Overview#
This is a compact orientation to Real-Time Intelligence — the hot-path/cold-path shape, the core items involved, and when to reach for streaming over batch. For the full practitioner-depth treatment — the Real-Time hub as catalog, the current Eventstream connector lineup, Activator's Copy-job automation, Outbound Access Protection, and internals — see Real-Time Intelligence: Fabric's End-to-End Streaming Platform.
Real-Time Intelligence in Fabric covers event ingestion, transformation, storage, analytics, visualization, and automated actions for streaming and event-driven scenarios [S1]. Eventstreams route data in motion, Eventhouses and KQL databases store and query high-volume events, and Activator can trigger actions from live conditions [S1] [S2].
Core concepts#
The Real-Time hub catalogs streams, events, and KQL database tables so users can discover and react to data in motion [S1] [S3]. Eventstreams collect, transform, and route high-volume event data to multiple destinations, including KQL databases and other Fabric targets [S1]. Eventhouses contain KQL databases optimized for structured, semi-structured, and unstructured event data, with hot cache and standard storage tiers controlling cost and performance [S2].
OneLake availability lets Eventhouse data appear as Delta data for other Fabric engines, creating a cold path for BI, Spark, warehouse, and lakehouse analysis [S2].
How it works and best practices#
Use the hot path for seconds-fresh operational monitoring, anomaly investigation, and actions driven by live events [S1] [S2]. Use the OneLake cold path when the same event history must be joined with batch data, consumed by Power BI Direct Lake, or analyzed through Spark and warehouse tools [S2] [S4].
Plan retention, cache windows, and OneLake batching together. A shorter hot-cache window reduces premium-tier cost, while OneLake availability uses batching that trades cold-path freshness against downstream file quality [S2].
Implementation example#
The core real-time design is to route the same event stream into a hot operational path and, when historical reuse matters, a OneLake cold path [S1] [S2]. Eventhouse/KQL serves live investigation, while OneLake availability exposes the data to other Fabric engines for historical joins and BI [S2].
// Illustrative KQL over event data in a KQL database.
DeviceEvents
| where Timestamp > ago(15m)
| summarize ErrorCount = countif(Level == "Error") by DeviceId, bin(Timestamp, 1m)
| where ErrorCount > 0
Inference: the query illustrates the kind of hot-path operational aggregation KQL enables. The sourced mechanics are Real-Time Intelligence event analytics, Eventhouse/KQL storage/query, and Activator/action patterns [S1] [S2].
streaming_routes:
eventstream:
source: mqtt_or_kafka_or_managed_stream
hot_destination: eventhouse_kql_database
cold_destination: onelake_availability
actions:
condition_source: kql_query_or_realtime_hub_event
target: activator_rule
The routing sketch is grounded in Eventstreams, Eventhouse/KQL, OneLake availability, and Activator capabilities; exact connector choices depend on the approved source and tenant configuration [S1] [S2].
What goes wrong#
Streaming is not a default replacement for batch analytics. If freshness needs are daily or slower, a lakehouse or warehouse pipeline can be simpler and cheaper than maintaining a hot event path [S1] [S2]. Turning on OneLake availability before schema and security decisions are stable can also create operational friction because some table changes require disabling and backfilling availability [S2].
Internals#
Architecture & design#
The architecture has a hot path through Eventstreams and Eventhouse/KQL plus a cold path through OneLake availability [S1] [S2]. Eventhouse storage separates a premium cache tier from a standard tier, and retention/caching policy choices decide what stays hot [S2].
How it works internally#
Eventhouse records are organized for KQL query over high-volume event data, while OneLake availability writes Delta-formatted data for other engines [S2]. Activator evaluates conditions over supported sources and can trigger notifications, Fabric jobs, functions, or Power Automate flows [S1].
Performance characteristics#
Hot queries benefit from cache-resident data, while cold-path freshness and file quality are shaped by OneLake availability batching [S2]. Eventhouse has throttling behavior under pressure, so workspace and capacity planning remain part of real-time design [S2].
Worked example#
A telemetry solution can route device events through Eventstreams into an Eventhouse for live dashboards and alerts, while enabling OneLake availability so long-history events can be joined with reference data in a lakehouse and reported through Power BI [S1] [S2] [S4].