Fique Em Casa Conf
Better reliability through SLOs
· Virtual
Service Level Objectives (SLOs) are a measurement of the reliability and general experience your end users and customers can expect. In this talk, you’ll learn how to define SLOs by choosing the correct service level indicators (SLIs) and defining appropriate agreements with stakeholders. We’ll explain the key concept of error budgets, which give you a solid, actionable metric for balancing innovation and velocity with reliability and safety. You’ll also learn how to have meaningful conversations around realistic availability, which will enable you to define high quality SLOs for your own organization.
Better Reliability With SLOs Daniel Maher Developer Relations
Better Reliability With SLOs Daniel “phrawzty” Maher Developer Relations
What we’ll cover today – SLIs, SLOs, and SLAs – Defining quality targets – Error budgets – Practical examples @phrawzty
SLIs, SLOs, & SLAs @phrawzty
A Service Level Indicator is a quantitative measurement that expresses an aspect of the service (commonly a metric). SLIs @phrawzty Image credit: Indicator by Nhor from the Noun Project
A Service Level Objective is a target value for a service, as measured via an SLI. SLOs @phrawzty Image credit: Objective by priyanka from the Noun Project
A Service Level Agreement is a contract that defines the results (and consequences) of meeting (or missing) one or more SLOs. SLAs @phrawzty Image credit: Checked document by Rockicon from the Noun Project
Multiple Stakeholders Product Managers Developers, SREs Executives Customers @phrawzty
Focus on user experience SLIs @phrawzty SLOs SLAs
Defining quality targets
User experience is everything ● How are they interacting with your product? @phrawzty
User experience is everything ● How are they interacting with your product? ● What is their workflow? @phrawzty
User experience is everything ● How are they interacting with your product? ● What is their workflow? ● What services do they interact with? @phrawzty
User experience is everything ● How are they interacting with your product? ● What is their workflow? ● What services do they interact with? ● What do they want? What do they expect? @phrawzty
Not all values make good SLIs @phrawzty – Free resources (CPU, Memory, Disk Space) – Quorum state (does the leader matter?) – Number of lines of code per commit.
Identifying good SLIs Response/Request Availability – Could the server respond to the request? Latency – How long did it take for the server to respond to the request? Throughput – How many requests can be handled? Storage Availability – Can the data be accessed on demand? Latency – How long does it take to read or write data? Durability – Is the data still there when it is needed? Pipeline Correctness – Was the right data returned? Freshness – How long does it take for new data or processed results to appear?
SLIs are applied values Indicators must have represent user experience. The number of requests to an endpoint that complete successfully. @phrawzty The number of requests to an endpoint that complete within 500ms.
SLOs are applied SLIs Objectives have both a target and a time window. Requests are 99.95% successful in the last 24 hours. @phrawzty 90% of requests complete under 500ms in the 30 days.
SLAs are applied SLOs Agreements address expectations and impacts. The customer expects a given service to have a 0.05% maximum error rate daily, or they’ll receive a rebate. @phrawzty The customer expects only 10% of monthly requests to take longer than 500ms to complete, or they’ll be reimbursed for the compute overage.
Error Budgets
Move fast and fix things! @phrawzty – Failure is unavoidable; how you respond is important. – Balance innovation and novelty with reliability and stability. – Similar to an SLA.
Building an Error Budget ● An SLO is identified by the product owner. @phrawzty
Building an Error Budget ● An SLO is identified by the product owner. ● The actual objective is measured by a neutral party (hint: a monitoring system). @phrawzty
Building an Error Budget ● An SLO is identified by the product owner. ● The actual objective is measured by a neutral party (hint: a monitoring system). ● The difference between the actual measurement and the objective is the error budget. @phrawzty
Using the budget Spend the budget Build the budget If the SLO is currently being met, you have room to move. Add new features! Deploy a new version! Trigger some planned downtime… If the budget is zero (or negative), you should concentrate on that. Freeze new features. Improve the observability story. Prioritise dealing with technical debt. @phrawzty
Practical examples
fotosite.neð (not real) – A fun site for uploading, sharing, and viewing photos. – Make friends! Build communities! – Buy print-quality photos. @phrawzty
Consider the user experience ● How are they interacting with your product? ○ Clicking, viewing photos, doing searches. @phrawzty
Consider the user experience ● How are they interacting with your product? ○ Clicking, viewing photos, doing searches. ● What is their workflow? ○ Log in, search photos, view, download. @phrawzty
Consider the user experience ● How are they interacting with your product? ○ Clicking, viewing photos, doing searches. ● What is their workflow? ○ Log in, search photos, view, download. ● What services do they interact with? ○ Accounts backend, upload service @phrawzty
Consider the user experience ● How are they interacting with your product? ○ Clicking, viewing photos, doing searches. ● What is their workflow? ○ Log in, search photos, view, download. ● What services do they interact with? ○ Accounts backend, upload service ● What do they want? What do they expect? ○ They want it to be fast!
fotosite.neð: Indicators Response/Request Availability – Could the server respond to the request? Latency – How long did it take for the server to respond to the request? Throughput – How many requests can be handled? Storage Availability – Can the data be accessed on demand? Latency – How long does it take to read or write data? Durability – Is the data still there when it is needed? Pipeline Correctness – Was the right data returned? Freshness – How long does it take for new data or processed results to appear?
fotosite.neð: Indicators Response/Request Availability – Could the server respond to the request? Latency – How long did it take for the server to respond to the request? Throughput – How many requests can be handled? Storage Availability – Can the data be accessed on demand? Latency – How long does it take to read or write data? Durability – Is the data still there when it is needed? Pipeline Correctness – Was the right data returned? Freshness – How long does it take for new data or processed results to appear?
fotosite.neð: Indicators Response/Request Availability – Could the server respond to the request? Latency – How long did it take for the server to respond to the request? Throughput – How many requests can be handled? Storage Availability – Can the data be accessed on demand? Latency – How long does it take to read or write data? Durability – Is the data still there when it is needed? Pipeline Correctness – Was the right data returned? Freshness – How long does it take for new data or processed results to appear?
Bonus: More tips!
Measuring SLIs with Datadog Monitor-based Event-based Based on monitors, which are generally tied to metrics. - Values within a set time frame. - ex. “99% of the time, latency for this request is less than 200ms” Based on events, which are more akin to statements. - Effectively a success ratio. - ex. “99% of requests have latency less than 200ms” @phrawzty
Four Golden Signals Latency Traffic Errors Saturation @phrawzty
– Step back from the internals. Focus on your users. @phrawzty – Define user stories / journeys first, then SLIs. – Involve all stakeholders, especially product.
– Gain experience; experiment by hand if that’s easier! Start small. – Build out data sets to establish reasonable baselines. – Error on the side of naïveté (at first). @phrawzty
SLOs change. @phrawzty – Re-evaluate your SLIs and SLOs as your environment evolves. – SLAs must have the capacity to evolve, too!
Tooling matters. @phrawzty
Obrigado!
Recording
Buzz and feedback
Já estamos ao vivo com o sensacional @phrawzty ! Bora lá acompanhar esse talk que será foda! #FiqueEmCasaConf #VAIIII https://t.co/q1aKYfVYKS https://t.co/XSjV4kWEnq — Jeferson Fernando (@badtux_) April 17, 2020
@badtux_