Ways to connect
Five ways to get your data to qloudpulse, from a read-only role that takes minutes to a client you run and redact yourself, which lets you use qloudpulse without granting it any access to your AWS accounts. Each one says what qloudpulse sees, what you run and who it suits. Most companies combine two or three.
The five ways
-
Connected AWS account
A read-only IAM role in each account, which qloudpulse assumes with your company's own ExternalId. A script you run in AWS CloudShell, from the AWS accounts page in the product, makes it.
- qloudpulse sees
- Every hour, what runs and what it costs. Once a day, your bill (CUR 2.0, read through Athena), the infrastructure behind it and carbon emissions. AWS's forecast and CloudWatch usage once an admin turns them on.
- You run
- One script per account, and one in the payer account for the bill. Nothing keeps running on your side.
- Suits
- Most teams: the least effort, and everything qloudpulse does.
-
Kubernetes agent Pilot
One pod per cluster running the upstream OpenTelemetry Collector, installed with the Helm chart or plain YAML from the public qloudpulse-public repository. It sends out over HTTPS; nothing connects into your cluster.
- qloudpulse sees
- Seven metrics every 30 seconds: what each container requests, what each pod uses, each node's allocatable CPU and memory, and each pod's phase, with the labels you choose to send.
- You run
- One pod per cluster, with a token that can write your company's metrics and do nothing else.
- Suits
- Every cluster you want priced per pod and namespace, in EKS or on your own hardware.
-
Offline collection New: early access
qloudpulse-collect, a client you run with your own credentials. It records what qloudpulse's collectors would read, redacts it by rules you write, and packs a bundle every hour. qloudpulse gets no access to your accounts.- qloudpulse sees
- Only what your bundles hold: the answers a connected account would give, after your redaction, for the hours they cover.
- You run
- One program on a schedule (cron, a systemd timer, a container or a Kubernetes CronJob), with a config file and a key you keep.
- Suits
- Teams that cannot grant cross-account access, or that want to see every byte before it leaves.
-
FOCUS file upload
The bills of your other clouds and providers, as the FOCUS files they export, the FinOps Foundation's open format. An admin uploads them in the product.
- qloudpulse sees
- Each day's cost per provider, account, service and region, in the currency billed, from FOCUS 1.0 to 1.4.
- You run
- An export at the provider, and an upload each month.
- Suits
- Azure, Google Cloud, Oracle Cloud or a SaaS bill beside your AWS bill.
-
CLI in CI
qloudpulse estimatereads a Pulumi preview and prints a pull request comment: what the change will cost a month, at your company's own rates, before it is applied.- qloudpulse sees
- The resource types in the preview and the attributes that set each one's price. Names, tags, config and secrets stay in your CI.
- You run
- One line in your pipeline, with a token you make under your name in the product.
- Suits
- Teams whose infrastructure changes arrive as pull requests, coding agents' included. It adds to a connected account rather than replacing one.
Compared
What each way asks of you, and what it lets you hold back.
| Way | Access qloudpulse needs | What it collects | Your effort | What you control |
|---|---|---|---|---|
| Connected account | A read-only role it assumes with your ExternalId. The role denies reading files, rows, secrets and logs. | What runs, hourly. The bill, infrastructure and carbon, daily. Forecast and usage metrics if you turn them on. | Least: one script per account. | Which accounts and which paid options. You read the role's policy before you run it. |
| Kubernetes agent | None into the cluster. The pod sends out with a write-only token. | Seven metrics per pod and node, every 30 seconds. | One install per cluster. | Which labels it sends. Its whole config is yours to read. |
| Offline collection New | None. An upload-only token, or nothing at all if you deliver the bundles yourself. | What a connected account gives, after your redaction, hour by hour. | Most: a scheduled job, a config file and a key to keep. | Every field. Tags you do not list are left out, values can be kept, dropped, replaced or pseudonymized, and you can open each bundle before it goes. |
| FOCUS upload | None. | Daily cost per provider, account, service and region. | An export and an upload each month. | Which files you upload. |
| CLI in CI | None. A token of yours lets the CLI send estimates. | A planned change's resource types and the attributes that set their price. | One line in CI. | Only what sets a price leaves, and --dry-run prints the request without sending it. |
Which one should I use?
Start from what your company will allow, then add what you need.
- You can grant a read-only role: connect your AWS accounts. It takes minutes and gives you every report.
- You cannot, or you want to read every byte that leaves: use offline collection instead, for the accounts it covers. It costs you a job to run and a config to keep.
- You run Kubernetes: add the agent to each cluster you want split by pod and namespace, beside either of the two above.
- You have bills from other clouds or providers: upload their FOCUS files.
- Your infrastructure changes come as pull requests: add the CLI to CI, to see a change's cost before it is applied.
- Nothing may leave your network at all: run all of qloudpulse yourself.
Each AWS account is either connected or fed by bundles, never both. Everything else combines freely, and you can change your mind later.
Offline collection New: early access
With qloudpulse-collect, you collect your AWS data yourself, decide what of it leaves, and deliver it. qloudpulse holds no role in your accounts and never calls them.
Early access. The client's first part is built, and the side of qloudpulse that takes bundles in is being built. Write to us if you want to try it; we will not promise a date until it works end to end.
What it reads
The same AWS calls qloudpulse's collectors make for a connected account, with your own credentials from the standard AWS chain: profiles, SSO, instance and container roles. It starts with what runs each hour; your bill's query results (never the raw bill files), the infrastructure behind them, carbon and AWS's forecast follow. It refuses any call that is not on a read-only list built into it, and the IAM policy it needs is made from the same list. Kubernetes stays with the agent.
You see it before it leaves
analyze, or the local page the client serves on your own machine, shows the tag keys and values it found and which resources carry them, what a rule would cost you (dropping Team turns off cost by team), and anything that looks like a key, a token, an email address or an IP address. It will not pack a bundle until each of those is redacted or allowed. The page also shows a record before and after your rules, side by side, and every record of a finished bundle.
Redaction rules you write
- In one TOML file, which you can review like any other config.
- Tags are left out unless listed. The starter file keeps the few that qloudpulse's reports use, such as
Name,EnvironmentandTeam. - Each value can be kept, dropped, replaced with a fixed value or a pattern, or pseudonymized with a key only you hold. The same value always gives the same pseudonym, so grouping survives, and qloudpulse cannot reverse it.
- Account and resource IDs can be pseudonymized in their own format: an account ID stays twelve digits and an instance ID keeps its
i-prefix and length, so the bill and the inventory still match. - Fields a price needs, such as the instance type and the region, cannot be dropped, and the client says why.
A bundle every hour
Each hour it writes a .tgz you can check: the recorded answers as plain-text NDJSON, and a manifest with a SHA-256 for every file and a summary of the rules applied. tar -xzf and jq are enough to read it.
Where it goes
- Your own S3 bucket, the only write the client makes to AWS.
- qloudpulse, with a token that can upload bundles for your company and do nothing else.
- A command of your own, run once per bundle, for any other route.
A build without the upload feature has no qloudpulse address in it at all, so you can check with grep, or with an egress policy that lets it reach AWS only, that it cannot send anything to us.
What qloudpulse keeps
Each bundle is checked when it arrives and fed through the same collectors as a connected account, so you get the same reports for the hours it covers. qloudpulse keeps only the newest bundle from each source, and deletes the one before once a newer one is in. A bundle more than 13 days old is refused, since hourly data is kept for 14 days. What stays is the data in your reports, not the files.
Running it all yourself
If no data may leave your network, run qloudpulse itself in your own Kubernetes cluster: one Helm chart in one namespace, with your data in your own database. It reads your clusters through the same agent, and your AWS accounts through the same read-only roles, with nothing reaching us. More about self-hosting.