Observability & SRE

Alerts that matter,
answers you can find

We tie your logs, metrics and traces together, agree reliability targets with your team and tune alerts to what users feel, so your team is called only when there’s something to do.

Thanks, we’ll be in touch soon.

{{ heroStatus }}

{{ heroSay }}

You’re in good company

A few of the companies we’ve worked with.

{{ wallSay }}

Fine on the dashboard, broken for a customer

Most monitoring watches machines: is the server up, is the disk full, is the database answering. Every one of those can be fine while a customer can’t sign in or pay, because what failed sits between them. So the first report comes from a customer. Then the search starts, across separate tools, matching lines by the clock and guessing which belong to the same request. And the alerts that do fire are mostly about machines. Most need nothing done, so the people on call, the ones who answer when an alert calls, whatever the hour, learn to doubt them.

Two reports of the same moment, in an example.

What monitoring says

  • Servers: fine
  • Database: fine
  • Network: fine
  • Disk space: fine

What a customer says

I can’t pay.

What we put in place

The parts that let a team see what’s happening, and decide what to do about it.

  • What users get, measured

    For each journey that matters, one thing a user comes to do, such as signing in or paying: whether it worked and how long it took, measured as close to the user as it can be.

  • Logs, metrics and traces, tied together

    One ID travels with each request through every service, so its logs, timings and errors line up as a trace: the record of one request’s path through every service it touched.

  • Instrumented to an open standard

    Your code reports what it’s doing in a standard format, so the data goes to the tools you choose, and moves with you if you change them.

  • Targets agreed with product

    A reliability target for each journey (often called an SLO): how often it should work, and how quickly, agreed ahead and short of perfect on purpose, with a written rule for what happens when it’s missed.

  • Alerts that mean act now

    An alert calls a person only when users are affected, or soon will be, and there’s something to do now. Everything else waits for working hours.

  • A first step for every alert

    Every alert that calls someone has a first step written down: what it means, what to check first and who to call next (a runbook), written with the people who answer it.

  • Kept for a reason

    What’s kept, in how much detail and for how long, decided by the questions it answers and what it costs, with personal data and secrets left out.

  • Incidents run with clear roles

    One person leads, one keeps everyone else informed, and what’s tried is written down as it happens.

  • A review after every incident

    What happened, what users felt, what changes and who owns each change, in plain words and without blame. It names systems and decisions, never a person at fault.

Hosting, releases and security? Cloud, DevOps & Security →

Software looked after, with agreed response times? Support & Maintenance →

Pages and screens that load slowly? Performance & Optimization →

A target short of perfect, on purpose

No system works every time, and each step closer to perfect tends to cost more than the one before. So for each journey we agree a target with your product team that’s short of perfect, on purpose. The gap between the target and perfect is an allowance (often called an error budget) your team can spend: on releases, experiments and the unexpected. While there’s allowance left, releases go out. When it’s going fast, someone is called. When it’s gone, reliability work comes first, by a rule everyone agreed before it was needed.

A target is for your team. What you promise customers in a contract (an SLA) is a separate decision, and yours.

A line from never works to always works. The target is marked a little short of always works, and the short stretch between the two is the allowance.
What the allowance decidesAn example
When the allowance isWhat happens
Mostly leftReleases go out as usual
Going fastSomeone is called now
Going slowlyA ticket, for working hours
GoneReleases pause, apart from urgent and security fixes. Reliability work comes first.

Written down, and agreed by engineering and product, before the first alert goes live.

One request, in one piece

When something is slow, the question is where. Each service can answer for itself, but the customer’s request passed through several. Followed in one piece, it shows where the time went.

An exampleA customer tracks a delivery

The request passed through several parts of the system, each called by the one before it. Each kept its own record, and every record carries the same ID.

  1. Website. Its record: Asked for the delivery’s status. Its share of the wait: a moment. Same ID
  2. Orders. Its record: Found the order. Its share of the wait: a moment. Same ID
  3. Deliveries. Its record: Asked the courier where it is. Its share of the wait: a moment. Same ID
  4. The call to the courier. Its record: Waited for an answer. Its share of the wait: most of the wait. Same ID

Most of the wait was here.

Each service keeps its own record. One ID, attached when the request arrives and passed along with every call, ties them into one.

The ID travels in a standard header that tools from different makers read the same way, so services written by different teams can pass it along.

We instrument to an open standard (the OpenTelemetry® specification), so the data isn’t tied to one tool.

Where a request leaves through a queue or a scheduled job, the code often has to carry the ID across. That’s part of the work.

Start with your last incident

We begin with your most recent incident and the alerts around it: what users felt, how your team found out, where the search went, and which alerts fired without needing anyone. Then we pick the one journey that matters most, and measure what users get from it. You get a written plan: that journey’s target and what happens when it’s missed, the alerts to keep, change and remove, and the first service instrumented with you, so the way is tried before the rest.

Your logs, metrics and traces stay in your accounts and tools, instrumented to an open standard so they can move, with personal data and secrets left out. Where our engineers reach that data, it’s for a task you’ve agreed, and it’s recorded.

Work with us as a dedicated team, on a fixed price project or a mix of both. Compare engagement models →

How each is paid for: Pricing →

The first question about an incident, in an example.

Your last incident

How did we find out?

  • A customer told us
  • An alert told us: The aim

Questions worth asking

Is observability just monitoring under a new name?

Not quite. Monitoring checks for things you already know can go wrong. Observability means your systems keep enough detail to answer a question nobody thought of ahead, such as why checkout is slow for some customers and not others. You need both.

What is SRE?

Site reliability engineering (SRE) is the practice of running systems as an engineering problem: targets agreed ahead, an allowance for failure, and alerts and reviews built around them. We do that work with your team, whatever your engineers’ job titles are.

Which tools do you use?

The ones you already have, where they do the job. We instrument your code to an open standard, so the data goes to the tools you choose and moves with you if you change them. We don’t sell or resell a monitoring product.

Will it make our monitoring bill bigger?

It can, if everything is kept in full detail forever. We decide with you what’s kept, in how much detail and for how long, by the questions it answers. For the costs themselves: Cloud Cost Optimization →

What should our target be?

That depends on what your users notice, and on what each step closer to perfect costs you. We don’t promise a number: no uptime, speed or recovery time. We measure what your users get through a full cycle of your business, then agree each target with your product team, short of perfect on purpose.

Who is on call?

Your team, ours, or both. Who answers a call, and in which hours, is agreed in writing before the first alert goes live. Where we look after your software, your support plan sets the first response time for each priority. Support & Maintenance →

Can you watch AI features too?

Yes. What users get from an AI feature is measured the same way, and what each use costs to run is set as a limit before launch. AI Product Engineering →

Is security monitoring part of this?

No. Watching for signs of attack is separate work, and so are dashboards for the business. This is about whether your software works for the people using it. For those dashboards: Dashboards & BI →

Do you run it for us, or hand it over?

Either. We can run it with you: watch your systems, respond when an alert needs a person, and go through the alerts and targets with you on a regular rhythm. Or we hand it over to your team, with every first step written down and the knowledge behind them. Cloud, DevOps & Security →

What does it cost?

It depends on how many services you run and how they’re watched today. The first call is free. Your proposal sets out the scope, the price or rate, and how you pay, before any work starts. Pricing →

OpenTelemetry is a registered trademark of The Linux Foundation in the United States and/or other countries.

Let’s run it better

Tell us how you found out about your last incident. We’ll suggest where to start.

Thanks, we’ll be in touch soon.

{{ ctaStatus }}