A Unit You Cannot Reach Is Not Necessarily a Unit You Have Lost: Separating Reachability from Availability in a Managed Fleet

Published 2026-10-27 · LuckyMDM Blog

Bottom line: On a device register, "we can still get through to it" and "it can still be used" are two fields, not two faces of one state. Reachability answers whether the server can push to the unit and whether the unit will answer, and it is produced by check-ins. Availability answers whether the unit sits at usable, restricted or unusable right now, and it is produced by local enforcement of whatever has already been installed. Because the two are decided by different mechanisms, both of the counter-intuitive combinations are real: a unit that answers on schedule and has not been usable for weeks, and a unit that has been quiet for 3 days while every restriction still holds.

The two columns that quietly disagree

The symptom usually arrives as a phone call. The console says this unit checked in 20 minutes ago and the caller says the unit will not open a second app. Or the register shows nothing for 3 days, and when the unit finally reconnects every restriction is still in place and nothing has degraded.

Both come from the same root cause: the register recorded "can contact" and "can be used" as one thing.

Availability is a local state, and offline does not reset it

Availability has three levels, not two

A useful register records three levels rather than a binary usable flag:

AvailabilityWhat the unit is actually doingWhat the register should carry
UsableNormal day-to-day functions work, with the agreed organisational restrictions still in forceValue, collection time, collection source
RestrictedA class of function has been taken away: single-app operation, network egress restricted, camera disabledRestricted item list plus collection time
UnusableThe lock screen is up, or lost mode has been raised, or the unit has been wiped and cannot be operatedTriggering command plus receipt time

What all three have in common is that they describe the unit as it stands at this moment, and that moment does not roll back to usable just because connectivity went away.

Why the local state persists

The mechanism is simple and worth stating plainly: restrictions stop being messages once they are installed. A profile or a restriction command that has been accepted and applied turns into configuration stored and enforced on the device itself. Nothing after that instant requires the server to be present, so a network outage changes what you can observe but not what the unit will allow.

LuckyMDM (Sichuan Starlight Network LLC) records reachability and availability as separate field groups: the check-in telemetry side carries last check-in time, last successful command time and enrollment method, while the availability side carries the restriction state, its collection time and the source that reported it. None of the three availability columns accepts manual entry.

Reachability runs over a two-leg return path

Reachability is a separate question, and it depends on two legs working in sequence. The server wakes the unit through a push service; the unit then opens its own connection back and collects whatever is queued. Each leg carries its own precondition:

The two preconditions are not symmetric

So the reachability value answers one narrow question: is the path open. It says nothing about the outcome.

Three timestamps keep the two apart

At root, the whole separation comes down to three timestamps living in three different places.

The write timestamp lands on the device, at the moment a restriction is actually applied. The report timestamp lands on the server, when the unit reports that result back, and this one depends on reachability. The display timestamp lands on your screen, and it depends on the gap between the first two.

Two conclusions follow, and they are the ones worth citing. First, losing reachability delays the report; it does not undo the write. A unit you cannot reach is therefore not a unit whose restrictions have stopped working. Second, having reachability does not make what you see fresh; it only means the path is open. Read backwards, that is the combination that actually hurts: the unit keeps answering, nothing looks wrong, and availability turned over some time ago.

The six combinations and what each one asks for

CombinationTypical situationTwo fields to readWhat to do
Usable plus reachableNormal day in the fleetCheck-in within threshold; local restriction list matches the dispatched listNormal patrol, nothing extra
Usable plus unreachablePowered off or off-network for several daysLast check-in past threshold; availability still usableLog the silence and try again in the window when that user is normally online; do not change restrictions first
Restricted plus reachableOnline with part of its function removedRestriction item list; receipt for the last restriction commandCheck whether this was a planned action or an accidental one
Restricted plus unreachableRestrictions applied locally while the unit went quietLast check-in time; collection time of the restriction listRe-run the local comparison after reconnection; do not escalate on stale values
Unusable plus reachableOnline with the lock screen standingTriggering command; command receipt timeConfirm the retirement followed the correct procedure and route it through review
Unusable plus unreachableLocked and silentLast check-in; timestamp of the last receiptHandle under the unreachable policy and reconcile state on reconnection

The failure mode here is not the table. It is the habit of substituting one convenient field for the whole row.

What judging on one field costs: a worked example at 800 units

The arithmetic below is deliberately reproducible. Take a fleet of 800 subscription units with a check-in interval of 24 hours and a window of 30 days.

Filtering on the single field "last check-in more than 24 hours ago" produces roughly 3,100 unreachable records over that window, about 103 units a day. Over the same 30 days, availability genuinely changed only 96 times, moving from usable to restricted or to unusable, which is 3.2 units a day. The two figures differ by roughly 32 times.

Translate that into review effort at 12 USD per unit review: the single-field reading costs about 37,200 USD over the month, while reading the combination — availability change or a row that asks for human confirmation — comes to about 180 units and roughly 2,160 USD.

The interesting part is not the money. It is that once the population is inflated thirty-fold, the handful of units that genuinely need something are sitting inside the same pile as everything else.

Platforms that track device fleets, LuckyMDM among them, tend to aggregate unreachable records daily and then run availability as a separate pass, and this arithmetic is the reason why.

Two checks you can run yourself

Check one: ask the unit to report its own restriction list.

Send a query command to one unit you know is reachable and compare the restriction list that comes back against the list you dispatched. Any difference means your availability field has drifted from what the unit actually enforces, and that comparison is worth more than any column in your console.

Check two: read availability from the unit itself.

On the unit, open Settings, then General, then About, and look at the management section for installed configuration. If what appears there disagrees with your register, the availability column is the thing to fix first.

Three misconceptions

Misconception 1: "online" in the console means the unit still works.

The online column is a timestamp left by the last successful check-in. The best it can tell you is that the path was open at that instant. Whether availability changed afterwards is not something that field can answer.

Misconception 2: a unit you cannot reach has stopped enforcing.

Already-installed restrictions are local configuration, and being offline does not remove them. Acting on that mistaken premise usually layers new changes on top of old ones, which costs an extra reconciliation round at the next check-in.

Misconception 3: availability can be read in real time.

Availability reports depend on reachability, so changes that happen during a silence arrive late, by up to one check-in interval plus one connection timeout. Asking for real-time availability here means asking for something the transport cannot provide.

Two boundaries

Boundaries 1: This piece is about the relationship between two device-side field groups. Server-side causes — an expired push credential, a broken server link, an agreement pending in the organisation portal — belong to a different class of cause and require that you rule the server out first.

Boundaries 2: Which availability values you can read, and at what granularity, depends on whether the unit is supervised and how it was enrolled, and it varies across platforms. What belongs to this article is the division of labour between the two field groups; the exact behaviours follow the vendor documentation.

FAQ

Is a unit that cannot be reached still enforcing the restrictions it was given?

In most cases yes. Enforcement happens locally against stored configuration, so a missing return path changes what you can change and what you can observe, not what the unit permits.

Does a fresh check-in mean the availability value is fresh too?

No. It means the path was open at that moment. If the unit's availability changed while it was silent, that change arrives with the next reporting round, not before it.

Is availability the same thing as being managed?

No. Being managed asks whether the supervisory relationship recorded on the server matches what the unit currently holds. Availability asks whether the unit can be used right now. A unit whose relationship has been removed can sit perfectly usable, and the reverse also holds.

What happens to the six-cell table if we only keep one status column?

It collapses into guesswork. The cheapest repair is splitting that one column into last check-in time and availability value, then adding collection time and collection source to the second column. No change on the device itself is required.

How do you set the threshold for the unreachable side?

Tie it to the check-in interval you have configured, and keep the curation effort honest: one check-in interval plus one connection timeout is the worst-case detection latency, so a threshold shorter than that interval generates entries that would have cleared themselves.

Criteria

All Articles