3.7 — Optimize Eventstreams and Eventhouses
DP-700 Domain 3 · Skill 3.7
What You Need to Know
Real-time performance has two halves. Getting events in quickly enough is about eventstream throughput and ingestion settings. Querying them quickly enough is about eventhouse policies and KQL habits.
Eventstreams
| Lever | Detail |
|---|---|
| Throughput level (eventstream settings) | Low (under 10 MB/s, the default), Medium (10–100 MB/s) or High (over 100 MB/s). Higher levels raise the ceilings for sources and destinations, and use more capacity |
| Source partitions | With an Azure Event Hubs source that has fewer than 4 partitions, the partition count caps throughput whatever level you pick |
| Retention | How long the eventstream keeps events (1 day by default, up to 90), so destinations can catch up after an outage |
| Process before you land | Filter out unneeded events and fields in the stream. Pre-aggregate with Group by windows when consumers only need summaries |
| Fan out instead of duplicating | Use one eventstream with several destinations, or derived streams, rather than several eventstreams reading the same source |
Eventhouses and KQL databases
| Lever | Detail |
|---|---|
| Caching policy (hot cache) | Keeps the recent period on fast local SSD. Queries over hot data are much faster. Set it to the period people actually query (for example 30 days), shorter than retention |
| Retention policy | How long data is kept at all. Longer retention costs storage but doesn't slow queries on hot data |
| Streaming vs queued (batched) ingestion | Streaming ingestion gives about one-second latency for small, frequent writes. Queued ingestion batches data for efficiency. Tune the ingestion batching policy (time, item count, size) to trade latency for efficiency |
| Update policies | Transform at ingestion so queries read clean, typed tables instead of parsing raw JSON every time |
| Materialized views | Precompute common aggregations or deduplication (arg_max) so dashboards don't scan raw data |
| Partitioning policy | Partitions the data by a high-cardinality key (such as TenantId) when most queries filter on it |
| OneLake availability latency | TargetLatencyInMinutes decides how quickly the Delta copy appears. Lower means fresher data but more small files |
KQL query habits
- Filter on time first (
where Timestamp > ago(1h)), then on other columns. Filters on the datetime column prune whole extents. - Prefer
hasovercontains(term index versus substring scan), and==over=~when the case is known. projectonly the columns you need, and as early as you can.summarizebeforejointo shrink both sides. Put the smaller table on the left of a join.- Use
materialize()or aletstatement when one subquery is reused.
Key Concept: Make eventstreams faster with throughput level and partitions. Make eventhouses faster with hot cache, transformation at ingestion and precomputed aggregates. Make KQL faster with early time filters and narrow projections.
Exam Traps
- Raising the eventstream throughput level while the Event Hubs source has only 1 or 2 partitions
- Lengthening retention to make queries faster (the caching policy controls hot data)
- Running heavy JSON parsing in every dashboard query instead of an update policy
- Using
containsfor whole-word searches over billions of rows
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.
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.
Sources
- Configure settings for a Fabric eventstream, Microsoft Learn
- Caching policy (hot and cold cache), Microsoft Learn
- Ingestion batching policy, Microsoft Learn
- Materialized views, Microsoft Learn
- Best practices for KQL queries, Microsoft Learn