Observability and SRE: Know when your users feel a problem
We agree an objective for checkout, sign-in and whatever else earns money, and page a person only when one is at risk.
Alert on what your users feel
Budget is left, and releases go out as planned.
- Your users
- Load balancer
- App
- Database
- Your users
- Load balancer
- App
- Database
Budget is left, and releases go out as planned.
A ticket for working hours. Nobody is woken.
The on-call engineer is paged and the runbook opens. If the budget runs out, releases wait.
What happened, without blame. Each action has an owner, and the alarm is tuned.
The objective is yours. We help you set it and measure it, and we do not promise it.
Four signals on every service
Latency
How long a request takes, failures timed apart.
Traffic
How much demand, in requests or sessions.
Errors
Failed requests, wrong answers included.
Saturation
How full the tightest part is.
- Gateway
- Orders
- Payments, the slow step
- Database
Instrumented with OpenTelemetry, so the same data can go to your cloud’s tools or to one you already pay for.
The tools, on each cloud
- Collect
- AWS Distro for OpenTelemetry, into Amazon CloudWatch and AWS X-Ray
- Objectives and alarms
- CloudWatch Application Signals, with burn-rate alarms
- AI first pass
- AWS DevOps Agent
Where you already run Prometheus and Grafana, we use each cloud’s managed Prometheus, and managed Grafana on AWS and Azure.
Customer work
- Managed Services
Prometheus, Jaeger and Grafana, with alerts in Microsoft Teams.
31 ms at p95Order serviceShwapnoRetail chain, online store, BangladeshRead the story: Shwapno (opens in a new tab) - Managed Services
Investigative tools and live metrics found the cause, then the server was tuned.
Problems resolvedDakterbhaiHealthcare platform, BangladeshRead the story: Dakterbhai (opens in a new tab) - Audit & Compliance
Resource monitoring, scaling and load balancing checked in a review.
A review across all six pillarsNorwegian bus operatorPublic transport, NorwayRead the story: Norwegian bus operator (opens in a new tab)
Where you will meet it
- Cloud Build & Launch (opens in a new tab)Monitoring and alarms are set before launch, from real load.
- Migration & Modernization (opens in a new tab)Each wave is watched through cutover, with a tested way back.
- Managed Services (opens in a new tab)An alert starts the AI first pass, and a named engineer answers.
- Audit & Compliance (opens in a new tab)Monitoring and alerting are checked against best practice.
Questions
Do you replace our on-call team?
No. An alert starts the AI first pass at any hour, an engineer answers at the service level you choose, and your team keeps the runbooks.
Do you promise uptime?
No. You set the objective with us and we measure it. On Managed Services we promise a first reply to each severity, and a first reply is not a fix.
Does the cloud’s AI agent change anything on its own?
Not in our setup. It proposes and a named engineer approves. On Azure we set the SRE Agent to Review mode and its other tools to ask.