Skip to content

Prices from your cloud bill

Out of the box, Seylo prices your cluster at public list prices. If you pay something else - reservations, savings plans, a negotiated discount - those figures overstate your cost. With a cloud account connected, Seylo reads what you actually pay from your own bill.

Cloud Machines Disks Read through
Azure Each machine type Managed-disk tiers Azure Cost Management, with the Seylo Reader role
AWS Each EC2 instance type EBS volume types AWS Cost Explorer, with one extra permission
Google Cloud Each machine type the bill names Persistent Disk types Your Cloud Billing export in BigQuery, which you turn on

The rules are the same for all three:

  • Your bill wins. For the same machine or disk type and region, a price from your bill replaces an uploaded rate and the public list price.
  • The public list fills the rest: a type that did not run in the window, an account that is not connected, a region not read.
  • Amortised. A reservation or savings plan is spread over what it covers, so a reserved machine is neither free nor charged in one day. On Google Cloud, committed-use discounts are left out instead, and shown beside the rate.
  • In US dollars, as the cloud states them. Seylo converts nothing; on Google Cloud it uses the US-dollar figure the export itself states.
  • Never invented. A charge with no dollar figure, no quantity, or a type Seylo does not recognise prices nothing, and is never priced from a similar type. The card counts each one left out, with the reason.

The Price book card on FinOps → Cost & savings → Cost then reads Prices: from your Azure bill, from your AWS bill, from your GCP bill, or a combination. Disk prices from your bill are also what Nightshift uses to state the cost of growing a nearly full volume.

Seylo reads Azure Cost Management with the same role you granted for the cloud account. There is nothing to install, no export and no key. It reads every connected Azure subscription whose last Test said Connected, again every six hours. How machine prices are read is on make the figures match your invoice.

Managed disks are priced per tier, from the same bill, over the same window. A tier’s monthly rate is used only from a charge Azure names exactly like a disk tier: P10 LRS Disk, E10 ZRS Disk, S10 LRS Disk and the other Premium SSD, Standard SSD and Standard HDD sizes, locally or zone redundant. These are not used as a tier price, and are listed on the card:

  • transactions, bursting, snapshots, and the shared-disk mount charge;
  • Premium SSD v2 and Ultra Disk, which bill per GiB, IOPS and throughput rather than per tier;
  • any charge Seylo does not recognise.

A disk’s region comes from Azure Resource Manager, not from the bill’s display name for it. If the disk read fails, machine prices and the bill are unaffected, and disks use the public list. The Your Azure bill row says, per subscription, how many machine and disk prices it read, for example Acme Production: 6 machine prices and 4 disk prices from the bill.

The connect stack adds one inline policy to Seylo’s role, seylo-read-costs, which allows ce:GetCostAndUsage and nothing else:

{"Version":"2012-10-17","Statement":[{"Effect":"Allow","Action":"ce:GetCostAndUsage","Resource":"*"}]}

AWS’s ViewOnlyAccess has no Cost Explorer action, so a stack deployed before this release cannot read the bill. Save the new template from the cloud account’s card and run the same aws cloudformation deploy command; see move an existing connection.

So Seylo reads as little as it can:

  • Once a day per account. Cost Explorer updates its data at least once every 24 hours, so reading more often buys nothing. A good read is kept for a day.
  • One query per region named on the connection (at most 5), with at most 5 pages each.
  • Plus one request each time you press Test.
  • A failed read is tried again with the six-hourly pass, never more often, and a refused one is not retried.

A steady account with one region costs about $0.01 a day, plus $0.01 per Test.

Over the 7 complete days before yesterday, for this account only (a payer account does not price from its members’ usage), credits and refunds left out:

Used Left out, and listed on the card
Instances Linux/UNIX instance hours: on-demand, and those covered by Reserved Instances or Savings Plans. Spot hours are priced apart, and a type’s spot price is used only where that type ran nothing but spot Windows and other licensed operating systems, Dedicated and Host hours, unused reservations, EBS-optimised hours, Local Zone usage
EBS volumes Per GB-month storage of gp3, gp2, io2, io1, st1, sc1 and magnetic (standard) Provisioned IOPS, provisioned throughput, I/O requests, snapshots, fast snapshot restore

A volume from the EBS CSI driver (ebs.csi.aws.com) is priced at its type’s rate times its size. When its StorageClass names no type, the driver’s default, gp3, is used. gp3 IOPS and throughput above the included 3,000 IOPS and 125 MB/s are not in the figure, and the cost view says so. io1 and io2 volumes are not priced, because their IOPS charge would be missing; the cost view lists each unpriced volume with the reason.

Nodes in a region not named on the connection use the public list. The AWS bill’s day-by-day history is not read yet, so whole-cloud cost dashboards stay Azure only.

The account stays connected, prices use AWS’s public list, and the Your AWS bill row and the Test’s cost-read step (badge list prices) say why:

You see What to do
Cost Explorer read is not allowed for role ... Redeploy the stack from the card, which adds seylo-read-costs
... refused ... by a service control policy of your AWS organisation (an explicit deny). Your AWS organisation’s administrator allows ce:GetCostAndUsage for this role
... refused ... by an explicit deny (a permissions boundary or another policy on the role). Remove the deny on ce:GetCostAndUsage
... has turned off Linked Account Access in Cost Explorer's preferences ... Someone with the management (payer) account turns it on
Cost Explorer is not enabled for account ... Someone with billing access opens Cost Explorer once in the AWS console. AWS does not allow enabling it by API; data is ready about 24 hours later
Cost Explorer has no data for account ... in this window yet ... Nothing: Seylo reads again tomorrow
AWS is limiting how often account ...'s costs are read ... Nothing: Seylo tries again later

Google Cloud: machines and Persistent Disks from your billing export

Section titled “Google Cloud: machines and Persistent Disks from your billing export”

Google has no API that says what a machine cost you. What you were charged is in the Cloud Billing export to BigQuery, which you turn on for your billing account. Seylo queries it through the same keyless connection, once a day.

If you have no export yet, Test says so and gives these steps:

  1. In BigQuery, create a dataset. Google recommends the US or EU multi-region.
  2. In the console, Billing → Billing export → BigQuery export → Standard usage cost → Edit settings, choose that project and dataset, and Save. This needs Billing Account Costs Manager or Billing Account Administrator, and BigQuery User on the project.
  3. Data appears within a few hours. A multi-region dataset also gets the current and previous month; a single-region dataset only from the day you turn it on.

The standard and the detailed export both work. Google names the table gcp_billing_export_v1_<billing account ID> (standard) or gcp_billing_export_resource_v1_<billing account ID> (detailed).

The Google Cloud form has an optional Billing export table field, project.dataset.table. Leave it empty and Test looks for a table with Google’s export name in the connected project; with more than one, it asks you to name one.

The connect commands enable the BigQuery API with the others. No new project role: Viewer already lets the service account run a query in the project and read its datasets. When you name a table, for example one in another project, the commands add one grant, read on that table only:

Terminal window
bq add-iam-policy-binding \
--member="serviceAccount:seylo-reader@<project ID>.iam.gserviceaccount.com" \
--role="roles/bigquery.dataViewer" --table=true \
<export project>:<dataset>.<table>

So Seylo reads as little as it can:

  • Once a day per project, a week of that project’s Compute Engine lines.
  • Only the days it needs. The query filters on the export’s daily partitions, never the table’s whole history.
  • At most 10 GiB a read. A read that would scan more fails, and BigQuery does not charge for it.
  • Test is a dry run, which BigQuery does not charge for. It says how much the daily read would scan and what that costs.

Over the 7 complete days before yesterday, for the connected project only:

Used Left out, and listed on the card
Machines Each machine type the bill names, priced from what you paid for its vCPUs and memory in that region, sustained-use discounts included Committed-use discounts (the commitment’s own fee is billed apart), free tier, promotions, spending-threshold discounts and other credits: their total is shown beside the rate
Persistent Disks Per GiB-month: pd-standard, pd-balanced and pd-ssd, zonal and regional Extreme Persistent Disk and Hyperdisk, which also bill provisioned IOPS and throughput

A machine type or region the bill does not show uses Google’s public list. The Your GCP bill row on the invoice card says, per project, how many machine and disk prices it read.

The project stays connected, prices use Google’s public list, and the Test’s cost-read step says why:

You see What to do
No Cloud Billing export table was found in project ... Turn on the export (above), then name the table and press Test
Project ... has ... billing export tables ...; Seylo will not pick one. Name the one to read and press Test again
The BigQuery API is not enabled in project ... Run the gcloud services enable command in the message, then Test
Seylo's service account may not read ... Run the bq add-iam-policy-binding command in the message, then Test
Seylo's service account may not run a query in project ... Check that the service account still holds roles/viewer and that no deny policy removes bigquery.jobs.create
... does not exist or Seylo cannot see it. Check the table’s name in BigQuery
... does not have the columns of a Cloud Billing export ... Name the table Google’s billing export writes
Reading a week of ... would scan more than Seylo's 10 GiB limit ... Tell Seylo support; prices stay on the public list