cost_tracking_enable

Description

The cost_tracking_enable parameter turns on the Rack's cost accumulator, which estimates each App's EC2 compute spend and powers convox cost and per-App budget caps. The accumulator runs inside the Rack API. On each tick, every 10 minutes, it charges each running pod the share of its node's hourly price that the pod's CPU or memory requests reserve, whichever is larger, or its share of the node's GPUs when it requests GPUs. The result is added to the App's spend for the current calendar month, which resets on the 1st, UTC. Spend is stored as an annotation on the App's namespace and surfaces in convox cost, the Convox Console cost views, and budget enforcement.

On AWS, spend is priced against us-east-1 Linux on-demand list prices, keyed by the node's node.kubernetes.io/instance-type label (for example m5.large).

Cost tracking is a prerequisite for budget caps. While it is off, the Rack rejects convox budget cap raise, and any convox budget set that sets a monthly cap, alert threshold or at-cap action, with HTTP 422. It also rejects every promote of an App whose convox.yml budget: block sets monthlyCapUsd, alertThresholdPercent or atCapAction. The error points at this parameter.

Default Value

The default value for cost_tracking_enable is false. Users must opt in to enable the cost accumulator and budget-cap surfaces.

Use Cases

Per-App cost visibility: See each App's estimated month-to-date spend by Service in convox cost and the Convox Console, without an external cost-management tool.
Budget caps: Set a monthly cap with convox budget set <app> --monthly-cap 1000 and have the Rack alert, block new deploys, or scale Services to zero when the App's month-to-date spend reaches it.
GPU spend by Service: A pod that requests GPUs is charged for its share of the node's GPUs, so each GPU Service's cost appears in its own row.

Capacity Considerations

The accumulator runs inside the Rack API and adds no pods to the cluster. The Console cost views read the Rack's cost data directly and do not depend on Console monitoring or Prometheus.

Console monitoring is a separate Console setting that installs a Prometheus chart. The capacity that chart needs on a Rack with small workload nodes is covered under gpu_observability_enable.

Setting Parameters

To enable cost tracking on an existing rack:

$ convox rack params set cost_tracking_enable=true -r rackName
Updating parameters... OK

The change updates the Rack API Deployment and creates no cloud resources. An App's spend first appears on the second tick after the update completes, about 10 minutes later.

To disable:

$ convox rack params set cost_tracking_enable=false -r rackName
Updating parameters... OK

The accumulator stops once the update has rolled the Rack API. Stored spend is kept, and convox cost shows the last snapshot with a notice that tracking is disabled. Deploy blocking and auto-shutdown stop with it, and Services that auto-shutdown already scaled to zero stay at zero until convox budget reset. A promote of an App whose budget: block sets monthlyCapUsd, alertThresholdPercent or atCapAction is rejected until the block is removed or tracking is enabled again, including the re-promote that convox apps params set, convox apps lock and convox apps unlock run.

Additional Information

Cost tracking is available on AWS from Rack version 3.24.6, on Azure from 3.25.1 and on GCP from 3.25.9. See cost_tracking_enable (Azure) and cost_tracking_enable (GCP) for provider details.
The accumulator runs inside the Rack API, not as a sidecar. It ticks once when it starts and then every 10 minutes, and no parameter changes the interval.
The pricing table ships with the Rack version and cannot be edited. convox cost --aggregate shows its date: pricing-table:2026-09-28 from 3.25.9.
Estimates use list prices, so discounts AWS applies to your bill, such as Savings Plans, Reserved Instances or an Enterprise Discount Program, are not reflected. The pricing adjustment on an App's budget scales its estimate by a multiplier from 0.1 to 1.5: convox budget set my-app --monthly-cap 1000 --pricing-adjustment 0.7 records 70% of the estimate.
A node labeled karpenter.sh/capacity-type=spot (Karpenter) or eks.amazonaws.com/capacityType=SPOT (an EKS spot node group) is priced at 0.30 of its instance type's on-demand rate. No AWS entry in the table has its own spot factor, so every AWS spot row uses that default. A node with no capacity label at all is priced on-demand and shows capacity unknown.
An instance type missing from the table adds no spend, and its pods are counted in warning-count. See Unpriced instance types.
Spend covers the current calendar month and resets on the 1st, UTC. The Rack keeps no earlier months.

gpu_observability_enable: Exports DCGM GPU utilization metrics. Cost tracking does not read them: it charges a GPU pod for the GPUs it requests, whether it uses them or not.
webhook_signing_key: Webhook deliveries from cost-tracking events (app:budget:auto-shutdown:armed, app:budget:auto-shutdown:fired) carry an HMAC signature when this is set, so receivers can verify authenticity.

Version Requirements

This parameter requires at least Convox rack version 3.24.6.