For the complete documentation index, see llms.txt. This page is also available as Markdown.

As Savings

Harness Autostopping automatically shuts down idle cloud resources and brings them back up on-demand, significantly reducing your cloud costs. This guide explains how savings are calculated, what to expect, and answers common questions.

How savings are calculated

Autostopping calculates savings by comparing two scenarios:

  1. Potential Cost: What you would have paid if the resource ran continuously without AutoStopping

  2. Actual Cost: What you actually paid based on real usage with Autostopping enabled

Savings = Potential Cost - Actual Cost
Savings = Potential Cost - Actual Cost
Savings Percentage = (Savings / Potential Cost) × 100  [undefined if Potential Cost = 0]

Example: If a VM costs $0.10/hour and Autostopping kept it off for 20 hours in a day:

  • Potential Cost: $0.10 × 24 hours = $2.40

  • Actual Cost: $0.10 × 4 hours = $0.40

  • Savings: $2.00 (83%)


Savings modes

Active mode (actual savings)

When an Autostopping rule is enabled and actively managing life cycle of resources, savings represent real money saved. The system tracks:

  • When resources are started (warm-up)

  • When resources are stopped (cool-down)

  • Actual cloud billing costs during active periods

Dry run mode (projected savings)

When a rule is in dry run mode, it monitors traffic patterns without actually stopping resources. Dry run savings show what you would save if the rule were fully enabled.

Use dry run mode to:

  • Evaluate potential savings before enabling Autostopping

  • Validate traffic patterns and idle detection

  • Build confidence before production deployment


Data sources

AutoStopping Savings calculations makes use of actual billing data from your cloud provider:

Cloud Provider
Data Source

AWS

Cost and Usage Report (CUR)

GCP

BigQuery Billing Export

Azure

Cost Management API

Billing data delay

Cloud providers typically have a 24-72 hour delay in billing data availability. This means:

  • Savings for today may not appear until 1-3 days later

  • This is normal cloud provider behavior, not an Autostopping limitation

  • Historical savings are backfilled once billing data arrives


Calculation timing

Costs (Spend/Savings/Freed-up Capacity) are calculated with a 2-day delay to ensure data accuracy.

  • Daily recomputation covers the previous 5 days

  • Monthly recomputation between the 15th–20th adjusts the last 45 days to accommodate changes happened on already computed savings numbers

Cost numbers may change as billing data from your cloud provider finalizes.


Understanding your savings dashboard

Key metrics

Metric
Description

Actual Savings

Money saved by Autostopping (active rules only)

Potential Cost

What you would have paid without Autostopping

Actual Cost

What you actually paid

Idle Hours

Hours resources were stopped

Savings Percentage

Efficiency of your Autostopping configuration

Unrealized Savings

Projected savings from dry run rules

Per-resource breakdown

Savings are tracked at the individual resource level:

  • Each VM, database, or container service has its own savings record

  • Resources can be grouped by rule, account, or cloud provider

  • Drill down to see daily, weekly, or monthly trends


Common scenarios

Scenario 1: Rule enabled mid-day

If you enable an Autostopping rule at 2:00 PM:

  • Savings calculation starts from 2:00 PM

  • Hours before enablement are not included in potential cost

  • First day shows partial savings

Scenario 2: Rule disabled mid-day

If you disable a rule at 10:00 AM:

  • Savings are calculated only until 10:00 AM

  • Remaining hours are excluded from the calculation

  • The rule must be re-enabled to resume savings tracking

Scenario 3: Resource running at midnight

If a resource is actively running when the day changes:

  • The session continues seamlessly into the next day

  • Savings calculation accounts for cross-day sessions

  • No manual intervention required

Scenario 4: Multiple start/stop cycles

If a resource starts and stops multiple times per day:

  • Each session is tracked individually

  • Total savings = sum of all idle periods

  • Frequent cycling may indicate idle time settings need adjustment


Frequently asked questions

Why are my savings showing as zero?

Savings may display as zero for several reasons:

  1. Billing data hasn't arrived yet

    • Allow 1-3 days for cloud billing data to become available

    • Check back after the billing delay period

  2. Rule was disabled

    • Disabled rules don't generate savings

    • Re-enable the rule to resume savings tracking

  3. Resources ran continuously

    • If there was no idle time, there are no savings

    • Review your idle time settings and traffic patterns

  4. New rule

    • Newly created rules need time to accumulate data

    • First savings appear after the billing delay period

  5. Resources covered by Reserved Instances or Savings Plans

    • If your resources are fully covered by RI/SP commitments, the hourly cost may show as $0

    • Savings = Potential Cost - Actual Cost, so if both are $0, savings will be $0

    • This is expected behavior — you're already getting discounts through your commitments

    • Check your cloud billing to verify RI/SP coverage

Why is there a delay in savings appearing?

Cloud providers (AWS, GCP, Azure) have inherent delays in their billing systems:

  • AWS CUR: Updated every 24 hours, may take up to 72 hours for complete data

  • GCP BigQuery: Typically 24-48 hour delay

  • Azure: 24-72 hour delay depending on resource type

This is standard cloud provider behavior. Savings will appear once billing data is available.

Can savings ever be negative?

No. Savings are always zero or positive. If actual costs temporarily exceed potential costs (due to billing adjustments, credits, or data corrections), savings display as zero rather than a negative value.

How accurate are the savings calculations?

Savings calculations use actual billing data from your cloud provider, making them highly accurate. The system:

  • Uses real hourly rates from your billing data

  • Accounts for pricing tiers, discounts, and reserved instances

  • Tracks actual resource state changes

What happens during cloud provider outages?

If billing data is temporarily unavailable due to cloud provider issues:

  • The system retries for up to 5 days

  • Once data becomes available, savings are calculated normally

  • No manual intervention is required

How do I maximize my savings?

  1. Optimize idle time settings: Set appropriate idle timeouts based on your usage patterns

  2. Use schedules: Configure fixed schedules for predictable workloads

  3. Enable more rules: Extend Autostopping to additional resources

  4. Review dry run data: Use dry run mode to identify savings opportunities

  5. Monitor regularly: Check the dashboard to identify underperforming rules


Savings calculation examples

Example 1: EC2 instance

Setup:

  • Instance type: m5.large ($0.096/hour)

  • Autostopping idle time: 15 minutes

  • Typical usage: 8 hours/day active, 16 hours/day idle

Daily Calculation:

Example 2: RDS database

Setup:

  • Instance type: db.r5.large ($0.24/hour)

  • Development database used only during business hours

  • Active: 10 hours/day (8 AM - 6 PM)

Daily Calculation:

Example 3: Kubernetes workload

Setup:

  • Namespace with 3 pods

  • Each pod: 2 vCPU, 4GB memory

  • Spot instances at $0.05/hour per pod

  • Active 6 hours/day for testing

Daily Calculation:


Troubleshooting

Savings Not Appearing

1. **Check rule status**: Ensure the rule is enabled and not in error state 2. Ensure billing enabled connector is available 3. **Verify billing integration**: Confirm cloud billing data is being ingested 4. **Wait for billing delay**: Allow 1-3 days for data to arrive 5. **Check resource activity**: Verify resources are actually being stopped

Unexpected Savings Values

1. **Review resource changes**: Instance type changes affect hourly rates 2. **Check for billing adjustments**: Credits or refunds may affect calculations 3. **Verify rule configuration**: Ensure correct resources are associated with the rule

Missing Historical Data

1. **Check rule creation date**: Savings only exist from when the rule was created 2. **Verify billing data availability**: Historical billing data may have retention limits 3. **Review recalculation status**: Monthly recalculations may be in progress

Glossary

Term
Definition

Actual Cost

The real cost incurred based on actual resource usage

Potential Cost

The cost that would have been incurred if resources ran 24/7

Idle Time

The configured duration of inactivity before resources are stopped

Warm-up

The process of starting stopped resources when traffic is detected

Cool-down

The process of stopping resources after the idle time expires

Dry Run

A mode that simulates Autostopping without actually stopping resources

Hourly Rate

The per-hour cost of a resource from cloud billing data

Billing Delay

The time between resource usage and billing data availability

Last updated

Was this helpful?