Skip to content
Nubent

/Observability and SRE

AWS observability and SRE: reliability measured with SLOs

We implement end-to-end observability on AWS and Site Reliability Engineering practices so you always know how your system is doing from the user's point of view. Metrics, logs and traces with CloudWatch, OpenTelemetry and Grafana, SLOs with error budgets, noise-free alerts and an incident process that learns.

/Diagnosis

Signs it is time

Customers tell you first

Production problems surface through support tickets or social media first, and only then does someone dig through the logs to figure out what happened.

Alert fatigue

There are hundreds of alarms, most get ignored, and the important ones get lost among false positives that wake the team up for nothing.

Debugging takes hours

Without distributed tracing or correlation between logs and metrics, finding the cause of an error that spans several services is painstaking manual work.

No agreement on reliability

Product wants more features, engineering wants stability, and there is no objective way to decide when to slow down and when to push ahead.

/What we deliver

Concrete deliverables, in your repository

  • Observability architecture

    Design of metrics, logs and traces with OpenTelemetry, destinations in CloudWatch or Amazon Managed Service for Prometheus, and a retention and cost policy.

  • Service instrumentation

    ADOT or OpenTelemetry SDKs in your applications, distributed tracing, structured logs and correlated business metrics.

  • SLOs and error budgets

    Service level indicators for each critical user flow, targets agreed with product, and error budget policies.

  • Dashboards and actionable alerts

    Dashboards in Amazon Managed Grafana or CloudWatch and alerts based on SLO burn rate, each one with its runbook.

  • Incident process

    Roles, severity levels, communication channels, a blameless postmortem template and tracking of corrective actions.

/How we work

Our process

  1. Observability assessment

    We review what is measured today, which alerts exist, what telemetry costs and how recent incidents were handled.

  2. Defining SLOs

    We identify your users' critical flows and agree on realistic indicators and targets with product and engineering.

  3. Instrumentation and alerting

    We instrument with OpenTelemetry, build dashboards, replace noisy alarms with SLO alerts and write runbooks.

  4. Ongoing SRE practice

    We support incidents, facilitate postmortems, review error budgets and prioritize reliability work with your team.

Observability on AWS: beyond monitoring

Monitoring answers questions you already know to ask: whether CPU is high or a service is responding. Observability lets you answer new questions about behavior you did not anticipate, by combining metrics, logs and traces with business context. In distributed systems on AWS, with containers, Lambda, queues and managed services, that capability is essential.

Our approach is built on OpenTelemetry as the open instrumentation standard, so your code is not tied to a vendor. With AWS Distro for OpenTelemetry we send telemetry to CloudWatch, X-Ray, Amazon Managed Service for Prometheus or whatever tool you already use. CloudWatch Application Signals adds a service-centric view with built-in SLOs for workloads on EKS, ECS and Lambda.

SLOs and error budgets: reliability as a business agreement

An SLO is a measurable reliability target for an important user flow, for example that checkout responds correctly under a latency threshold for the vast majority of requests over a 28-day window. The indicator being measured is the SLI, and the gap between the target and perfection is the error budget.

The error budget turns reliability into a shared decision. While budget remains, the team can deploy more often and experiment. Once it is spent, stability work takes priority. This agreement ends subjective arguments between product and engineering and gives everyone a clear basis for planning.

Alerts change too. Instead of paging on every CPU spike, we alert when the error budget burn rate shows the target is at risk. A 3 a.m. page now means a user is affected, or soon will be.

Incident management and blameless postmortems

A good incident process defines severity levels, roles such as incident commander and communications lead, dedicated channels and customer communication templates. With that in place, nobody improvises when something breaks and mitigation starts right away. We integrate alerts with PagerDuty or Opsgenie, and runbooks live next to the code.

After every significant incident we facilitate a blameless postmortem: a timeline, the impact, contributing causes and concrete actions with an owner and a due date. The focus is on changing the system so the same problem does not happen again, not on finding someone to blame. The actions are prioritized alongside the rest of the backlog.

Keeping observability costs under control

Telemetry can become one of the most expensive lines on the bill if nobody manages it. Overly verbose logs, high-cardinality custom metrics and indefinite retention are the usual culprits. We review which data is generated, which is actually queried and what each one costs.

We apply log classes based on query frequency, smart trace sampling, retention tiered by data type and metric aggregation before ingestion. The goal is to have the information you need to operate with confidence without paying for data nobody looks at.

/FAQ

Observability and SRE: frequently asked questions

Another question? Write to hello@nubent.com and an engineer will answer.

What is the difference between monitoring and observability?

Monitoring watches known indicators and alerts when they cross a threshold. Observability lets you understand why a system behaves the way it does, even in the face of new problems, by combining metrics, logs and traces with context. Monitoring is part of observability, but it is not enough to debug complex distributed systems.

What is an SLO and how do you define one?

An SLO, or service level objective, is a measurable reliability target for a service, expressed as the percentage of good events over a time window. You define it from critical user flows, choosing indicators such as availability or latency and agreeing with product on a realistic target, which does not have to be one hundred percent.

Should you use OpenTelemetry on AWS?

Yes, in most cases. OpenTelemetry is an open standard that keeps your instrumentation from being tied to a vendor, and AWS supports it through AWS Distro for OpenTelemetry, CloudWatch and X-Ray. That means you can switch or combine analysis tools, such as CloudWatch and Grafana, without re-instrumenting your applications.

CloudWatch or Grafana for observability on AWS?

They are not mutually exclusive. CloudWatch natively collects metrics, logs and traces from almost every AWS service and offers Application Signals with SLOs. Amazon Managed Grafana is very flexible for dashboards that combine multiple sources, such as Prometheus and databases. A common setup uses CloudWatch as the source and Grafana as the visualization layer.

What is a blameless postmortem?

It is a post-incident analysis that focuses on how the system and processes allowed the failure, without singling out a person. It documents the timeline, the impact, contributing causes and improvement actions with an owner and a due date. It encourages the team to share information honestly and learn from every incident.

Next step

Let's talk about your Observability and SRE project

30 minutes with an engineer. Concrete next steps, whether or not you work with us.