Power BI error decoder

How to use the decoder

Error text varies a little between Desktop, the Service and versions, so each entry gives the key phrase to search for rather than a claim about the exact wording. Every entry follows the same shape: what it means, common causes, how to diagnose, how to fix, and where to practise.

Power Query

"The column 'X' of the table wasn't found"

• Means: a step refers to a column the previous step no longer has.
• Common causes: the source renamed or dropped a column; a header row changed; promoted headers picked up a different row.
• Diagnose: click through the Applied Steps until the error appears; compare Table.ColumnNames of the source with what the step expects.
• Fix: select and rename columns with MissingField.UseNull or MissingField.Ignore; agree the schema with the source owner. See schema drift.
• Practise: Sprint 08 refresh outage.

"DataFormat.Error: We couldn't convert to Number" (or Date)

• Means: a type change hit a value that isn't a number or date in the expected format.
• Common causes: currency symbols, thousands separators, "N/A", dates in another locale, a total row at the bottom of an export.
• Diagnose: View → Column quality and Column distribution; Table.SelectRowsWithErrors.
• Fix: clean before typing (remove symbols, replace "N/A" with null), set the culture on the type change, remove junk rows on purpose. See errors.
• Practise: Beginner Power Query with RawOrdersExport.

"Expression.Error: The key didn't match any rows in the table"

• Means: a navigation step (Source{[Name="Sheet1"]}[Data], a database table by name) found nothing.
• Common causes: a renamed sheet, table or database object; a different file with a different structure.
• Fix: navigate by a stable property, add a check that fails with a clear message, or agree naming with the file's owner.

"Formula.Firewall: Query … references other queries or steps, so it may not directly access a data source"

• Means: the privacy firewall stopped data from one source being combined with another in a way that could leak it.
• Common causes: using a value from one source (a parameter table in Excel) to build a query against another (a SQL server); privacy levels not set.
• Fix: set privacy levels for each source; restructure so each query reads one source and a separate query combines them. See combining sources safely.

"This dataset includes a dynamic data source" (the Service won't refresh)

• Means: the Service can't determine the data source before running the query, so it can't check credentials and won't schedule refresh.
• Common causes: a URL or connection string built by concatenating values from data.
• Fix: keep the base URL static in Web.Contents and pass the variable parts through RelativePath and Query. See APIs and Web.Contents.

"DataSource.Error: Web.Contents failed to get contents from … (4xx/5xx)"

• Means: the HTTP call failed. The status code says why: 401/403 credentials or permissions, 404 wrong path, 429 rate limited, 5xx the API is unhealthy.
• Fix: check the credential type on the data source, the path, and the API's paging and rate limits; add a retry policy upstream if the API is flaky. See pagination.

DAX and modeling

"A table of multiple values was supplied where a single value was expected"

• Means: an expression that needs one value got a column with several (for example VALUES or a column reference in a measure).
• Fix: use SELECTEDVALUE ( column, fallback ), an aggregation (MAX, SUM), or iterate with SUMX/MAXX. Decide what the result should be when several values are visible.

"A circular dependency was detected"

• Means: two objects depend on each other: a calculated column that refers to a measure that uses that column, or calculated columns on both sides of a relationship that reference each other.
• Fix: move the logic upstream, or rewrite so one of the two doesn't depend on the other (ALLEXCEPT and RELATED in calculated columns are frequent culprits).

"The value for 'Column' cannot be determined… no current row"

• Means: a measure referenced a column directly, as if there were a current row. Measures have no row context.
• Fix: aggregate it (SUM ( FactSales[Quantity] )) or iterate (SUMX ( FactSales, … )). See measures vs calculated columns.

"You can't create a relationship between these two columns because one of the columns must have unique values"

• Means: neither side is unique, so a one-to-many relationship isn't possible.
• Common causes: duplicate keys in the dimension, blank keys, or two facts at different grains being related directly.
• Fix: fix the dimension's duplicates; relate facts through a shared dimension; resist many-to-many as a way to make the error go away. See relationship not filtering.

"Function 'CALCULATE' has been used in a True/False expression that is used as a table filter expression"

• Means: a boolean filter argument contains a measure or CALCULATE, which simple filter predicates don't allow.
• Fix: compute the value in a variable first, or use FILTER ( ALL ( column ), condition ).

"Visual has exceeded the available resources"

• Means: the query for one visual used more memory or time than the capacity allows.
• Common causes: a table visual with high-cardinality columns and no filter; a measure that materialises huge intermediate tables; many-to-many or bi-directional filters exploding row counts.
• Fix: filter the visual, reduce columns, rewrite the measure (DAX Studio's Server Timings will show the materialisation). See Performance Clinic.

"The resultset of a query to external data source has exceeded the maximum allowed size of '1000000' rows"

• Means: a DirectQuery visual asked the source for more than a million rows of intermediate results.
• Fix: aggregate at the source, add filters, use aggregation tables, or switch the table to Import. See the DirectQuery section of the clinic.

Refresh

"Scheduled refresh has been disabled"

• Means: the Service turned off the schedule. It does this after repeated consecutive failures, and pauses schedules on models nobody has visited for a long time.
• Fix: fix the underlying failure first (refresh history has the error), then re-enable the schedule; set up failure notifications to a group, not one person.
• Practise: Sprint 08.

"Resource Governing: This operation was canceled because there wasn't enough memory"

• Means: the refresh needed more memory than the capacity allows. Refresh can need roughly double the model's size while the new copy is built.
• Fix: remove unused columns and high-cardinality text, use incremental refresh, refresh tables or partitions separately (XMLA), or move to a larger capacity. See model performance.

Refresh timed out

• Means: the refresh ran longer than the limit (shorter on shared capacity than on Premium or Fabric capacity).
• Fix: make queries fold, filter early, use incremental refresh, push heavy transformation upstream.

Gateway and credentials

"The credentials provided for the … source are invalid"

• Means: the stored credential no longer works: a password changed, an account was disabled, an OAuth token expired.
• Fix: update the credential on the semantic model's settings or the gateway connection; switch to a service account or service principal with a documented owner. See the production runbook.

Gateway offline or unreachable

• Means: the Service can't reach the on-premises data gateway.
• Common causes: the gateway machine is off, asleep or patched and not restarted; the service account's password changed; outbound network rules changed; the gateway version is too old.
• Fix: check the gateway's status in Manage connections and gateways, restart the gateway service, check outbound connectivity, keep gateways in a cluster of at least two members. See gateway offline.

"A network-related or instance-specific error occurred while establishing a connection to SQL Server"

• Means: the gateway (or Desktop) can't reach the database server.
• Fix: test from the gateway machine itself (not your laptop): server name, port, firewall, DNS, and whether the gateway service account can authenticate.

Security and access

A user gets "You don't have access" or sees an empty report

• Means: missing permission on the workspace, app, report or underlying semantic model (Build permission for a report in another workspace), or RLS returns no rows for them.
• Diagnose: check access in this order: app audience → workspace role → item sharing → semantic model permissions → RLS role membership. Test as role shows what RLS lets them see.
• Fix: see the security field guide.

RLS seems to do nothing

• Means: usually not an error at all: RLS doesn't restrict workspace Admins, Members or Contributors, only users with Viewer access (or app and share access).
• Fix: test with a Viewer account or View as; never judge RLS from a developer's account. See RLS testing checklist.

Deployment

Report in Production shows Test data, or the wrong numbers after deployment

• Means: parameters or data source rules weren't applied to the target stage, or a report still points at the Test semantic model.
• Fix: set deployment rules (parameter and data source rules) for each stage, and add a post-deployment check that queries one known number. Sprint 07 is this incident.

PBIP or Git merge conflicts in TMDL or report JSON

• Means: two people changed the same object.
• Fix: resolve in the text files (TMDL is designed for it), reopen in Desktop to validate, and agree smaller, more frequent merges. Sprint 06 walks through one.

Fabric and REST API

401 Unauthorized / 403 Forbidden from the Power BI or Fabric REST API

• Means: the token is invalid or for the wrong resource (401), or the identity isn't allowed (403).
• Common causes: wrong scope (https://analysis.windows.net/powerbi/api/.default for Power BI APIs); the tenant setting that lets service principals call the APIs is off or doesn't include the principal's security group; the service principal isn't a member of the workspace.
• Fix: check the tenant settings under the admin portal's developer settings and the workspace access list. See automate Power BI.

429 Too Many Requests

• Means: you hit an API rate limit.
• Fix: honour the Retry-After header, back off exponentially, batch requests, and use the admin scanner APIs for tenant-wide inventory instead of calling per item.

Capacity throttling: interactive requests delayed or rejected

• Means: the capacity used more compute than it has over the smoothing window, so Fabric delays and then rejects work.
• Diagnose: the Fabric Capacity Metrics app: find the items and operations using the most CU, and when.
• Fix: see capacity overloaded.

Direct Lake model errors instead of falling back

• Means: a Direct Lake on OneLake model can't answer a query in Direct Lake (for example a guardrail is exceeded) and, by design, doesn't fall back to DirectQuery.
• Fix: check the table sizes against the capacity's guardrails, reduce row groups and columns, or reconsider the mode. See Fabric architecture.