation": 2.14,
      "severity": "low"
    },
    {
      "period": "2025-09-01",
      "service": "Amazon Elastic Compute Cloud - Compute",
      "expected": 857.6,
      "actual": 323.43,
      "deviation": -2.0,
      "severity": "low"
    }
  ]
}

• severity: "high" (|z| >= 4.0), "medium" (|z| >= 3.0), "low" (|z| >= 2.0).
• deviation can be negative (unexpected cost drop) or positive (cost spike). Both are anomalies.
• expected is the historical mean cost for that service.

Compare JSON

{
  "report_type": "compare",
  "generated_at": "2026-03-19T20:07:45Z",
  "current_period": { "start": "2026-01-01", "end": "2026-03-01" },
  "previous_period": { "start": "2025-10-01", "end": "2025-12-01" },
  "total_current": 29410.86,
  "total_previous": 28929.97,
  "total_change": 480.89,
  "total_change_pct": 1.7,
  "currency": "USD",
  "service_deltas": [
    {
      "service": "Amazon Relational Database Service",
      "previous_amount": 5529.03,
      "current_amount": 4734.77,
      "absolute_change": -794.26,
      "percent_change": -14.4,
      "currency": "USD"
    }
  ]
}

Resources JSON

{
  "report_type": "resources",
  "generated_at": "2026-03-19T21:26:43Z",
  "total_count": 7,
  "resources": [
    {
      "service": "Amazon Relational Database Service",
      "resource_type": "rds:db",
      "resource_id": "arn:aws:rds:eu-west-1:557981700545:db:rds-app-pro",
      "name": "rds-app-pro",
      "region": "eu-west-1",
      "state": "available",
      "account_id": "557981700545",
      "spec": "{\"engine\":\"postgres\",\"engine_version\":\"17.7\",\"instance_class\":\"db.r8g.large\",\"multi_az\":true,\"storage_encrypted\":false,\"storage_gb\":100}",
      "tags": "{}"
    },
    {
      "service": "AWS Lambda",
      "resource_type": "lambda:function",
      "resource_id": "arn:aws:lambda:eu-west-1:557981700545:function:my-function",
      "name": "my-function",
      "region": "eu-west-1",
      "account_id": "557981700545",
      "spec": "{\"code_size\":3137,\"handler\":\"handler.lambda_handler\",\"memory_mb\":128,\"runtime\":\"python3.12\",\"timeout_s\":63}",
      "tags": "{}"
    }
  ]
}

Key notes:

• state is optional/null — only present for EC2 (running/stopped), RDS (available), CloudFront (enabled/disabled), ElastiCache (available). Lambda, S3, and NAT Gateways do not have state.
• spec and tags are JSON-encoded strings — parse them as nested JSON for structured data.
• spec content varies by resource type (see Resource Types below).
• tags is often "{}" (empty JSON object string) when no tags are set.

Resource Types and Spec Fields

AWS Service Names Reference

Service names in the CLI are exact Cost Explorer names. Use the full name when filtering. Common services:

Tip: Always run finops report top-services --output json --file - --limit 50 first to discover the exact service names available in the dataset, then use those names for --service filters.

Common Agent Patterns

Full Cost Audit

Scan data, then generate key reports for a complete picture:

finops scan --provider aws
finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'
finops report anomalies --output json --file - 2>/dev/null | sed '/^Report saved to:/d'
finops report top-services --output json --file - --limit 20 2>/dev/null | sed '/^Report saved to:/d'

Parse the summary for total spend and distribution, check anomalies for spikes, and identify the highest-cost services.

Investigate a Cost Spike

When anomalies are detected, drill into the specific service:

# 1. Detect anomalies
finops report anomalies --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

# 2. Check the trend for the flagged service (use exact name from anomaly output)
finops report trend --output json --file - --service "AWS Lambda" 2>/dev/null | sed '/^Report saved to:/d'

# 3. List resources for that service to identify what's running
finops report resources --output json --file - --service "AWS Lambda" 2>/dev/null | sed '/^Report saved to:/d'

Resource Right-Sizing Analysis

Find potentially over-provisioned resources:

# Get all EC2 instances
finops report resources --output json --file - --service "Amazon Elastic Compute Cloud - Compute" 2>/dev/null | sed '/^Report saved to:/d'

# Get all RDS databases
finops report resources --output json --file - --service "Amazon Relational Database Service" 2>/dev/null | sed '/^Report saved to:/d'

Parse spec JSON strings to analyze instance types, memory allocation, and storage. Cross-reference with cost data from the summary to identify high-cost, potentially over-provisioned resources.

Stopped/Idle Resource Detection

finops report resources --output json --file - --service "Amazon Elastic Compute Cloud - Compute" 2>/dev/null | sed '/^Report saved to:/d'

Filter resources where state is "stopped" — these incur EBS storage costs without providing compute value. Also check for EBS volumes with state: "available" (unattached volumes).

Multi-Account Cost Breakdown

For AWS Organizations, the summary groups costs by account:

finops scan --provider aws
finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

Analyze cost_by_account to identify which accounts drive the most spend, and cost_by_account[].top_services for per-account service breakdown.

Commitment Utilization Check

finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

Check commitments:

• avg_utilization_pct below 80% → over-provisioned commitments, money being wasted
• avg_utilization_pct at 100% → fully utilized, potentially room for more savings
• Compare on_demand_equivalent vs total_commitment per type to quantify actual savings

Regional Cost Analysis

finops report summary --output json --file - 2>/dev/null | sed '/^Report saved to:/d'

Analyze cost_by_region with service_costs breakdown to identify:

• Regions with unexpectedly high costs
• Services running in expensive regions that could be relocated
• Resource sprawl across too many regions

Data Freshness Check

Always verify data is available before generating reports:

finops db stats

If Cost records: 0 or Last sync is stale, run finops scan --provider aws first.

Error Handling

Errors return exit code 1. Success returns exit code 0 (even with empty results).

Important Notes

• Reports are local-only. They read from SQLite and never call AWS APIs. Only finops scan contacts AWS.
• Run scan first. Reports will error with "no data found" if the database is empty.
• Incremental sync. Subsequent scans only fetch new data since the last sync date.
• Service names are exact AWS Cost Explorer names. Use top-services to discover the exact names before using --service filters.
• Date format. Dates use YYYY-MM-DD. Report period defaults to last 6 months.
• Database location. Data is stored at ~/.finops/data.db.
• Retention. Default 12 months. Change with finops db prune --retention N.
• Multi-account errors are non-fatal. If some accounts lack permissions, scan continues with warnings and partial data.
• Monetary values are float64 rounded to 2 decimal places. Currency is always in the JSON output.
• spec and tags in resources are JSON-encoded strings inside the JSON — they must be parsed as nested JSON.
• Null vs empty arrays. When no data matches a filter, arrays may be null instead of []. Always handle both cases.
• Report saved to: - is always printed to stdout. Use sed '/^Report saved to:/d' to strip it for clean JSON parsing.