Identifying When a Gateway Is Required
A sales report works perfectly in Power BI Desktop. The analyst publishes it to a workspace, which is a collaborative container in the Power BI service for reports, semantic models, and related items. The report initially displays data, but the service can’t obtain the latest sales records. Should the analyst immediately install an on-premises data gateway, or would that add infrastructure without solving the real problem?
The evidence begins with a connectivity question. When the Power BI service needs data, can the service reach every required source directly? If it can, a gateway shouldn’t be added merely because gateways are familiar or because the interface offers gateway-related settings. If a required source isn’t directly accessible through cloud connectivity, particularly because it’s on premises or behind a protected network boundary, then the service needs a gateway path to reach it.
That distinction is the core of the PL-300 skill in this lesson: identify when a gateway is required. Notice that the skill isn’t simply to recognize the word gateway or remember the location of a setting. You’re being asked to judge whether the service’s path to the data requires one. That judgment starts with the data sources and their network boundaries, not with a memorized interface procedure.
Consider a fictional manufacturer called Northwind Cycles. Its sales analysts build a governed reporting solution from two sources. Current order records are stored in a database inside the company network. Regional targets come from a cloud source that the Power BI service can reach directly. A semantic model, meaning the governed Power BI data model used by reports and other consumers, combines information from both sources.
An analyst’s computer can reach the internal order database while connected to the company network. As a result, work in Power BI Desktop can appear completely healthy. Power Query can retrieve the expected rows, the model can contain the expected tables, calculations can return plausible values, and report visuals can respond to filters. None of those observations proves that the Power BI service has the same network path as the analyst’s computer.
After publication, the service becomes the relevant operating environment. It can already reach the cloud source directly, so that source alone doesn’t justify installing a gateway. The internal order database is different.
If the service can’t directly access that protected source, a gateway is required for the service to reach it. One semantic model can therefore require careful source-by-source analysis. Calling the whole solution cloud-based or on-premises without identifying each source hides the decision that matters.
This is also why a working report isn’t sufficient evidence. The report may be displaying data that was already loaded into the semantic model. A consumer can open it, select a region, and see numbers even though the service currently lacks a working path to retrieve newer order records. Report rendering and access to previously loaded data aren’t the same test as successful service access to the underlying source.
The nearest wrong answer is to treat a gateway as a general repair tool. Suppose Northwind Cycles had moved both its orders and targets to cloud sources that the service could reach directly. Installing a gateway anyway wouldn’t be justified by the requirement. The decision rule remains straightforward: inventory every source used by the semantic model, identify the relevant network boundary for each one, and add a gateway only where the service needs a path to data it can’t reach directly.
It helps to separate this service decision from work performed in other Power BI layers. Power Query shapes and transforms data when its queries run. The semantic model defines the governed model that reports use. Data Analysis Expressions, or DAX, calculates results in model context.
The report layer controls how consumers interact with and view those results. A gateway requirement belongs to service connectivity. Changing a visual, rewriting a measure, or adding an unrelated query transformation doesn’t create a network path from the service to a protected source.
Imagine that Northwind’s report shows an old total for yesterday’s orders. An analyst suspects a DAX calculation because the total appears in a card visual. Before changing the measure, the analyst should determine whether newer source data ever reached the semantic model. If the service couldn’t reach the order database, the calculation might be operating correctly on outdated data. Editing DAX could conceal the symptom temporarily while leaving the actual connectivity problem untouched.
The same discipline prevents the opposite mistake. If the service successfully reaches the source and obtains the expected data, but a measure produces an incorrect total under one region filter, gateway work isn’t supported by that evidence. The observable failure is then associated with model or calculation behavior, not source reachability. A gateway addresses the service’s need to reach inaccessible data. It doesn’t replace diagnosis in Power Query, model relationships, DAX, report interactions, or service permissions.
Return to the original Northwind model. The analyst lists the internal orders database and the cloud target source separately. The cloud target source is directly reachable by the service.
The internal database sits behind a network boundary that prevents direct cloud connectivity. That inventory provides a reasoned answer: the model requires a gateway path because one required source can’t otherwise be reached by the service. The gateway decision is tied to that source and boundary, not merely to the report’s subject or the number of visuals it contains.
Now suppose another analyst says, “The report opened after publication, so the configuration is correct.” That conclusion goes beyond the evidence. Opening the report verifies one consumer experience. It may show that the report and semantic model are available to that consumer, but it doesn’t demonstrate that the service can retrieve current data from every source. Correct validation has to observe the behavior at the layer where the requirement operates.
For a gateway decision, useful evidence concerns refresh or access state. Can the service reach the required source through the intended path? When retrieval occurs, does the semantic model receive the expected current data? Does the relevant refresh or access state indicate success rather than a connectivity failure?
Those observations are stronger than the fact that a configuration page accepted a setting. A successful click proves that an interface action completed. It doesn’t, by itself, prove end-to-end data access.
A controlled test makes the distinction clearer. Assume the internal order database contains a known new batch of records that should affect the reporting result. Before testing, the analyst records the expected outcome. The service should retrieve those records, the semantic model should reflect them, and the report should then show their effect under the appropriate filter. If the retrieval fails before the new records enter the model, changing the report filter won’t repair the source path.
The analyst reproduces the failing condition from the service rather than relying on a successful Desktop result. The cloud targets remain available, but the latest internal orders don’t arrive. That split result is important.
It suggests that the model isn’t universally broken. One source path works while another doesn’t. The analyst inspects the connectivity arrangement governing the protected internal source instead of redesigning the cloud connection or editing unrelated visual formatting.
Once the source can be reached through the required gateway path, the analyst repeats the same test. This time, the expected order batch arrives and the semantic model’s data reflects it. The relevant report result changes accordingly. The analyst also checks an ordinary case that worked before, such as a previously available region, to make sure the correction addressed the failed condition without disturbing the normal one. Verification now consists of observable behavior across the path, not confidence in the configuration screen.
There are several opportunities to stop too early. The first is when gateway-related settings become available. Availability doesn’t prove appropriateness. The second is when a source entry appears to be associated with a model.
Association doesn’t prove that the service can retrieve the required data. The third is when a report opens. Report access doesn’t prove that the displayed data is current. Each stopping point confirms something, but none alone confirms the complete business outcome.
Operational intent matters here. Northwind’s requirement isn’t “use a gateway.” The requirement is to keep a governed sales solution supplied with data from an internal database that the service can’t directly reach.
The gateway is required because of that source location and network boundary. If the source architecture later changes so that the service can reach the data directly through supported cloud connectivity, the original reason for requiring that gateway path may no longer apply. The decision should be revisited against the current sources rather than preserved as a ritual.
You should also resist deciding from a source’s friendly display name. A connection called Cloud Sales doesn’t prove direct cloud accessibility. A source called Corporate Orders doesn’t, by its name alone, prove that a gateway is required. Determine where the actual data is available from the service’s perspective and what boundaries stand between the service and that data. Labels can help people organize connections, but they aren’t evidence of network reachability.
Mixed-source models deserve special attention because the easiest source can make the entire solution look healthy. In Northwind’s case, cloud targets might update while protected order data remains old. A report that compares actual sales with targets could then combine fresh values from one source with stale values from another. The visual might still look polished and calculations might still execute. Without source-level validation, the consumer has no obvious reason to know that the two sides don’t represent the same update state.
That consequence explains why the source inventory needs to be complete. Start with the semantic model and identify every source it depends on, including a source that supplies only one small supporting table. For each source, ask whether the Power BI service can reach it directly. If the answer is no because the source is on premises or protected by a network boundary, that source creates the gateway requirement. If the answer is yes, don’t route it through a gateway merely to make every source look administratively similar.
Troubleshooting should follow the same map. First, state the expected business result in observable terms. For Northwind, the latest internal order batch should become available to the governed model and affect the relevant report result.
Next, identify where actual behavior first diverges from that expectation. Did Power Query produce the expected output when retrieval ran? Did the semantic model receive the expected data?
Did the calculation return the right result from that data? Did the report interaction display it under the selected filter? Or did the service fail to reach the source before those later layers were involved?
This sequence keeps you from altering whichever component is easiest to see. Report visuals are visible, so they often attract attention first. Yet a visual can’t display data that never arrived.
DAX measures can be edited, but a correct measure can only evaluate the data available to the model. Power Query can shape records, but a transformation isn’t a substitute for the service having a viable path to retrieve them. The earliest failed layer usually provides the most useful evidence about where to investigate.
Suppose the gateway path has been addressed and the refresh or access state now reports success, but the Northwind report still omits the new orders under one regional filter. That changes the troubleshooting focus. The source is reachable, and the expected records have entered the model. The remaining discrepancy must be validated in the model, DAX, or report context that controls the filtered result. Continuing to change gateway configuration would ignore the evidence already collected.
The reverse is equally important. If the report result is wrong because the new orders never entered the model and the service reports that it can’t access the protected source, redesigning the filter interaction is premature. Restore the required service-to-source path, confirm that the expected data arrives, and only then judge downstream behavior. This approach doesn’t assume that every failure has one cause. It isolates the first broken part of the path before moving farther along it.
A normal case and an edge case give you better confidence than a single successful observation. The normal case might be an existing region whose previously loaded data already appears. The edge case might be the newly added order batch that requires current source access.
If only the normal case works, you haven’t demonstrated current retrieval. After correcting the gateway path, both cases should behave as expected at their respective layers. If one still fails, the difference tells you where additional diagnosis is needed.
Consumer experience is the final practical check, but it shouldn’t be confused with the whole validation process. The business user cares that the sales report presents the expected current information. The analyst needs to connect that result backward through the report, calculation, semantic model, retrieval operation, and service access state. When those observations agree, there’s a defensible chain of evidence. When they don’t, the disagreement prevents a premature claim that the solution is correct.
This is why there isn’t one universally correct gateway setting for every Power BI report. Two reports can look nearly identical while depending on very different source arrangements. One model might use sources the service can reach directly.
Another might depend on a database isolated behind a company network boundary. The report design alone doesn’t decide the requirement. The source inventory and the service’s actual connectivity do.
For an applied PL-300 question, look past distracting details and identify the responsible layer. If the requirement concerns shaping columns or rows, that points toward query-time work. If it concerns model calculations or filter context, inspect the semantic model and DAX. If it concerns what a consumer sees or selects, inspect report behavior. If the Power BI service must obtain data that it can’t reach directly, the gateway requirement is a service connectivity decision.
Then demand evidence appropriate to that decision. Don’t settle for “the report published,” “the setting saved,” or “it worked in Desktop.” Determine whether every required source has a viable service access path.
Observe whether expected data reaches the semantic model. Check the relevant refresh or access state, and confirm that the resulting consumer behavior matches the stated requirement. Those checks turn a plausible configuration into a validated solution.
Northwind’s answer is now specific. The cloud targets don’t require a gateway merely because they participate in the same model. The protected internal order database does require a gateway when the service can’t reach it directly. A successful Desktop connection doesn’t erase that service boundary, and an opening report doesn’t prove that current orders arrived. Source-by-source connectivity analysis identifies the requirement, while service and data observations verify that the chosen path actually works.
A gateway decision starts with the service’s path to every source, not with a report’s appearance or an available setting. If the service can reach a cloud source directly, don’t add a gateway without a connectivity reason. If required data is on premises or behind a protected boundary that the service can’t cross directly, a gateway is required for that path.
Then verify more than configuration success. Confirm that the service reaches the source, the expected data enters the semantic model, and the consumer receives the intended result. That combination of source inventory, network-boundary analysis, and observable validation lets you distinguish a necessary gateway from unnecessary infrastructure and a working data path from a report that merely opens.
Decide whether a gateway is required by tracing where the source resides and how the service reaches it. What connection or refresh result would confirm that the gateway path is working? Consult the matching episode reference when you are ready to inspect the terminology and reproduce the verification process. This lesson was produced by Bare Metal Cyber dot com.