Plan update waves, soak times and a halt rule
Choose how much of the fleet each wave reaches and how long to wait between waves. The planner shows devices per wave, the day each wave starts and when to stop.
Key characteristics
- Inputs
- Fleet size, three wave shares, soak time, failure and halt rates
- Outputs
- Devices and start day per wave, expected rollbacks, halt count
- Default example
- A fictional fleet of 12,000 devices in four waves
- Runs
- In your browser with nothing sent
Calculator: Plan update waves, soak times and a halt rule
A staged rollout sends a new firmware release to a small part of the fleet first, waits, checks what happened and only then widens. Google’s site reliability engineers call that first step a canary: a partial, time-limited deployment that is evaluated before the rollout continues. Devices raise the stakes: a bad server release is rolled back in minutes, while a bad device release may need someone to drive out to a field. This planner turns your wave sizes and soak time into a schedule, and your halt threshold into a device count you can write down before the first wave goes out.
How to use it
- Enter the number of devices that will be offered the release.
- Set waves 1 to 3 as cumulative shares of the fleet. Wave 4 always offers the update to every remaining device.
- Set the soak time: how long you watch one wave before starting the next.
- Enter the failure rate you expect and the rate at which you would halt.
What each input means
| Input | What it means | Where to get a real number |
|---|---|---|
| Devices in the fleet | Devices that will be offered this release | Your device registry, minus units in storage or returned |
| Wave shares | How much of the fleet has been offered the update by the end of each wave, counting earlier waves | Start small, with devices you can reach easily, ideally your own |
| Soak time | Hours between the start of one wave and the start of the next | Long enough for the wave to check in, update and run through at least one full daily cycle |
| Expected update failures | Share of devices you expect to roll back | Rollback counts from earlier releases and from your test rig |
| Halt threshold | Failure rate at which you stop and investigate | Agreed before the release, as a row in your update policy worksheet |
Reading the default example
Example (fictional). A fleet of 12,000 devices gets a release in waves that reach 1%, 10% and 50% of the fleet, then everyone, with 48 hours of soak between waves and a 0.2% expected failure rate. Wave 1 is 120 devices and starts on day 0, with about 0.2 expected rollbacks. Wave 2 adds 1,080 devices on day 2 (about 2.2 rollbacks), wave 3 adds 4,800 on day 4 (about 9.6) and wave 4 the last 6,000 on day 6 (about 12.0). The whole fleet has been offered the update by day 6. With a 1% halt threshold, more than 1 rollback among the 120 devices of wave 1 stops the rollout. About 24 rollbacks are expected across the fleet.
Read the halt line closely. With a 1% threshold and 120 devices, one rollback (0.8%) is within plan and two (1.7%) are not. A small first wave is also a weak detector: if the true failure rate is 0.2%, the chance that wave 1 shows even one rollback is only about 21%. A clean first wave is not proof of a good release, which is why the soak time and the other signals you watch (reconnects, crash reports, battery readings) matter as much as the rollback count.
Loamwick’s own trial is far smaller: 40 probes in three waves of 4, 12 and 24, with 48 hours between them, halting if any probe rolls back or goes silent. At that size a rule about single devices works better than a percentage.
How it calculates
Each wave’s cumulative size is the fleet multiplied by its share, rounded up. The devices in a wave are that number minus the previous wave’s cumulative size. Each wave starts one soak period after the previous one, so wave n starts on day (n minus 1) times the soak time in days. Expected rollbacks are the devices in a wave multiplied by the expected failure rate. The halt count for wave 1 is its size multiplied by the halt threshold, rounded down; any count above it halts the rollout.
Limits
- Always four waves. The planner uses three shares you choose plus a final wave to everyone. A small fleet like Loamwick’s may want fewer waves; plan those by hand.
- Offered is not updated. Battery devices that check in once a day, or devices out of coverage, stretch every wave. Base the soak time on when most of a wave has actually updated.
- Failures are rarely random. A release that breaks one hardware revision or one gateway model can fail all of those devices and none of the rest. Make sure wave 1 includes every hardware revision and connection path in the fleet.
- Rollback only catches what the device can detect. MCUboot boots a new image as a test and reverts to the previous one if the application never confirms it. A release that confirms itself and then misbehaves, for example by draining the battery, stays installed; only your telemetry will show it.
What to do next
Write the plan down with owners in the update policy worksheet, and check each release against the firmware release checklist. If your devices use cellular, the over-the-air data cost calculator shows what a rollout costs. Our over-the-air updates service sets up the waves, the soak signals and the halt rule with you, and the updates and releases section of Field Update has more on release engineering.
Send us your board and your plan.
A firmware review reads your schematics, code, flash layout and update plan, and ends with a written plan: what to build first, what could go wrong in the field and how updates will reach your devices.
Or email sales@fumbar.com, or call (720) 675-4202.