Azure Cost Management Basics

Cost Management answers “what did we spend and on what,” reservations and the Well-Architected Framework’s cost pillar answer “how do we spend less without breaking something else.” All three matter together — cutting cost without reading the trade-off is how a Well-Architected review finds an over-optimized deployment months later.

Reading a cost breakdown

az costmanagement query \
  --type ActualCost \
  --timeframe MonthToDate \
  --scope "/subscriptions/3b1f9c2a-6d4e-4a7b-9c11-2f5e8a0b7d33" \
  --dataset-aggregation '{"totalCost":{"name":"PreTaxCost","function":"Sum"}}' \
  --dataset-grouping name="ResourceGroup" type="Dimension" \
  --output table

$ az costmanagement query --type ActualCost --timeframe MonthToDate --scope "/subscriptions/3b1f9c2a-6d4e-4a7b-9c11-2f5e8a0b7d33" --dataset-aggregation '{"totalCost":{"name":"PreTaxCost","function":"Sum"}}' --dataset-grouping name="ResourceGroup" type="Dimension" --output table
PreTaxCost    ResourceGroup    Currency
------------  ---------------  ----------
14.82         learn-rg         USD
2.03          NetworkWatcherRG  USD

NetworkWatcherRG is a resource group Azure creates automatically the first time Network Watcher is used in a region (triggered by the test-ip-flow call in the virtual-network lesson) — a common source of “where did this resource group come from” confusion in a cost report.

A real pay-as-you-go vs. reserved instance comparison

Rather than quoting a discount percentage from memory, pull actual current prices from Azure’s public Retail Prices API for one concrete SKU — Standard_D2s_v5 (2 vCPU, 8 GiB), Linux, East US:

curl -s "https://prices.azure.com/api/retail/prices?\$filter=armRegionName eq 'eastus' and armSkuName eq 'Standard_D2s_v5' and priceType eq 'Consumption'" \
  | jq '.Items[] | select(.productName=="Virtual Machines Dsv5 Series") | {meterName, retailPrice, unitOfMeasure}'

$ curl -s "https://prices.azure.com/api/retail/prices?..." | jq '...'
{
  "meterName": "D2s v5",
  "retailPrice": 0.096,
  "unitOfMeasure": "1 Hour"
}

curl -s "https://prices.azure.com/api/retail/prices?\$filter=armRegionName eq 'eastus' and armSkuName eq 'Standard_D2s_v5' and priceType eq 'Reservation'" \
  | jq '.Items[] | {reservationTerm, retailPrice}'

$ curl -s "https://prices.azure.com/api/retail/prices?..." | jq '...'
{ "reservationTerm": "1 Year", "retailPrice": 519.0 }
{ "reservationTerm": "3 Years", "retailPrice": 997.0 }

Fetched from Azure’s live Retail Prices API on 2026-09-13, region eastus, SKU Standard_D2s_v5. The reservation retailPrice is the total cost for the whole term, paid upfront — not a per-hour rate — so the comparison needs converting to the same unit:

(8,760 = 24 × 365 hours/year; savings = (840.96 − annualized_reserved) / 840.96.)

The 3-year number looks like the obvious choice, but the trade-off is commitment: a reserved instance bills its full amount whether or not the VM runs, so it only pays off if the workload is genuinely steady for the term. A workload that might be decommissioned in eight months should stay on pay-as-you-go even at the worse hourly rate — this is exactly the kind of cost-vs-flexibility trade-off the Well-Architected Framework’s cost pillar asks you to state explicitly rather than defaulting to “cheapest number.”

The Well-Architected Framework’s five pillars, at a glance

The framework’s own point is that these pillars pull against each other: a 3-year reservation improves Cost Optimization but reduces the flexibility that Performance Efficiency wants when load patterns change, and locking RBAC down tightly (Security) can slow down Operational Excellence if every routine task needs a new role request. A Well-Architected review names the trade-off being made, rather than treating one pillar as free to maximize.

Practice

az costmanagement query --type ActualCost --timeframe MonthToDate --scope "/subscriptions/3b1f9c2a-6d4e-4a7b-9c11-2f5e8a0b7d33" --dataset-aggregation '{"totalCost":{"name":"PreTaxCost","function":"Sum"}}' --dataset-grouping name="ServiceName" type="Dimension" --output table

$ az costmanagement query ... --dataset-grouping name="ServiceName" type="Dimension" --output table
PreTaxCost    ServiceName          Currency
------------  -------------------  ----------
9.14          Virtual Machines     USD
4.02          Storage              USD
2.31          SQL Database         USD
1.38          Azure Monitor        USD

az consumption budget create \
  --budget-name learn-monthly-budget \
  --amount 50 \
  --time-grain Monthly \
  --start-date 2026-09-01 \
  --end-date 2027-09-01 \
  --notifications '{"actual_GreaterThan_80_Percent":{"enabled":true,"operator":"GreaterThan","threshold":80,"contactEmails":["reader@example.com"]}}'

$ az consumption budget create --budget-name learn-monthly-budget --amount 50 --time-grain Monthly ...
{
  "name": "learn-monthly-budget",
  "amount": 50.0,
  "timeGrain": "Monthly",
  "category": "Cost"
}

curl -s "https://prices.azure.com/api/retail/prices?\$filter=armRegionName eq 'eastus' and armSkuName eq 'Standard_B1s' and priceType eq 'Consumption'" | jq '.Items[] | select(.productName=="Virtual Machines B Series") | {meterName, retailPrice}'

$ curl -s "https://prices.azure.com/api/retail/prices?..." | jq '...'
{ "meterName": "B1s", "retailPrice": 0.0104 }

az costmanagement query --type ActualCost --timeframe MonthToDate --scope "/subscriptions/3b1f9c2a-6d4e-4a7b-9c11-2f5e8a0b7d33" --dataset-aggregation '{"totalCost":{"name":"PreTaxCost","function":"Sum"}}' --dataset-grouping name="ResourceId" type="Dimension" --query "sort_by(properties.rows, &[0])[-1]"

$ az costmanagement query ... --query "sort_by(properties.rows, &[0])[-1]"
[8.42, "/subscriptions/.../resourceGroups/learn-rg/providers/Microsoft.Compute/virtualMachines/learn-vm", "USD"]

Decision: run Azure SQL Database on the Basic tier (5 DTU, no zone redundancy) instead of a
zone-redundant Premium tier.

Cost: Basic is billed at a small fraction of Premium's rate.
Reliability given up: no automatic failover across availability zones — a zone outage in
  the region causes downtime until Azure recovers the single instance, rather than an
  automatic failover to a healthy replica.
Acceptable because: this is a learning environment with no uptime requirement; the same
  choice on a production order-processing database would trade away reliability the
  business actually needs, and Premium's cost would be justified instead.

This closes the Azure track. From here: terraform covers provisioning Azure resources (alongside AWS and GCP) as code with HCL instead of the CLI, and cloud-interviews covers scenario questions that span all three major clouds.