Engineering

Release cadence: how often should your team ship?

7 min read

A release cadence is the regular rhythm on which a team puts new versions in front of users: as each change is ready, every week, every four weeks, a few times a year. The right one is the fastest rhythm your testing, your rollback path and your communication can keep up with, and it is a separate decision from how often you deploy. Large projects publish their schedules, which makes them useful reference points when you pick your own.

ProjectCadenceSource
ChromeA new milestone every 4 weeks since Chrome 94 (2021), with an 8-week Extended Stable optionChromium blog
FirefoxEvery 4 weeks, phased in from Firefox 71, previously every 6 to 8 weeksMozilla
KubernetesAbout three minor releases a year, each cycle around 15 weeksKEP-2572
PythonOne feature version a year, in OctoberPEP 602
Node.jsTwo majors a year through Node.js 26, with even-numbered ones becoming LTS; from Node.js 27, one major a year, each becoming LTSNode.js releases
UbuntuAn interim release every six months, an LTS release every two yearsUbuntu release cycle

What is a release cadence?

It is the schedule, written or habitual, that decides when users get new versions. Some teams set a date and ship whatever is ready on it; others release each change as soon as it is finished. Both are cadences, and both work when everyone knows which one is in force.

Cadence is easy to confuse with deployment frequency, which DORA’s software delivery metrics define as “the number of deployments over a given period or the time between deployments.” A team can deploy twenty times a day and still release features to users once a month, with flags keeping the unfinished ones hidden. The difference is covered in release vs deployment; this article is about the second number, the one users notice.

Which release rhythms are most common?

Most products fall into one of five rhythms, and many run two at once for different audiences.

  • Continuous. Each change is released when it is ready, often behind a flag that is switched on later. Common for web applications, where users never install anything.
  • A fixed train every week or two. The release leaves on its date with whatever is ready, and anything that misses waits for the next one.
  • Every four to six weeks. The browser rhythm: long enough for a beta channel to catch problems, short enough that no feature waits long.
  • A few times a year. Platforms that other teams build on, like Kubernetes, where every release means upgrade work for someone downstream.
  • Once a year. Programming languages and runtimes, where each version has to be supported for years after it ships.

The two-track pattern shows up across the table. Chrome pairs its 4-week milestones with Extended Stable, Ubuntu pairs six-monthly interim releases with an LTS every two years, and Node.js marks its majors for long-term support. Rather than slowing everyone down for the customers who need stability, these projects give those customers a slower track. Ubuntu even puts the month in its version numbers, one of the schemes compared in software versioning.

How did Chrome, Firefox and Kubernetes choose their cadence?

Each one changed its cadence for a stated reason, and the reasons point in two directions: faster once testing could support it, slower when people downstream needed predictability.

Chrome had shipped “a new milestone every 6 weeks” for more than a decade. In its 2021 announcement the team wrote that as it “improved our testing and release processes,” it “became clear that we could shorten our release cycle and deliver new features more quickly.” Mozilla wanted to take features to market sooner when it moved Firefox from 6 to 8 weeks down to 4, and told projects built on Firefox that the new releases “will have fewer changes each so they should be correspondingly easier to integrate.”

Kubernetes went the other way. KEP-2572 proposed moving “from 4 down to 3 releases per year,” formalizing a schedule that had started in 2020 during the pandemic, because contributors and end users wanted to plan ahead and expected “a predictable release cadence.” Python’s PEP 602 moved from an 18-month cycle to an annual one and listed both kinds of benefit: it “makes releases smaller,” “puts features and bug fixes in hands of users sooner,” and creates a predictable calendar where the final release always lands in October.

Should you release on a schedule or when features are ready?

On a schedule, for most teams that ship installed software or ship to other developers, and per change for most web products. The schedule matters less than the rule for what happens when a feature is late.

A time-based release ships on its date with whatever is finished. Nobody argues about whether one more feature can squeeze in, because the next release is never far away, and the date is something support, documentation and customers can plan around. A feature-based release waits until a set of features is done, which sounds tidier and in practice tends to slip, since each delay pushes every other finished change back with it.

Feature flags take most of the cost out of the time-based model. Unfinished work merges and deploys switched off, so the train can leave without it and nothing has to be pulled from a branch at the last minute. Frameworks that plan on a fixed rhythm, such as an agile release train, separate the planning cadence from the release decision, so releases can go out whenever the business is ready.

How do you choose a release cadence for your team?

Start from who has to act when a release goes out, then check what your pipeline can support. A few questions settle most cases:

  1. Does anyone have to upgrade? If users of a web application get changes without doing anything, frequent releases cost them little. If customers install, integrate or migrate, every release is work for them, which is why platforms and languages release less often.
  2. How long does testing take? The cadence cannot be shorter than the time it takes to be confident in a build. Chrome shortened its cycle after its testing improved.
  3. How quickly can you roll back? A fast, routine rollback makes a short cadence safe. A rollback that needs a meeting argues for fewer, better-tested releases.
  4. Who has to prepare? Support, documentation and sales need to know what is coming. A cadence they cannot keep up with produces releases nobody can explain.
  5. Do some customers need a slower track? Offer one, the way Extended Stable and LTS releases do, instead of setting the whole cadence by your most cautious customer.

Then write it down: the cadence, the cut-off for a release, and who decides when something misses it. The release management process article covers the steps and owners around each release.

How does release cadence affect release notes?

A faster cadence means each release has fewer changes to describe, and it also means you should stop sending an announcement for every one. Record every release in the changelog, and save the release notes and emails for releases with something users will notice.

Notes that arrive on a strict schedule regardless of content teach readers to skip them; how to write release notes covers when to publish and when to roll entries into the next set. The changelog itself can keep pace with any cadence if entries are written when changes merge rather than when the release goes out. In Changeloop, each user-facing merged pull request becomes a draft entry, so on release day the work is reviewing and approving drafts, not writing from memory; the docs describe the flow. For releases big enough to announce, the release notes template gives the notes a consistent shape.

How do you know whether your cadence is working?

Watch whether releases go out on time and whether they go out cleanly. Missed dates mean the cadence is too fast for your process; growing batches and nervous release days mean it is probably too slow.

DORA’s change fail rate, “the ratio of deployments that require immediate intervention following a deployment,” shows whether the pace is hurting quality. Two numbers of your own complete the picture: how often a finished feature misses its planned release, and the time from a change merging to users getting it.

FAQ

What is the difference between release cadence and deployment frequency? Deployment frequency counts how often code reaches production. Release cadence is how often users get new versions or features. They match only when every deploy releases its changes immediately, with no flags or staged rollouts in between.

Is releasing more often always better? No. Faster releases are smaller and easier to fix, but each one costs something for people who have to upgrade or integrate. That is why Chrome and Firefox ship every four weeks while Kubernetes ships about three times a year and Python once.

How should you announce a new release schedule? Months in advance, with dates. Chrome announced its 4-week cycle in March 2021 for a change starting in the third quarter of that year, and Mozilla published Firefox’s new schedule with a table of planned release dates.


The technical claims in this article have not been independently reviewed. If something here is wrong, tell us and we will correct it.

Related on Changeloop: Developer docs, Release notes template

changeloop
The team building a closed-loop changelog. Your users ask, your team ships, the person who asked gets told.