Deckard
Contact
Contact

Cloud setup you can rebuild

Cloud architecture and DevOps

Your cloud written as code, with each change tested and approved before it goes live. Changes ship more often, and new features reach paying customers sooner.

Start a conversationSee it step by step
Production change, reviewedIn production
Plan: add 3, change 1, remove 1
Diff posted on the pull request5 resources
Tests on the pull request
Merge blocked until they passPassed
Policy checks: tags, encryption, ingress
Each resource tagged to the system that paysPassed
Second-engineer approval
Authors cannot approve their own planApproved
Apply from the main branch
Recorded with the plan it appliedReversible

Each night, the live setup is checked against the code.

A production change, checked by two people

TodayLive servers, set up by hand in a web console

Known only to whoever built it

01Write
AddsA database copy
Files changed2
Other changesWait their turn
02Check
Checks passed14:12
Cost estimate+$212 a month
Approved byPlatform lead, 14:29
03Go live
Live at14:32
UndoOne step
Changes this quarter41, all reviewed
OutRebuilt from codeRepeatable
Last rebuildStaging, Sep 14, 22 min
Hand edits0 this quarter
EnvironmentsTest, staging, live

Three layers of protection around your data

Cloud architecture as three trust rings: public edge, private compute and the data core, each entered at one gate, with one request crossing all three and an audit log outside the rings One request from users enters the public edge at the load balancer, private compute at IAM, and the data core at the database proxy, passing the services on its way to the database. Inside private compute the services also reach the workers and the queue without crossing a gate. Each gate the request crosses writes who and when to the audit log, which sits outside all three rings.
  1. A mistake stays in its ringThe public edge, private servers and the data core each have their own permissions.
  2. Each ring has one way inThe load balancer, access rules (IAM) and the database proxy are the only ways in.
  3. One request, logged at each gateEach gate a request crosses writes who and when to the audit log.

Set up in your cloud

The setup is written in Terraform or Pulumi, for Google Cloud, AWS or Azure.

  1. 01New cloud designs, and reviews of what already runs
  2. 02Moves to Google Cloud, AWS or Azure, with a way back
  3. 03Every environment defined in code, from test to live
  4. 04Automatic releases that stop when a test fails
  5. 05Cost tagged to the system that spends it
  6. 06A disaster recovery drill you have run
Yours at the end

infra/

terraform/7 modules: network, servers, data

dr/Recovery steps and the drill script

ci/Tests, policy checks and release steps

cost/Tag rules and monthly budget alarms

See it work

One reviewed change to production

How it goes live

  1. 01 · Optional

    Architecture review

    What runs today, what it costs and what a mistake reaches

  2. 02 · Optional

    Pilot environment

    One service defined in code, beside the one you run now

  3. 03 · Cloud and DevOps

    Move with a way back

    Every environment in code, moved one service at a time

  4. 04 · Ongoing

    Day-to-day running

    Your engineers run it, or we do

Related services

The production change, used by three other services

Each approved change is filed as evidence for the auditor.

Security and compliance engineering

Each change is watched closely for an hour once live.

Infrastructure monitoring and SRE

Hosts every customer account on the same version.

SaaS product development

All solutions

By industry

A recovery drill, a cloud bill and a weekend sale, all in code

Portal failover drillQuarterly
Portal rebuilt at site 2
7 modules, no console steps02:11
Database restored
From last night’s backup02:18
Traffic switched to site 2
DNS and load balancer moved02:24
One secret still set by hand
Vaulted before the next drillFix logged

The drill runs from the code every quarter.

Data centers and colocation

Cloud spend by clientMonth to date

$4,180

Spend in each client’s own account, tagged to its engagement, Sep 1 to 23.

Tagged

11 of 11

Untagged

$0

Sandboxes torn down

3

Professional services and consulting

Weekend saleBefore the freeze
Change opened as a pull request
Adds 2 checkout instances
Tests and policy checks pass
Plan read by a second engineer
3Freeze starts Friday
Held for the platform lead’s approval
4Apply, then scale back Monday
Both by pull request

During the freeze, every change needs a named approver.

Retail and e-commerce

Put your cloud setup in code

The scope comes with a plan to undo each step.

Start a conversation