ZipLoom
Home
PricingSign InGet Your Live URL

operate

Staged / Canary Rollout

1% → 10% → 50% → 100% With An Automatic Error Gate — On Google Cloud Run.

Updated

At a glance

ZipLoom moves Cloud Run traffic to a new revision in steps (1%, 10%, 50%, 100% by default) and reverts if its errors exceed your threshold and the old one's.

  • Automatic Error Gate
  • Configurable Stages
  • You Stay In Control
Sample. On Google Cloud Run, a new revision takes traffic in stages. At each stage ZipLoom compares its error rate with the previous version; over your threshold and worse, confirmed twice, all traffic goes back.

Why it matters

A full cutover means every user meets a bad release at the same moment. The difference between a minor incident and an outage is usually how many people were exposed before anyone noticed.

On Google Cloud Run, ZipLoom deploys the new revision with no traffic, health-checks it, then moves traffic in stages: 1%, 10%, 50% and 100% by default, or your own stages and timings. Every 5 minutes it sends its own requests to the new and the previous version and compares their error rates. If the new one is over your threshold and worse, confirmed by a second check, all traffic goes back and you get an email. If both are failing, it holds, because the release is not the clear cause. Owners and Admins can pause, resume, skip a stage or abort, and each action is audited.

A Release That Earns Its TrafficIllustration · sample data
New revision0% traffic1%10%50%100%Previouscomparederror ratesworse:all backpause · resumeskip · abort

What you get

Automatic Error Gate

At each stage, ZipLoom sends its own requests to the new version and to the previous version and compares their error rates. If the new one is over your threshold and worse than the previous one, confirmed by a second check, all traffic goes back to the previous version and you get an email. If both are failing, nothing moves, because the release is not the clear cause. ZipLoom measures this itself; it does not read your real visitors' traffic.

Configurable Stages

Default stages: 1% → 10% → 50% → 100%. Per project, you set the percentages (up to 8 stages, ending at 100), the minutes at each stage (5 to 60), and the error rate threshold. Stages are checked every 5 minutes, so a stage lasts at least its minutes and can run a few minutes longer. Some releases warrant a longer soak at 10%; some can go straight to 50%.

You Stay In Control

An Owner or Admin can pause a rollout, resume it, skip the rest of a stage, or abort it and send all traffic back to the previous version. Every one of those actions is recorded in the audit log. A paused rollout is still watched, so a real breach still reverts it.

How Staged Rollout Works

  1. 1

    Connect Google Cloud Run as the project's deploy target. Staged rollout is for Cloud Run only; projects on any other host deploy normally. It is included on the Business plan and above.

  2. 2

    Enable staged rollout for the project on the Staged Rollout page, and set your stages, minutes per stage, and error threshold.

  3. 3

    Deploy. ZipLoom deploys the new Cloud Run revision with no traffic, health-checks it on its own address, then sends the first stage's share to it. The first deploy of a service, and a service with no earlier good deploy, deploys normally.

  4. 4

    Every 5 minutes ZipLoom checks the new and the previous version. When a stage's time is up and the new version is healthy, it moves to the next stage; at the last stage the new version takes 100%.

  5. 5

    If the new version is over the threshold and worse than the previous one, confirmed by a second check, all traffic returns to the previous version.

  6. 6

    You receive an email when a rollout finishes or reverts. You can pause, resume, skip a stage, or abort from the Staged Rollout page at any time.

What this check inspects

  • Whether the new revision responds before any traffic is sent to it.

  • The new version's error rate and the previous version's, measured by ZipLoom's own requests to each.

  • The comparison between the two, so a regression is judged against how the previous release is doing right now.

  • Whether the rollout should advance, hold, or revert, decided on those measurements and your threshold, not on a timer alone.

What it means when this fails

A full-cutover deploy means every user meets your bug at the same moment. A staged rollout means a fraction do, briefly, and the release stops there. The difference between a minor incident and an outage is usually just how many people were exposed before someone noticed.

What it does not do

  • It works on Google Cloud Run only. Projects on other hosts deploy normally.

  • It measures its own test requests, not your real visitors' traffic.

  • It is included on the Business plan and above.

Questions

Does this slow down shipping?
Only in wall-clock minutes, and only on the way out. You're trading a few minutes of gradual rollout for not having to undo a bad release under pressure.
What triggers a revert?
The new version's error rate going over the threshold you set and being worse than the previous version's, confirmed by a second check. The threshold is a setting you can see and change. If both versions are failing, ZipLoom holds instead of reverting.
Which hosts does it work on?
Google Cloud Run. Other hosts deploy normally, without staged rollout.
Can I skip it for a small fix?
Yes. It is a per-project setting: turn it off, or keep it on and use Skip Stage to move a running rollout along. It is most valuable on releases that touch data, auth, or payments.

What it replaces

Better Stack Uptime

Uptime monitoring, on-call and status pages. Tells you it broke; doesn't roll the release back.

$29/mo

Checkly Starter

Synthetic API and Playwright browser checks. Team plan $64/mo for all regions.

$24/mo

ZipLoom Solo — every deploy check included

Deploy, security scan, and monitoring in one flat price. Guild $40/mo · Studio $70/mo · LoomWorks $149/mo · Business $299/mo.

$216/yr($20/mo)

Related features