Task 3.2: Identify and resolve errors

• ARaise the Lookup activity's timeout

  Incorrect

  The failure is output size, not run time.

• BTurn on retries with an increasing delay for the Lookup

  Incorrect

  Retrying returns the same oversized output and fails again.

• CSwitch the ForEach to Sequential so it reads the three columns in smaller parts

  Incorrect

  The Lookup fails before the ForEach reads anything.

• DMake the Lookup query return only the three needed columns

  Correct

  Smaller output stays under the 4 MB limit on what an activity can return.

Activities can return at most about 4 MB of execution output (error 2001). Reducing what the activity returns, such as selecting only the columns that downstream activities need, keeps it under the limit; bulk data should move with a Copy activity rather than through activity outputs.

• AConvert the value to an array in the expression the ForEach uses

  Correct

  The payload reaches the child as a string, so an array function restores the type.

• BChange the child's parameter type from Array to Object

  Incorrect

  The value still arrives as a string, and ForEach needs an array.

• CPass the tables as a comma-separated string and loop over it in the ForEach expression

  Incorrect

  ForEach can't iterate a plain string either.

• DMark the Invoke pipeline activity to wait on completion

  Incorrect

  Waiting changes the parent's flow, not the parameter's type.

When a parent pipeline passes an array parameter to a child pipeline, the payload is passed as a string, so a ForEach over it fails with a type error. Microsoft's troubleshooting guide resolves this by applying an array function in the child's expression so the ForEach receives an array.

• AMove the %pip install cell to the end of the notebook

  Incorrect

  Inline installs are disabled in pipeline runs wherever the cell is.

• BRestart the pipeline with retries enabled on the Notebook activity

  Incorrect

  The import fails the same way on every attempt.

• CPublish the library in the environment the notebook is attached to

  Correct

  Environment libraries install for every session, including pipeline runs.

• DTurn on high concurrency for pipelines in the workspace

  Incorrect

  Session sharing doesn't install missing libraries.

Inline installation commands such as %pip install are disabled by default in pipeline runs (you can turn them on with the _inlineInstallationEnabled parameter), and libraries installed with %pip aren't retained across runs. Publishing the package in the notebook's environment makes it available to interactive and scheduled sessions alike.

• AQuery queryinsights.exec_requests_history in the eventhouse

  Incorrect

  Query insights belongs to warehouses and SQL endpoints.

• BIncrease the table's retention policy so ingestion errors are kept longer

  Incorrect

  Retention controls how long data is kept, not whether it's ingested.

• CTurn on OneLake availability for the table

  Incorrect

  This setting doesn't change how events are mapped or ingested.

• DRun .show ingestion failures on the database to see the failed operations and their errors

  Correct

  This command lists ingestion failures with error details.

• EUpdate the table's ingestion mapping so its JSON paths match the new property names

  Correct

  Mappings map source fields to columns; outdated paths break or empty the ingestion.

.show ingestion failures returns failed ingestion operations with details such as the error code and the source. Ingestion mappings map incoming data, such as JSON paths, to table columns; when producers change property names, the mapping must be updated so the data lands in the right columns.

• ASet the transaction isolation level to READ UNCOMMITTED for both pipelines

  Incorrect

  The warehouse enforces snapshot isolation on all transactions.

• BRun the two MERGEs one after the other instead of at the same time

  Correct

  Conflicts happen when two transactions modify the same table concurrently.

• CAdd a NOT ENFORCED primary key to dbo.FactSales

  Incorrect

  Unenforced constraints don't prevent concurrent writes.

• DGrant both pipelines' identities the db_owner role

  Incorrect

  Permissions aren't the cause of an update conflict.

Fabric Data Warehouse uses snapshot isolation for all transactions. When two transactions UPDATE, DELETE, MERGE or TRUNCATE the same table at the same time, one can fail with write-write conflict errors such as 24556 or 24706; avoid concurrent writers on a table, or retry the failed transaction.

• ADelete the twenty shortcuts and create them again with a new connection

  Incorrect

  Recreating every shortcut is the work you wanted to avoid.

• BTurn on the shortcut cache so reads come from Fabric

  Incorrect

  Caching doesn't fix the authorization of a broken connection.

• CReplace the connection on the Manage OneLake shortcut connections page

  Correct

  This page shows broken connections and lets you replace a connection used by many shortcuts.

• DGive every reader the Storage Blob Data Reader role on the account

  Incorrect

  External shortcuts authenticate through the connection, not each reader.

Shortcuts use shared cloud connections. The Manage OneLake shortcut connections page for a lakehouse lists every connection, highlights broken ones under Action required, shows how many shortcuts share each, and lets you replace a connection, for example when its creator left the organization.

• AWith Workspace identity authentication, after the partner grants your workspace identity Storage Blob Data Reader

  Incorrect

  Workspace identity has the same cross-tenant limit as the organizational account that already failed.

• BThrough an external data share that the partner creates for the folder

  Incorrect

  External data sharing shares Fabric item data and needs Fabric in both tenants, which the partner doesn't have.

• CWith Service principal authentication, using a service principal that holds Storage Blob Data Reader on the partner's account

  Correct

  Shortcuts to a storage account in another Entra tenant need a service principal or a SAS token, and SAS is ruled out.

• DWith the Storage Blob Delegator role and folder ACLs for the guest account

  Incorrect

  The guest still signs in as an organizational account, which is what fails across tenants.

ADLS shortcuts to storage accounts in a different Microsoft Entra tenant than the Fabric tenant require service principal or SAS authentication; organizational account and workspace identity authentication don't work across tenants, whatever roles or ACLs they hold. External data sharing is the cross-tenant option for data already in Fabric items, and it requires Fabric capacity in both tenants.

• ARetry the command immediately over the same connection in a tight loop

  Incorrect

  Immediate retries are discouraged for transient faults.

• BRecreate the warehouse from a backup

  Incorrect

  Nothing is wrong with the warehouse itself.

• CWait 5 to 10 minutes, open a new connection, then retry

  Correct

  Transient faults often clear within a minute or so during reconfiguration.

• DChange the workspace's capacity region

  Incorrect

  Region moves don't fix transient faults.

If a connection that used to work starts to fail without changes, it may be a transient fault, for example during hardware reconfiguration. Don't retry immediately: wait 5 to 10 minutes, establish a new connection and retry.

• A$Fabric

  Incorrect

  There's no built-in $Fabric consumer group.

• B$Eventstream

  Incorrect

  Eventstreams don't create a consumer group with that name.

• CA new consumer group named after the eventstream

  Incorrect

  The source uses the consumer group you enter, $Default by default.

• D$Default

  Correct

  The default consumer group for an event hub is $Default.

An Azure Event Hubs source asks for a consumer group; the default consumer group for an event hub is $Default. Separate readers should use separate consumer groups so they don't compete for the same position.

• AOnly the lakehouse's view of the file is hidden from the storage account

  Incorrect

  The delete reaches the target, not a local view.

• BIt's deleted from the storage account as well

  Correct

  Deleting a file inside a shortcut deletes it in the target when permitted.

• CThe delete fails as shortcuts are read-only

  Incorrect

  Shortcuts can be written to when the target permissions allow.

• DThe file moves to the lakehouse's recycle bin

  Incorrect

  Shortcut deletes don't move files into the lakehouse.

Deleting a file or folder inside a shortcut, with permissions in the target to do so, deletes that file or folder in the target.

• AERRORFILE works only inside an explicit transaction

  Incorrect

  Transactions don't control error file output.

• BRejected-row files are produced only for CSV and JSONL ingestion

  Correct

  The error files aren't generated for Parquet files.

• CERRORFILE must point to the warehouse's own Files folder

  Incorrect

  The location isn't why nothing was written for Parquet.

• DRejected-row files appear only after the warehouse restarts overnight

  Incorrect

  Warehouses aren't restarted to produce error files.

Rejected-row files are produced only for CSV or JSONL files ingested with T-SQL commands such as COPY INTO, BULK INSERT and OPENROWSET. No rejected-row files are generated for Parquet files, external tools such as pipelines, or Delta tables.