Skip to main content

SRE Revisited: SLO in the Age of Microservices

Dotan Horovits
Logz.io

Site Reliability Engineering (SRE) practice was established by Google nearly 20 years ago and was popularized with Google's monumental SRE Book. Everyone's been attempting to follow that iconic path ever since.

What Are SLI, SLO and SLA Anyway?

SRE practice revolves around the concepts of SLO, SLI, SLA, and the related "error budget." The Google SRE Book, which laid the foundations for the practice, defines them as follows:

Service Level Indicator (SLI) is a carefully defined quantitative measure of some aspect of the level of service that is provided. Common examples include latency, error rate, request throughput and availability.

Service Level Objective (SLO) is a target value or range of values for a service level that is measured by an SLI. For example, an SLO can state that the average latency per request should be under 120 milliseconds. Many companies set SLO targets as a number of nines, for example "five nines" uptime means 99.999% uptime, which means a maximum of 5.26 minutes downtime per year.

Service Level Agreement (SLA) is an explicit or implicit contract with your users that includes consequences of meeting (or missing) the SLOs they contain. Put simply, if you've got a penalty attached to breaching an SLO — you're talking SLA.

SRE typically doesn't deal with SLA directly, as it's more commercial in nature. Rather, the SRE is concerned with meeting the SLOs that underlie the SLA.

Error budget is an important related concept, which determines the rate at which the SLOs can be missed. Error budget is an important indicator for meeting the SLO goals consistently, and SRE teams track it weekly and even daily. Error budget is also important, as it enables downtime windows that can be used for maintenance, experimenting with new features, and other innovation.

So How Is SRE Changing in the Age of Microservices?

The SRE concepts remain valid these days as they've been before. But then again, their system of reference has drastically changed.

SLOs have traditionally referred to assets such as the application's binary or other deployment artifacts. That made perfect sense when the binary encapsulated the entire application, or a major piece of it such as the business logic tier.

In the world of microservices, however, the deployed artifacts are, well — microservices. And as the name suggests, each individual micro-service executes only a tiny piece of functionality.
In such systems it makes little sense to track SLO on the individual artifacts. SLO should refer to the functionality that is meaningful to the users.

Therefore, the practice of monitoring SLO on the service binary is no longer relevant.

In microservice architecture, rather than the individual service, it is the mesh of services that needs to play well together to execute. In these systems, tens and even hundreds of services may be traversed in order to satisfy a single user request. In these systems, rather than monitoring individual services, SLI and SLO should be defined for flows and user journeys.

The complex request flows brought about the rise of Distributed Tracing discipline. It enables tracing a single request through the call chain between the different interacting microservices, by propagating context along the request path. It's typically used to pinpoint where failures occur and what causes poor performance. However, tracing can also be used to trace FinOps and business workflows.

Google Dapper is in fact one of the forefathers of modern distributed tracing, which served as inspiration for tools such as Jaeger and Zipkin. According to Ramón Medrano Llamas, a staff site reliability engineer at Google who leads the SRE team in charge of Google's identity services, "it's the closest tool we have right now for a definite observability tool."

To make things more interesting, these user journeys not only span several services, but may actually involve a few requests. Let's look for an example at the Google sign-in process: as Ramón from Google shares, what used to be a matter of three calls in the pre-microservices era, is today a full business flow, consisting of multiple steps: loading page, then putting username & password, then second factor auth. and so on. Each individual step is a request, triggering a chain of microservices to satisfy. And the overall multi-step flow isn't stateful, it depends on the end user to carry it over from step to step. Monitoring these business flows is less trivial, but that's what we need to measure to ensure the functionality we deliver to the user.

How should you go about defining SLO then?

■ Start by identifying the main flows and user journeys in your system. These should be well defined as part of the system and product specifications.

■ Then for each flow, clarify the expected end result from the user perspective, to define the SLI. This should also be a direct derivative of the product flow definition.

■ If needed, add missing instrumentation to track, collect and aggregate the relevant metrics to measure the defined SLI.

■ Make sure that the relevant flow context is propagated throughout your system. This will serve to associate a metric with the respective tenant, account or similar entities.

■ Carefully define the SLO around each SLI, to meet technical and business goals. This often involves additional stakeholders, to factor in the tradeoff between stability and velocity of innovation.

■ Start monitoring the SLI and set relevant alerting according to the threshold defined by the SLO. Track your error budget regularly to see how close you get to the SLO bar.

Dotan Horovits is Principal Developer Advocate at Logz.io

Hot Topics

The Latest

As businesses increasingly rely on high-performance applications to deliver seamless user experiences, the demand for fast, reliable, and scalable data storage systems has never been greater. Redis — an open-source, in-memory data structure store — has emerged as a popular choice for use cases ranging from caching to real-time analytics. But with great performance comes the need for vigilant monitoring ...

Kubernetes was not initially designed with AI's vast resource variability in mind, and the rapid rise of AI has exposed Kubernetes limitations, particularly when it comes to cost and resource efficiency. Indeed, AI workloads differ from traditional applications in that they require a staggering amount and variety of compute resources, and their consumption is far less consistent than traditional workloads ... Considering the speed of AI innovation, teams cannot afford to be bogged down by these constant infrastructure concerns. A solution is needed ...

AI is the catalyst for significant investment in data teams as enterprises require higher-quality data to power their AI applications, according to the State of Analytics Engineering Report from dbt Labs ...

Misaligned architecture can lead to business consequences, with 93% of respondents reporting negative outcomes such as service disruptions, high operational costs and security challenges ...

A Gartner analyst recently suggested that GenAI tools could create 25% time savings for network operational teams. Where might these time savings come from? How are GenAI tools helping NetOps teams today, and what other tasks might they take on in the future as models continue improving? In general, these savings come from automating or streamlining manual NetOps tasks ...

IT and line-of-business teams are increasingly aligned in their efforts to close the data gap and drive greater collaboration to alleviate IT bottlenecks and offload growing demands on IT teams, according to The 2025 Automation Benchmark Report: Insights from IT Leaders on Enterprise Automation & the Future of AI-Driven Businesses from Jitterbit ...

A large majority (86%) of data management and AI decision makers cite protecting data privacy as a top concern, with 76% of respondents citing ROI on data privacy and AI initiatives across their organization, according to a new Harris Poll from Collibra ...

According to Gartner, Inc. the following six trends will shape the future of cloud over the next four years, ultimately resulting in new ways of working that are digital in nature and transformative in impact ...

2020 was the equivalent of a wedding with a top-shelf open bar. As businesses scrambled to adjust to remote work, digital transformation accelerated at breakneck speed. New software categories emerged overnight. Tech stacks ballooned with all sorts of SaaS apps solving ALL the problems — often with little oversight or long-term integration planning, and yes frequently a lot of duplicated functionality ... But now the music's faded. The lights are on. Everyone from the CIO to the CFO is checking the bill. Welcome to the Great SaaS Hangover ...

Regardless of OpenShift being a scalable and flexible software, it can be a pain to monitor since complete visibility into the underlying operations is not guaranteed ... To effectively monitor an OpenShift environment, IT administrators should focus on these five key elements and their associated metrics ...

SRE Revisited: SLO in the Age of Microservices

Dotan Horovits
Logz.io

Site Reliability Engineering (SRE) practice was established by Google nearly 20 years ago and was popularized with Google's monumental SRE Book. Everyone's been attempting to follow that iconic path ever since.

What Are SLI, SLO and SLA Anyway?

SRE practice revolves around the concepts of SLO, SLI, SLA, and the related "error budget." The Google SRE Book, which laid the foundations for the practice, defines them as follows:

Service Level Indicator (SLI) is a carefully defined quantitative measure of some aspect of the level of service that is provided. Common examples include latency, error rate, request throughput and availability.

Service Level Objective (SLO) is a target value or range of values for a service level that is measured by an SLI. For example, an SLO can state that the average latency per request should be under 120 milliseconds. Many companies set SLO targets as a number of nines, for example "five nines" uptime means 99.999% uptime, which means a maximum of 5.26 minutes downtime per year.

Service Level Agreement (SLA) is an explicit or implicit contract with your users that includes consequences of meeting (or missing) the SLOs they contain. Put simply, if you've got a penalty attached to breaching an SLO — you're talking SLA.

SRE typically doesn't deal with SLA directly, as it's more commercial in nature. Rather, the SRE is concerned with meeting the SLOs that underlie the SLA.

Error budget is an important related concept, which determines the rate at which the SLOs can be missed. Error budget is an important indicator for meeting the SLO goals consistently, and SRE teams track it weekly and even daily. Error budget is also important, as it enables downtime windows that can be used for maintenance, experimenting with new features, and other innovation.

So How Is SRE Changing in the Age of Microservices?

The SRE concepts remain valid these days as they've been before. But then again, their system of reference has drastically changed.

SLOs have traditionally referred to assets such as the application's binary or other deployment artifacts. That made perfect sense when the binary encapsulated the entire application, or a major piece of it such as the business logic tier.

In the world of microservices, however, the deployed artifacts are, well — microservices. And as the name suggests, each individual micro-service executes only a tiny piece of functionality.
In such systems it makes little sense to track SLO on the individual artifacts. SLO should refer to the functionality that is meaningful to the users.

Therefore, the practice of monitoring SLO on the service binary is no longer relevant.

In microservice architecture, rather than the individual service, it is the mesh of services that needs to play well together to execute. In these systems, tens and even hundreds of services may be traversed in order to satisfy a single user request. In these systems, rather than monitoring individual services, SLI and SLO should be defined for flows and user journeys.

The complex request flows brought about the rise of Distributed Tracing discipline. It enables tracing a single request through the call chain between the different interacting microservices, by propagating context along the request path. It's typically used to pinpoint where failures occur and what causes poor performance. However, tracing can also be used to trace FinOps and business workflows.

Google Dapper is in fact one of the forefathers of modern distributed tracing, which served as inspiration for tools such as Jaeger and Zipkin. According to Ramón Medrano Llamas, a staff site reliability engineer at Google who leads the SRE team in charge of Google's identity services, "it's the closest tool we have right now for a definite observability tool."

To make things more interesting, these user journeys not only span several services, but may actually involve a few requests. Let's look for an example at the Google sign-in process: as Ramón from Google shares, what used to be a matter of three calls in the pre-microservices era, is today a full business flow, consisting of multiple steps: loading page, then putting username & password, then second factor auth. and so on. Each individual step is a request, triggering a chain of microservices to satisfy. And the overall multi-step flow isn't stateful, it depends on the end user to carry it over from step to step. Monitoring these business flows is less trivial, but that's what we need to measure to ensure the functionality we deliver to the user.

How should you go about defining SLO then?

■ Start by identifying the main flows and user journeys in your system. These should be well defined as part of the system and product specifications.

■ Then for each flow, clarify the expected end result from the user perspective, to define the SLI. This should also be a direct derivative of the product flow definition.

■ If needed, add missing instrumentation to track, collect and aggregate the relevant metrics to measure the defined SLI.

■ Make sure that the relevant flow context is propagated throughout your system. This will serve to associate a metric with the respective tenant, account or similar entities.

■ Carefully define the SLO around each SLI, to meet technical and business goals. This often involves additional stakeholders, to factor in the tradeoff between stability and velocity of innovation.

■ Start monitoring the SLI and set relevant alerting according to the threshold defined by the SLO. Track your error budget regularly to see how close you get to the SLO bar.

Dotan Horovits is Principal Developer Advocate at Logz.io

Hot Topics

The Latest

As businesses increasingly rely on high-performance applications to deliver seamless user experiences, the demand for fast, reliable, and scalable data storage systems has never been greater. Redis — an open-source, in-memory data structure store — has emerged as a popular choice for use cases ranging from caching to real-time analytics. But with great performance comes the need for vigilant monitoring ...

Kubernetes was not initially designed with AI's vast resource variability in mind, and the rapid rise of AI has exposed Kubernetes limitations, particularly when it comes to cost and resource efficiency. Indeed, AI workloads differ from traditional applications in that they require a staggering amount and variety of compute resources, and their consumption is far less consistent than traditional workloads ... Considering the speed of AI innovation, teams cannot afford to be bogged down by these constant infrastructure concerns. A solution is needed ...

AI is the catalyst for significant investment in data teams as enterprises require higher-quality data to power their AI applications, according to the State of Analytics Engineering Report from dbt Labs ...

Misaligned architecture can lead to business consequences, with 93% of respondents reporting negative outcomes such as service disruptions, high operational costs and security challenges ...

A Gartner analyst recently suggested that GenAI tools could create 25% time savings for network operational teams. Where might these time savings come from? How are GenAI tools helping NetOps teams today, and what other tasks might they take on in the future as models continue improving? In general, these savings come from automating or streamlining manual NetOps tasks ...

IT and line-of-business teams are increasingly aligned in their efforts to close the data gap and drive greater collaboration to alleviate IT bottlenecks and offload growing demands on IT teams, according to The 2025 Automation Benchmark Report: Insights from IT Leaders on Enterprise Automation & the Future of AI-Driven Businesses from Jitterbit ...

A large majority (86%) of data management and AI decision makers cite protecting data privacy as a top concern, with 76% of respondents citing ROI on data privacy and AI initiatives across their organization, according to a new Harris Poll from Collibra ...

According to Gartner, Inc. the following six trends will shape the future of cloud over the next four years, ultimately resulting in new ways of working that are digital in nature and transformative in impact ...

2020 was the equivalent of a wedding with a top-shelf open bar. As businesses scrambled to adjust to remote work, digital transformation accelerated at breakneck speed. New software categories emerged overnight. Tech stacks ballooned with all sorts of SaaS apps solving ALL the problems — often with little oversight or long-term integration planning, and yes frequently a lot of duplicated functionality ... But now the music's faded. The lights are on. Everyone from the CIO to the CFO is checking the bill. Welcome to the Great SaaS Hangover ...

Regardless of OpenShift being a scalable and flexible software, it can be a pain to monitor since complete visibility into the underlying operations is not guaranteed ... To effectively monitor an OpenShift environment, IT administrators should focus on these five key elements and their associated metrics ...