A cloud multi tenant management test plan proves site isolation, scoped roles, per-tenant records and degraded mode with named probes.

Δημοσιεύτηκε στις: Μαϊ 14, 2024

Cloud Multi Tenant Management: The Test Plan a Buyer Should Demand

Cloud Multi Tenant Management: The Test Plan a Buyer Should Demand

An integration failure is usually a sequence failure, so cloud multi tenant management test plan is traced here from handover to commissioning. A multi-tenant console is easy to demonstrate and hard to accept. The demonstration is always the same: a property selector, two buildings whose room numbers happen to be identical, a list of doors that looks right, and a promise that the permission model keeps the properties apart. The buyer's next question is the only one that matters — show me that the first property cannot read the second — and the vendor shows the menu again. Isolation is a property of every path that touches a door, a credential, a person or a record, and a menu demonstrates none of it.

Direct answer. In brief, this article covers cloud multi tenant management test plan from cloud multi tenant management test plan to cloud-multi-tenant-management. It states what to confirm about cloud multi tenant management test plan on the dated product page where a figure is involved. A cloud multi tenant management test plan is an integration guide. Every component has an obligation to the next. cloud multi tenant management test plan is the first one tested for cloud multi tenant management test plan.

  • Trace the interface — write down what hands over between lock, door and system at cloud multi tenant management test plan
  • Verify the handover — record the sequence that proves cloud-multi-tenant-management works end to end
  • Preserve the trail — store the commissioning evidence with the file for shape of run
  • For cloud multi tenant management test plan, cloud multi tenant management is what each step has to answer to.

A cloud multi tenant management test plan is the document that turns isolation from a claim into a check: a register of the operations somebody might attempt, each with a precondition, an action, an observation, a pass criterion and an artefact, arranged into gates that close on evidence.

Nothing here describes a supplier, a platform behaviour, a protocol, a certification route or a commercial term. Every product-side question is left open on purpose, to be answered in writing on the dated product page for the specific SKU. Read the plan as five gates, each with an entry condition, probes and an exit artefact.

The shape of the run

The plan runs in one direction. Gate 0 settles what a tenant is, a definition the supplier cannot write for you. Gates 1 to 4 prove the behaviour that definition makes testable. Gate 5 proves what survives when the service is not there. What can be supplied is listed in the smart door locks.

flowchart TD
    G0[Gate 0 tenant definition signed] --> G1[Gate 1 isolation probes closed]
    G1 --> G2[Gate 2 collision probes closed]
    G2 --> G3[Gate 3 role probes closed]
    G3 --> G4[Gate 4 record probes closed]
    G4 --> G5[Gate 5 degraded hour staged]
    G5 --> EV[Evidence pack signed by the client]
    G1 --> DEF[Defect logged with a named owner]
    G2 --> DEF
    G3 --> DEF
    G4 --> DEF
    G5 --> DEF
    DEF --> FIX[Fix and retest under identical conditions]
    FIX --> G1
GateEntry conditionWhat the gate settlesExit artefact
0The estate and its sites are namedWhat a tenant keys on, and how objects are scopedSigned tenant definition
1Gate 0 closedWhether scoping is enforced on every read and write pathProbe results and captures
2Gate 1 closedWhether colliding identifiers resolve inside their scopeCollision probe set and movement rule
3Gates 1 and 2 closedWhich powers exist, for whom, and how support access is containedRole matrix and support procedure
4Gate 3 closedWhat is written, who owns the clock, and who may read itRetrieval export and reader list
5Gate 4 closedWhat the doors and records do when the cloud is absentDegraded hour record per door class

The requirement these gates satisfy is set out in Cloud Multi Tenant Management for Smart Lock Projects: What Buyers Must Define First. A test plan cannot test a tenant definition that does not exist yet.

Gate 0 — what a tenant is, written before anything is installed

The word tenant is a label until someone states what it keys on. A physical site is the obvious candidate and the wrong one for many operators. A customer contract, a legal entity, a brand account or a regional grouping key differently, and the difference decides which objects sit inside a boundary.

Two decisions dominate. The first is shape: a flat estate with one tenant per site, or a hierarchy in which a group sits above its sites. The second is identifier scope: whether a door number, a credential number or a staff record is unique platform-wide or only within one tenant. If identifiers are unique only within a tenant, every lookup must carry a tenant key, and any code path that resolves an object from a bare identifier is a leak path.

DecisionOptions a supplier may proposeWhat the choice does to the planDecision owner
What a tenant keys onA site, a contract, a legal entity, a brand account, a regionDecides which objects sit inside a boundaryClient, with the integrator
Shape of the estateFlat, one tenant per site, or hierarchicalA hierarchy adds a cross-site read path that needs its own probeClient
Identifier scopeUnique platform-wide, or unique within a tenantDecides whether a bare lookup can ever be safeSupplier, in writing
Shared and moving objectsAn object tied to two scopes, or a device that changes siteDecides what a report counts and where history followsClient and supplier

Two questions belong to the client rather than the supplier. Does the operator want group-level visibility at all, and if so over which sites? And what happens to the record of a site that leaves the estate — transferred, retained, or separated at a named boundary? The fields belong in Cloud Multi Tenant Management Requirements Document Template.

Gate 1 — the negative probes that carry the isolation claim

A suite that proves legitimate requests succeed proves only that the platform works when it behaves. Isolation is a negative property: it holds when the request that should not work does not work. The pass criterion is written as the failure itself — a closed door, a refused write, a response carrying nothing belonging to the other tenant.

The path list matters more than the count. A single-object read path can be scoped while the list, search, export and scheduled job paths are not, and the scheduled job is the one that runs under a service identity nobody watches. A job carrying the tenant key in its payload is safe by construction; a job that re-reads a shared queue and filters afterwards is safe only while nobody changes the filter.

erDiagram
    TENANT ||--o{ SITE : contains
    TENANT ||--o{ ROLE : grants
    SITE ||--o{ DOOR : has
    DOOR ||--o{ CREDENTIAL : authorises
    SITE ||--o{ EVENT : records
    ROLE ||--o{ SUPPORT : opens
    SUPPORT }o--|| SITE : scoped_to

The response shape matters as much as the outcome. Three failures are distinguishable: the object is not found, the object is found and access is denied, and the request errors. The first is the correct closed door, because it does not confirm that the other tenant's object exists. The second confirms existence and becomes an enumeration channel; the third may leak scope in its message, in a reference number, or in how quickly it returns.

ProbeOperation attempted from tenant AObservation requiredArtefact keptRepair owner
P-01Read by identifier an object that exists only in tenant BThe request closes and no field of tenant B appearsRequest and response capturePlatform software owner
P-04Attach a credential belonging to tenant A to a door belonging to tenant BThe write is refused and nothing is createdWrite attempt and refusalPlatform software owner
P-07Request an export of all access eventsThe export holds tenant A events only, labelled with its scopeExport header and row countPlatform software owner
P-11Trigger a background task that resends queued workThe task carries the scope of the work it was raised forTask definition and outputPlatform software owner
P-13Open the same visible name in two scopesEvery operation resolves inside the scope that created itRequest and response captureIntegrator configuration
P-15Open support access for a fault in one siteOnly the affected site is reachable, and the session is loggedSupport session logSupplier field support

The export and the background task carry most of the weight, because both run outside the user's own request context. An export file with no scope label is a document that can be forwarded to the wrong recipient a year later.

Gate 2 — the identifiers that collide on the way

Multi-tenant estates break on collisions, not on permissions. Room 201 exists in every property. Card series restart at each site. Two properties employ people with the same name. A device identifier that was at the first site last season is at the second site this season. All of these become security decisions the moment a lookup resolves by name alone.

Each class is tested with the same shape: create an object in tenant A whose visible name duplicates one in tenant B, then attempt every operation on the A object from a B session and from an unscoped interface. The pass criterion is that the duplicate never resolves to the wrong record.| Object | Typical collision between two sites | Required behaviour ||---|---|---|| Room or door number | The same number exists in every property | A name resolves only inside its own scope || Credential or card number | Series restart per property | The number is scoped, and no bare lookup resolves it || Device identifier | A device moves between properties | History follows a stated rule and the mapping change is logged |

Movement deserves its own line, because it is the case where a correct system produces a wrong answer. When a device moves between tenants the mapping connecting it to a site changes, and the question is what happens to events arriving afterwards and to history that arrived before. The rule must be written down before the first move. The portfolio version of the problem is in Cloud Multi Tenant Management for a Multifamily or Office Portfolio, the room-by-room version in Cloud Multi Tenant Management for a Hotel Property.

Gate 3 — three kinds of power at the administration surface

The isolation claim is only as good as the roles that reach across it. Gate 3 asks a different question from Gate 1: not whether the platform refuses a bad request, but whether a legitimate request can grant a person more than the operator intended.

There are three kinds of power, and the mistake is to treat them as one permission level. The first is operational: running a site day to day, issuing credentials, changing access. The second is oversight: seeing across several sites at once, a legitimate need for a group operator and precisely the capability the operator wanted to protect, so it has to be scoped, read only, and visible to each site underneath it. The third is support: the supplier's own engineers fixing a fault inside a customer's tenant.

RoleTypical scopeReadWriteSeen in the audit trail
Site operatorOne siteThat site's doors, credentials and recordsAccess changes inside that siteYes, attributed to a person
Group oversightA named set of sitesA defined view across those sitesNothing, unless grantedYes, and visible to each site
Service supportOne site at a time, bound to a faultOnly what the fault needsOnly what the fault needsYes, with who, when and which objects
Platform operatorThe supplier's own environmentTenant metadata, not tenant contentConfiguration, disclosedYes, and disclosed

The support session is where the plan earns its keep, and the requirements worth writing down are unglamorous: support is a separate role and not an administrator impersonating a customer account; the grant is time-boxed to the fault and the client can revoke it unilaterally; it is bound to a ticket reference; the session is scoped to the affected site; every object viewed is logged and visible in the client's own audit view; and the account is never shared.

The second part of the gate is the invitation path. An invitation that carries only an email address is a route into the platform that bypasses the tenant boundary entirely. Probe it by inviting an address from the client's domain, completing the flow from a session belonging to a different site, and observing which tenant the new account lands in. The questions behind this gate are gathered in Cloud Multi Tenant Management Security Questions to Ask About Cloud Multi Tenant Management.

Gate 4 — the record, the clock and the reader list

An audit log is a record of an event with a time and a source. Event memory is the buffer a lock keeps when it cannot reach the server. They are not the same, and the difference is exactly where events are lost. In a multi-tenant estate the buffer has to be per site as well, so a site that was offline for a morning returns its own events to its own trail.

The write path and the retrieval path fail differently. The write path asks whether every expected event arrives with its time, its source and its scope. The retrieval path asks whether what was written can be produced later to somebody who was not present, in a form that cannot be confused with another tenant's records.

Two clocks will always exist. The lock or controller stamps the moment an event happened, the platform stamps the moment it received it, and a third clock may stamp the moment it reached storage. The plan names the authoritative clock, states the drift allowance the buyer accepts, and flags anything outside it. That allowance belongs to the integrator and the client, not to a supplier's material.

Element of the recordWhat the probe establishesArtefact
Event scopeEvery stored event carries the site it belongs toEvent query grouped by scope
Event completenessExpected events arrive, none duplicated on replayRun sheet compared with export
Buffer behaviourA filled buffer drains in order and loses nothingBefore and after exports
RetrievalA third party receives the record, with its scopeExport file and header

A door log at an operating site holds personal data about identified people: who presented what, at which door, at what time. The retention period, the reader list and the handling of test data belong to the operator's own legal position, not to the platform and not to the integrator.

CONFIRM-WITH-LOCAL-DATA-PROTECTION-REVIEW — which bodies in the jurisdiction of each site set a minimum or maximum retention period for access records, whether an entry log must be retained at all, who inside the operator may read one, and whether events generated by test credentials during commissioning are handled under the same rules as live records. These are per-operator and per-jurisdiction questions.

CONFIRM-WITH-LOCAL-LEGAL — whether the platform operator and the site owner are treated as one processing arrangement or two, which party is the controller of the door-side record, and what notice is owed to the people whose movements the trail records. Settle this before the retention field is filled in.

The separation between the two record types, and the handover of a trail somebody else will read later, are covered in Documenting Cloud Multi Tenant Management for Handover and Audit, and the lock-side record in Smart Lock Audit Logs.

Gate 5 — the hour the cloud is not there

Every integration has a degraded mode, and it must be named before the integration ships. The offline case is the design, not the exception. For a multi-tenant estate the degraded hour is stranger than for a single site, because a partial outage is ordinary: one site loses its link while the rest of the estate is fine.

Three losses are staged separately. The platform becomes unreachable: reports and exports stop, and the question is what else stops, including whether an authority issued minutes before the loss reached the door. The site's link drops: the site keeps working if the design puts the state on the device, and the backlog waits on the platform side, where the queue holding it must be tagged per site. The controller drops: that door stops being a controlled door for as long as the loss lasts.

Loss stagedWhat the plan must stateEvidence kept
Platform unreachableWhich operations stop, and what happens to an authority issued just beforeOperation log and door record
Site link downWhether the door keeps working, and how the backlog is scopedDoor log and queued work list
Controller unreachableThe state the door holds, and the route in for operatorsSigned observation per door class
Link restoredEvents arrive once, in order, and land in the right scopeBefore and after exports

Centralised and standalone are different designs, and the difference decides what the door does when the network is down: where the lock's state lives on a server, a severed link changes what the door can do; where it lives on the device, the door keeps its own record and the link is only about reporting.

Two boundaries sit outside the plan. The escape side and the fire interface follow the building's approved strategy: release on the escape side is the default behaviour of the exit device, and a request to exit signal tells the system that it happened rather than commanding a release. Release is not the lock's decision, so the plan tests the plumbing. A door position contact reports open or closed, and it reports a state, not a cause.

CONFIRM-WITH-LOCAL-FIRE-AUTHORITY — which authority accepts the fire-alarm interface for each opening, whether that interface is commissioned as part of the fire system rather than the access system, and what evidence that commissioning must produce. A per-building question.

CONFIRM-WITH-LOCAL-BUILDING-CONTROL-BODY — whether the escape-side release and the fail-safe or fail-secure state of each opening are recorded in the approved classification for that building. Fail-safe and fail-secure describe the state a lock takes when power is removed; the door's role and the approval decide which is correct, not the product.

CONFIRM-WITH-LOCAL-ACCESSIBILITY-REVIEW — which operable force, motion and release requirements apply to each opening in its jurisdiction, and which body signs off the installed assembly.

Where the reader and controller sit inside a tenant's estate, the wiring convention decides what evidence you can collect. Wiegand is a legacy wiring convention that says nothing about encryption or device identity, while OSDP is a specified, addressed, bidirectional protocol whose implementation is a per-SKU question to confirm on the dated product page for the specific SKU.

How each gate is closed

Collect the probes into a single register before testing starts, and choose the repair owner for each one before the run: a case that fails at eleven at night in the week before a portfolio goes live will not wait for the project to work out who was wrong. Which gate a probe belongs to is not a detail either, because a probe run before the decision it depends on has been written down produces an answer that cannot be relied on later.

A gate closes on evidence, not on a meeting: the probe results, the defect log, and a short certificate naming the scope tested, the probes that ran, the ones that did not, and the person who signed.

Findings are graded before the exit rule is written, never after. A blocking finding leaves a tenant able to read another tenant's doors, credentials or records, or leaves a record that cannot be produced. A major finding is a probe that fails inside the scoping layer but with no evidence of data crossing the boundary. A minor finding is a discrepancy in an artefact, a missing label, a wording gap in a report. The exit rule then reads: no blocking finding open, every major finding closed or carried with a named owner and a signed risk, every minor finding logged.

Retesting has one rule and one trigger. A retest re-runs the probe under the same preconditions with the same artefacts, because a retest that changes the conditions proves nothing about the original failure. Any release on either side that could touch the scoping layer re-opens the affected probes, and without that trigger the plan has quietly assumed that a passing test is permanent. The client should witness at least one negative probe, and the catalogue of ways these estates fail in production is in Cloud Multi Tenant Management Failure Patterns and How to Recover.

The five questions the estate has to answer anyway

Every integration has to answer five questions, and a cloud multi tenant management test plan rests on the same five. The gates above are the delivery mechanism; these questions are the reason the plan exists.

  • What does each side send, and what happens when it does not arrive? Gate 5, with the partition and catch-up probes. A failure looks like a queued instruction landing in the wrong trail.
  • Who owns the clock? Gate 4, with the drift probe. A failure looks like two systems producing two versions of one event.
  • What is written down, where does it live, and who can read it? Gate 4, with the retrieval probe. A failure looks like a trail that cannot be produced, or one that arrives without its scope.
  • What happens when the network, the controller or the server is unavailable? Gate 5, with the degraded-hour record. A failure looks like a door doing something nobody approved.
  • Which side owns the fix? The owner column of every probe. A failure looks like a project that discovers who was responsible only after the failure.

The distinction between a local interface and a cloud interface decides where the boundary sits before a single probe runs. A local API on a server inside the building puts an outage on your network and your support rota. A cloud API puts the outage on the supplier's service status and on an escalation path you have to know exists before you need it.

What a green result does not prove

A closed plan is evidence about the operations it names, on the version that was live when it ran, for the scopes that were configured at the time. It is not evidence about paths nobody thought to probe, nor about the periods before the plan existed. If the platform had been running for a year before the probes were written, ask for the scope configuration history for that period rather than assume it was clean.

It also does not transfer between platforms. Isolation is implemented, not inherited, and a platform that passes these probes in one estate proves nothing about a second estate with a different tenant key or a different support model.

Nor does it replace the contract. A test plan shows what was proved on one date; the obligation to keep the separation is a term of the agreement between the parties. The division of that obligation between supplier, integrator and client is the question in Who Owns Cloud Multi Tenant Management: Supplier, Integrator or Client.

Where the plan goes next

Take the register, not the framework. The register is a set of operations somebody performed and a set of results somebody signed, and it is the part that has value after the project closes.

The wider context sits in Smart Lock Cloud Management for the platform layer, and Smart Lock Cloud vs Local Management for the choice that decides what a door does when the link is down. Samples can be raised through Order Samples so the probes are written against hardware in hand rather than a datasheet, and the specification detail for the exact SKU is confirmed on the dated product page for the specific SKU. When the register is finished, the commercial conversation belongs at Request a Quote.

Το καλάθι σας
Η εμπειρία σας σε αυτόν τον ιστότοπο θα βελτιωθεί με την ενεργοποίηση των cookies Πολιτική cookies
Chat on WhatsApp