Skip to content
Technology transformation expertise firm · ARDIANE Group Innovate, secure, succeed

Cloud and platforms

Observability and SRE: managing reliability through service level objectives

Reliability cannot be decreed and is not measured by the number of green lights. We help you define it from your users' point of view, measure it and make it a criterion for trade-offs between new features and stability.

Our principles

  • Service objectives negotiated with the business
  • Alerts that call for action, and only those
  • Incidents the organisation learns from
In brief

SRE (Site Reliability Engineering) applies engineering methods to operations in order to maintain an agreed level of reliability, and observability provides the signals that make it possible to understand the real state of a system. Management Square helps organisations define their service level objectives, instrument their applications and structure incident management.

Scope

How we contribute

  1. Defining SLIs and SLOs

    Choice of critical user journeys, of the indicators that reflect their quality and of the objectives to be met, agreed with product owners and not by operations alone.

  2. Error budget policy

    A written rule, accepted by the business and technical sides, stating what happens when the budget is eroded or exhausted: slowing down changes, strengthening testing, addressing the cause.

  3. Instrumentation and observability

    Consistent collection of metrics, logs and traces, with an open standard such as OpenTelemetry, and correlation of the three signals to shorten diagnosis.

  4. Alert redesign

    Review of existing alerts, removal of those that call for no action, alignment of the others with error budget consumption.

  5. Incident management

    Severity levels, roles during the incident, coordination channel, communication to users, organisation of a sustainable on-call rota.

  6. Blameless postmortems

    A lessons-learned method focused on facts and contributing causes, with actions that are assigned, dated and tracked through to closure.

Page contents

When monitoring is green and users are complaining

The need for SRE and observability most often emerges from a well-known scene. The monitoring dashboards show no anomaly, the servers respond, the disks are not full. Meanwhile, the support desk is receiving calls: payments are failing, pages take too long to display, a batch job has not produced its result.

The gap comes from what is being measured. Traditional monitoring looks at components. The user, however, experiences a journey that crosses many components, and no indicator light describes that journey. As long as reliability is not defined from the user’s point of view, operations and the business are not talking about the same thing.

A second difficulty adds to this: trade-offs. Product owners want to release quickly, operations want stability. Without a common criterion, every production release is negotiated as a trial of strength.

SLI, SLO, error budget: the vocabulary that changes the conversation

SRE provides that common criterion. Three notions are enough to get started. Our glossary entry on the definition of SRE presents them in detail.

The service level indicator (SLI) measures what the user experiences. Its most useful form is a ratio: the number of satisfactory events divided by the number of valid events. For example, the proportion of payment requests completed in less than one second.

The service level objective (SLO) sets the value to be met for that indicator over a defined measurement window. It is negotiated. An objective that is too low dissatisfies users, an objective that is too high is costly and rules out any change. A 100% objective makes no sense: it assumes that nothing will ever change.

The error budget is the complement of the objective. With a 99.9% objective, the budget is 0.1% of events over the window. This budget is consumed, by incidents as well as by risky production releases. It turns a quarrel of principle into a management rule: as long as budget remains, we release; when it is exhausted, we stabilise.

Metrics, logs, traces: what observability adds

Measuring an objective is not enough. When it degrades, you need to understand why, quickly. That is the role of observability, which rests on three complementary signals.

Signal What it provides Its limit Main use
Metrics Numerical values aggregated over time, inexpensive to retain No detail on any particular request Trends, alerts, SLO calculation
Logs The timestamped detail of each event High volume and cost, hard to exploit if not structured Fine-grained diagnosis, evidence, audit
Traces The path of a request across services Requires instrumenting the code, sampling needed Locating slowness or an error in a distributed system

The value comes from correlation. An alert on a metric should lead to the trace of a faulty request, then to the logs of the service concerned, without switching tools three times or copying an identifier by hand. An open standard such as OpenTelemetry facilitates this consistency and limits dependence on a vendor.

Managing the incident, then learning from it

A badly run incident lasts longer than it should, for organisational rather than technical reasons: nobody coordinates, ten people work on the same component, leadership questions the engineers busy making repairs.

Our consultants help your teams put a simple framework in place. A severity scale states who is mobilised. Three roles are distinguished: the person who coordinates, the person who leads the technical diagnosis and the person who communicates. A single channel concentrates exchanges and serves as the timeline. This framework complements, without replacing, the incident and problem management processes derived from ITIL, where they are in place.

After the incident comes the blameless postmortem. It establishes the facts, looks for contributing causes rather than a single cause, and concludes with assigned and dated actions. The absence of blame is a condition of accuracy: a person who fears sanction does not recount what they actually did.

Our SRE and observability approach: start from a journey

Our consultants work within your teams; decisions remain yours.

Choose a critical journey. We start from a service whose unavailability is visible, with a product owner ready to commit. Two or three indicators are enough to describe it.

Measure before setting. We record the level actually achieved over a past period. The objective is set on the basis of that finding and of users’ expectations, not of a figure borrowed from elsewhere.

Write the error budget rule. Who decides to slow down releases, and on what signal? The rule is signed by the business and technical sides before the first breach.

Align the alerts. Alerts are rebuilt around the rate of budget consumption. Those that trigger no action are removed.

Extend, then reduce toil. The following services adopt the same model, integrated into the golden paths of the platform described in our DevOps and platform engineering offering. Recurring manual tasks are inventoried and automated in order of frequency.

What drains an SRE approach of its meaning

  • Objectives everywhere. Setting an SLO for every component recreates the monitoring you wanted to move beyond. Objectives relate to what the user sees.
  • An objective without consequence. An SLO whose breach changes no decision is just one more indicator.
  • Mistaking the tool for the practice. Buying an observability suite produces neither relevant indicators nor instrumentation of the code.
  • Alert fatigue. An on-call rota woken for alerts that lead nowhere ends up no longer reading them, including the one that matters.
  • Postmortems without follow-up. A lessons-learned review whose actions are never carried out teaches teams that the exercise is a formality.

Major incidents, which involve business continuity and crisis communication, fall within a broader set-up described on our page on cyber resilience.

How to measure progress

We track observable signs with you, without promising a target value:

  • the share of critical services with an SLO, an owner and an error budget rule;
  • the share of incidents detected by an alert before being reported by a user;
  • the share of alerts that led to an action;
  • the evolution of detection time and recovery time;
  • the number of on-call call-outs outside working hours;
  • the share of postmortem actions closed by the planned deadline.

The most telling sign is not quantified: it is the moment when a product owner checks the error budget of their own accord before scheduling a production release.

Typical work

What our consultants contribute to

SLO catalogue

For each critical service: indicators, objectives, measurement window, owner and data source.

Error budget policy

The thresholds and the decisions they trigger, approved by the business and technical owners.

Incident management guide

Roles, severity levels, escalation procedures and communication templates.

Postmortem template and action log

The lessons-learned format and the tracking of the actions that result from it.

Management Square in figures

300+
projects completed
43+
clients supported
Since2015
serving our clients
  • Allianz
  • Euler Hermes
  • Groupe BPCE
  • ODDO BHF
  • Candriam
  • Ledger
  • Saur
  • Veolia
  • Voltalia
  • emeis
  • Bonpoint
  • Legal Suite

FAQ

Frequently asked questions

Another question? Ask us

What is the difference between an SLI, an SLO and an SLA?

The SLI (service level indicator) is a measure of the quality perceived by the user, for example the proportion of requests served correctly. The SLO (service level objective) is the target set internally for that indicator over a given period. The SLA (service level agreement) is a contractual commitment to a customer, generally less demanding than the SLO and accompanied by consequences in the event of a breach.

What is an error budget?

The error budget is the share of failure that a service level objective tolerates over a period. If the objective is for 99.9% of requests to succeed, the error budget is 0.1% of requests. As long as it has not been consumed, the team can release changes. When it is exhausted, priority returns to reliability.

What does a blameless postmortem mean?

A blameless postmortem is a lessons-learned review that seeks to understand how the system and the organisation made the incident possible, rather than to identify a culprit. People describe their actions without fear of sanction, which produces an accurate timeline. Blameless does not mean without follow-up: each corrective action has an owner and a deadline.

Should you create a dedicated SRE team?

Not necessarily. SRE practices can be carried by product teams with the support of a few specialists, by a central team, or by a mixed model. The choice depends on the number of critical services and the maturity of the teams. An operations team renamed SRE without any change of mandate brings none of the expected benefits.

A project to bring to completion?

A first conversation is enough to understand your challenge, identify the right people and tell you frankly what we can bring.