Beta

Docs

How qloudpulse works, where your data flows, how to connect your AWS accounts and your Kubernetes clusters, and how to use a report, hosted or self-hosted. The same page is under Help in the product, shorter; this one is the full text.

How it works

From your AWS accounts to a page in your browser, in four steps. Nothing is installed in your accounts, and nothing is fetched when you open a page.

  1. Your accounts and clusters

    For an AWS account you create one read-only role, with a script you run in AWS CloudShell. It trusts qloudpulse only with your company's own ExternalId, and it denies reading what is stored: files, rows, secrets, logs. For a Kubernetes cluster you install one small pod, which sends out what each pod requests and uses, and nothing else. Delete the role or the pod and the access ends.

  2. The collector

    Each role is checked before anything is read. Then, every hour, the collector lists what is running and prices it at the AWS list price, with its usage from CloudWatch once you turn usage metrics on. A cluster's nodes are priced from their EC2 cost, or on your own hardware from the price per core-hour and per GiB-hour you set, and each node's cost is split across its pods. Once a day it reads your bill, if you connect it, and works out what you could change to spend less.

  3. Your company's database

    Every run is stored in a database your company has to itself: ours, in Ireland, on the hosted plan, or your own, on a self-hosted install in your own Kubernetes cluster. Opening a page asks nothing of AWS: what you see is what the last run stored, so the pages are fast, and a slow AWS API never slows you down.

  4. The reports

    A report adds up the stored figures for the hour or day you look at, split and filtered the way you ask. Every report is built from the same blocks, so once you know one you know them all, and anyone in your company can build a report of their own on the same data. Every report, in a sentence each.

Connecting AWS and clusters

Connect AWS, under Settings in the product, has three steps: the bill, the accounts, and the options. Each step hands you a script to download and run in AWS CloudShell, already holding your company's two values, the role qloudpulse reads through and your ExternalId. The scripts create roles and an export; nothing on the page runs anything in your accounts. If your company will not grant a role, Ways to connect compares the others, offline collection with a client you run yourself among them.

The bill

Do this first. The script runs in your payer account, the management account of an AWS Organization. It puts a daily export of your bill, CUR 2.0, into an S3 bucket, ours or one you already have, makes an Athena table over it, and creates the role qloudpulse-cost-reader, which can read that export and nothing else. Everything else starts from the bill: only the regions an account's bill shows in use are read, and an account with no usage is skipped. The first data lands within about 24 hours.

The accounts

The same kind of script, once per account you want read, the payer included. It creates the role cost-collector, which reads metadata and metrics only. Then you enter each account's id on the AWS accounts page, nothing else, since the roles' names are fixed. Each role is checked before anything is read: it has to let qloudpulse in with your company's ExternalId and refuse it without. A role that fails the check says why on the page, and you can check it again once it is fixed.

What the roles can read, and cannot

What runs and how much it is used; how it is connected; names and settings, never contents. The role's own policy denies reading what is stored, so a broader policy attached to it later still could not open your files, rows, secrets or logs. The full lists are on the start page under What we read and do not read, and every statement of every role, the cluster agent's included, is on the Permissions page and on the Connect AWS page, for whoever approves it in your company. Check them before you run anything: the Permissions page has a question to ask an AI assistant about each policy, and AWS's own IAM Access Analyzer checks a policy for free.

Options

Whether to collect your resource tags, so the inventory can be split by your Environment, Team and Company tags. Optional, and free.

What adds to your AWS bill is off until one of your admins turns it on, under Options on the AWS accounts page, where each switch says what AWS charges: usage metrics (CloudWatch, within AWS's free API tier for most accounts) and AWS's month forecast (Cost Explorer, about $0.35 per account a month). Reading the bill and sampling flow logs are the other two things AWS charges for, and each starts only when you connect it. The pricing page lists all four.

Clusters

Clusters, under Settings, lists your Kubernetes clusters and makes a token for each. You install one small pod per cluster with Helm or plain YAML, from the public qloudpulse-public repository, with that token: it reads the cluster's API and each node's kubelet and sends out what each pod requests and uses, over outbound HTTPS. Nothing connects into your cluster, so a private API endpoint works too. An EKS node's cost comes from its EC2 instance; a node on your own hardware is priced from what you say a core-hour and a GiB-hour cost you, so a cluster needs no AWS account. Reports then split each node's cost across its pods, by namespace and workload. What the pod may read in the cluster is on the Permissions page. On the hosted plan the pod sends to our service; on a self-hosted install it stays inside your cluster.

DigitalOcean

DigitalOcean is a pilot on the hosted plan. Under Settings, Clouds and services connects a DigitalOcean team with one personal access token, made inside that team with custom scopes and read scopes only; the page lists the scopes. qloudpulse tries each read once when you connect, and says which scope is missing if one is. Once a day it reads the team's daily costs and finished invoices, its Droplets, volumes, databases, clusters, firewalls and load balancers, the findings of DigitalOcean's own security scans if the team runs them, and Droplet CPU, and finds the waste and the open doors in them. The token is stored encrypted and only ever read. DigitalOcean cannot show a token's own scopes, so qloudpulse cannot check that yours holds read scopes only: that is up to you when you make it. Delete the token in DigitalOcean and the access ends.

Datadog

Datadog is a pilot on the hosted plan. Under Settings, Clouds and services connects a Datadog org with its API key and an application key owned by a Datadog service account that holds six read scopes only; the page lists them and gives a command that makes the key. qloudpulse tries each read once when you connect and keeps the connection only if all six answer. Once a day it reads the org's hourly usage, its monthly cost, its custom metrics with their volumes and which tag keys are queried, its log indexes, APM retention filters and hosts, and finds custom metrics nobody queries, metrics with more tags than anyone uses, and on-demand spend a commitment would cover, priced at your own contract's rates. The keys are stored encrypted and only ever read. Revoke the application key in Datadog and the access ends.

On-prem

A Kubernetes cluster on your own hardware connects the same way as one in EKS: one pod, one token. Its nodes have no cloud bill behind them, so you give each cluster a price per core-hour and per GiB-hour of what its nodes make allocatable, worked out from what the hardware costs you, and the reports price every pod from that.

A self-hosted install also reads VMware vSphere, as a pilot, every hour, from each vCenter you name in its settings, with a vCenter login that holds vCenter's built-in Read-only role: hosts, VMs, datastores, snapshots and what each VM used. You give each vSphere cluster a price per core-hour and per GiB-hour of its hosts and per GiB-month of its datastores; a VM's cost is its share of its cluster by what it is allocated, and what no VM is allocated shows as idle. The hosted service cannot reach a vCenter inside your network, so vSphere needs the self-hosted install.

Self-hosted

The product itself can also run in your own cluster: one Helm chart in one namespace, your data in your own database, sign-in through the install's own Keycloak. It reads your Kubernetes clusters and your AWS accounts, and needs no AWS account if it only reads clusters. Its licence comes with the first month free; write to us and we set you up. More about self-hosting.

Taking it back

Delete the roles in your accounts and the access ends at once. Your company's page in the product shows how, and deleting the company does the rest: its database, its logins and its settings are gone within 30 days.

Getting started

  1. Sign up

    Your name, your work email and a password, or Continue with Google. Right after the form you sign in with that password, and a link goes to your address. Opening it proves the address is yours and takes you on, signed in. With Google there is no link, since Google has checked the address.

  2. Name your company

    The name everyone in it will see. You become its admin; a company of one is fine. During early access only the addresses on our list can create a company: the page says so, and support@qloudpulse.com adds you.

  3. Connect AWS

    Your start page lists the steps until the first run lands: connect the bill, add the read-only role to each account, wait for the first run. Each step opens the page that does it, and the list says how far you are.

  4. Your reports

    The first report lands within about an hour of the first account passing its check. Everything you see is your company's, and nobody in another company can see it. As its admin you invite colleagues on your company's page, under Settings: they join as members, who see every report, while admins connect accounts and change settings. An admin can make a member read-only, a viewer who reads every report and changes nothing. Everyone is asked to add a second step to their sign-in, an authenticator app or a security key such as a YubiKey, and an admin can require it for the whole company.

Using a report

Every report has the same controls.

  • Time. The range picker sets how far back the page goes: the last 14 days, 2 months, 90 days or 13 months. Every block shows the newest run; select a point on a trend, or a cell on the heatmap, to show that hour or day instead.
  • Compare. The headline says how the figure moved since 1 hour, 8 hours, 24 hours or 7 days ago, and a ledger under it lists what went down and what went up, so the groups add up to the change.
  • Filters. One under the banner for each way the report is split. "Only these" keeps the ticked values, "All but these" leaves them out, and "Clear filters" removes them all. Selecting a bar, slice or node in a chart filters to it.
  • Share it. The address holds the report, the range, the shown hour, the comparison and every filter: copy it from the browser, and a colleague sees the same view after signing in.
  • Your accounts. Accounts I follow, under your name, opens every report filtered to the accounts you tick. The start page counts only them.
  • Your tokens. Your tokens, under your name, makes a token of your own, for 30, 90 or 365 days, to price an infrastructure change with qloudpulse estimate and send its deploy. Any role can make one.
  • Take it with you. Download gives the page as a PDF, and its tables as CSV. Make a copy turns the report into a draft of your own.
  • The (i). In the banner's corner, it explains the report: how its data is collected, how the figures add up, and what each block shows.

Every report is built from the same blocks, in the order its definition gives.

BlockWhat it showsWhat you can do
Summary figuresThe figure at the shown point, with its trend over the week before.Read only.
TableItems ranked by cost. Where the report knows how a cost was made, an item zooms into its parts and their priced lines.Select an item to zoom in or to filter to it; the ↗ opens it in the AWS console.
TrendThe figure over the whole range, stacked by a dimension where the report splits it.Select a point to show that hour or day on the whole page.
Cost flowHow the total splits, column by column, such as account → environment → service.Point at a band to trace it; select a node to filter to it.
Bars and donutThe figure per value of one dimension, or each value's share of the total.Select one to filter to it.
HeatmapThe figure by hour of day and by day, for the last 14 days.Select a cell to show that hour.
Network mapWhat the internet reaches: the ways in, and what they reach inside each account, VPC and subnet.Point at anything to trace it; select it to filter to it.
CalendarDated events on three months, such as scheduled AWS changes.Select a day to go to its event.
To-do listWhat to change to spend less or to fix, largest first: what it is, what to do, what to watch, and what each item is worth. A >_ holds the commands, a preview and the change; the page never runs them.Select a chapter to filter to it, >_ for its commands, ↗ to open the resource.
Savings PlannerWhat a Savings Plan would have done over the range, for a commitment you choose. Nothing is bought.Choose a way to buy and a commitment, or select a point of the curve.
TextA summary, how the figures are made, or a caveat; its bold figures follow the shown point and the filters.Read only.

Missed runs are gaps

An hour whose collection run failed has no figure. The report shows it as a gap, never as $0: the resources were still running and still costing money. A day is the mean of the hours that ran, times 24, so a missed hour does not pull the day down either. A missed hour is a break in the trend, a hollow cell on the heatmap, and the headline says "There was no run". Daily reports, such as the bill, have no hours to miss.

Collection and alerts

Collection, in the product's top bar, shows the jobs that gather your company's data, when each last ran and how it went, and lets you start one now with Run now. The dot beside it says the same at a glance: grey until a run has happened, pulsing while one runs, red when a job's last run failed, green when the last runs succeeded.

Alerts, under Settings, sends an email, a Slack message or a PagerDuty alert when a service's cost in one of your accounts jumps against the same hour a day before and a week before, naming the resources that rose most. Because the run rate is collected every hour, the check runs every 15 minutes, and an alert goes out within about half an hour of the hour that jumped. A jump has to pass both a percentage and an amount a month, 20% and $100 unless you change them, and a drop never alerts. Instead of one message per hour, you can take a digest each morning.

The same channels tell you when a budget's spend or projection passes the percentages you set for it, and, once a day, when the inventory finds a new resource failing a rule you chose to hear about. Every alert is your own: you pick the kinds, your thresholds and your finding rules, and where they go, your own address and up to five channels of your own, an email address, a Slack webhook or a PagerDuty key. Nobody else sees your channels, and nobody needs to be an admin to set them up.

Budgets and unit costs

Budgets, in the top bar, sets a monthly amount on some accounts, some services, both, or everything, with an amount of its own for any month that needs one. Each budget shows what the month has spent so far and what it will end at, from the bill where AWS has settled it and the run rate after that, and says whether it is on track, at risk or over. Your company's admins add and change budgets; everyone in the company sees them.

Unit costs, beside it, divides that same cost by a number your business counts, such as orders, customers or thousands of API calls, so you see what one of them costs in AWS. An admin names the metric, gives it the same kind of scope as a budget and, if wanted, a target, then pastes its values, one day or month per line, as they come out of a spreadsheet. The page shows the latest cost per unit and how it moved, a chart over the last 90 days or 12 months with the target as a line, and the cost, units and cost per unit of every period. A period with no value is a gap, never a zero. A budget and a unit cost on the same scope agree to the cent.

Building your own report

Anyone in your company can build a report on the data that is already here, with New report on the start page or Make a copy on any report. No code: it is saved in qloudpulse and listed straight away.

  • Data from: a published report. Yours can use its dimensions and measures, and no others.
  • Blocks: add them, in sections if you like, and order them with the arrows. The (i) on each block explains its settings.
  • Preview: below the form, on real data, drawn as the saved report will be.
  • Save: if something cannot be drawn, you are told what to fix and nothing is saved.

Everyone in your company can open your report and see who built it; only you can change or delete it. If the source report later changes under it, yours says what is missing, and Edit repairs it.

Your data

  • Read-only, by construction. The roles you create can only read, and deny reading what is stored. You can read every line of their policies before you run the scripts, and check them with tools you trust.
  • Apart from every other company. Each company has a database of its own, and every request is checked against the company of the person signed in. Reports hold totals about your infrastructure, never data about individual people.
  • In the EU. The hosted plan runs on AWS in Ireland, and your data is stored there. Self-hosted, it stays in your own cluster and your own database.
  • Recorded. Every change to a company's users and settings, and every look our staff take at a company's data, goes to an audit trail that nobody can change. We are working towards SOC 2.
  • Yours to take. Your own scripts read every report through the qloudpulse API, as JSON or CSV, and take your whole bill as FOCUS or everything you have in one zip, with a key of your own, in every plan.
  • Yours to end. Delete the roles and the access ends at once; delete the company and its data is gone within 30 days. The privacy policy has the rest.

Questions

What is the run rate, and what is it not?

What your running resources cost at the moment they were collected: usage × the AWS list price, excluding VAT, per hour. That is what we mean by real time: a run every hour, so the figure on the page is never much more than an hour old, where the bill is a day or more behind. What cannot be measured per hour, such as data transfer, snapshots and request charges, is taken from the bill, two days behind, and monthly fees are spread over the hours of their month. It is not the bill: credits, refunds, tax and discounts are only on the bill.

Why does the bill lag, and why does it change?

AWS delivers the billing export about a day late and keeps revising the current month until it closes. The bill report shows days, and its recent days can still move.

Why does a resource have no ↗ link?

The ↗ opens the resource in the AWS console, in the account that holds it, so sign in there as you usually do. There is none where a line names no single resource, such as data transfer, where the console has no page for it, or where one name is several resources in different accounts: filter to one account and it appears.

How do I join a company someone else created?

Ask one of its admins to invite your address. The email's link works for 12 hours: open it, give your name, verify your address and choose a password, and you land in that company's reports.

I work for two companies

A login belongs to one company, so sign up a second time with the other company's address.

Can qloudpulse staff see my company?

During early access, staff can sign in as one of your company's users to help you, and every such look is recorded. That ends with early access. Otherwise only your company's own logins reach its reports.

Something missing here? Write to support@qloudpulse.com, or ask for it with Suggest a feature in the product: every request is read and answered.