Launch Deal · 50% off 1st month ·

3.4 — Identify and Resolve Eventhouse, Eventstream and T-SQL Errors

DP-700 Domain 3 · Skill 3.4

What You Need to Know

Real-time and SQL items fail differently from batch jobs. A stream can keep running while it silently drops events, and a T-SQL statement can fail because of a concurrency rule that doesn't exist in Spark.

Eventhouses (KQL databases)

Symptom Where to look Common fixes
Data doesn't appear in a table .show ingestion failures; the IngestionResults table in workspace monitoring Fix the ingestion mapping (JSON path or column order), the data format, or the target table's schema
Some columns are always empty A mapping or column name mismatch Correct the mapping; check casing (KQL is case-sensitive)
An update policy's target is missing rows The policy's function fails on some batches Run the function on its own to see the error; make the policy transactional if source and target must stay consistent
Queries time out or hit memory limits Very broad scans (no time filter), contains on huge tables, or joining large tables both ways Filter by time first; use has rather than contains; put the smaller table on the left of a join; pre-aggregate with materialized views
Old data is "gone" The retention policy removed it Lengthen the retention period on the table or database

Eventstreams

Symptom Where to look Common fixes
A source shows an error or no incoming events The source node's status; Data insights; Runtime logs Fix the source credentials or connection string; check the consumer group; restart the source
Events reach the stream but not the destination The destination node's errors in Runtime logs Make the event schema fit the destination table (type mismatches and missing fields); add a Manage fields step to cast them
Events dropped after a transformation The operator's configuration (a filter that's too strict, Union with fields that don't match) Preview the test results on each node in Edit mode; align the field names and types
Lag grows during peaks Throughput is set too low for the volume Raise the eventstream's throughput level; check how many partitions the source has
Changes don't take effect The edits weren't published Publish them (pause and resume nodes if the setting needs it)

T-SQL in a warehouse or SQL analytics endpoint

Error or symptom Cause Fix
Update conflict, or a transaction aborted on concurrent writes Snapshot isolation: two transactions changed the same table at the same time Serialize the writers (for example by scheduling them apart); retry; keep transactions short
"Not supported" for a T-SQL feature Features differ in Fabric (some data types, triggers, some DDL) Use supported alternatives, such as varchar instead of text, or CTAS patterns
INSERT, UPDATE or DELETE fails on a lakehouse table The SQL analytics endpoint is read-only Write with Spark, or load the data into a warehouse
New lakehouse rows don't show up in SQL yet Metadata sync lag on the SQL endpoint Wait, or run Refresh SQL analytics endpoint (there's also a pipeline activity for it)
Permission denied Missing GRANT or ReadData, or RLS filtered the rows Grant the object permission; check the security policies
A slow or failing query Find it in queryinsights.exec_requests_history (status and error); see what's running in sys.dm_exec_requests; end a runaway with KILL

Key Concept: For streams, check each node's status and runtime logs, then whether the schema fits the destination. For eventhouses, .show ingestion failures and mappings. For warehouses, snapshot-isolation conflicts, the read-only lakehouse endpoint and Query insights.

Exam Traps

  • Trying to UPDATE a lakehouse table through its SQL analytics endpoint
  • Expecting an eventstream edit to apply before it's published
  • Treating a write-write conflict as a permission problem
  • Querying an eventhouse table without a time filter and then blaming capacity

Check your understanding

Practice Scenario

🔒 This practice scenario is part of Skillmarked Pro

The lessons stay free. Pro unlocks the other 16 DP-700 practice scenarios and the 50-question timed mock exam, plus everything for every other exam.

See plans

Launch Deal: first month €9.95, then €19.90/month. Enter code LAUNCH50 at checkout. Ends 1 December.

Cancel anytime. Or €199 a year, which saves 17%.

Not sure yet? Try 10 free mock exam questions.

Already Pro? Log in.