Real-Time Analytics Platform  ·  View 13 of 21  ·  Runtime

Event to Insight — Critical Path

One event from SDK call to dashboard pixel, with the latency budget spent at each hop against the 5-second p95 target.

Editable source SVG draw.io All views
Producer SDK
Producer SDK
API Management
API Management
Event Collector
Event Collector
Azure Event Hubs
Azure Event Hubs
Databricks Stream
Databricks Stream
Azure Data Explorer
Azure Data Explorer
Dashboard
Dashboard
1. POST /v1/events · batch 500
1. POST /v1/events · batch 500
2. validate token, quota
2. validate token, quota
3. forward with tenant claim
3. forward with tenant claim
4. schema check, event_id
4. schema check, event_id
5. publish · key = tenant
5. publish · key = tenant
6. offset committed
6. offset committed
7. 202 Accepted · 200 ms
7. 202 Accepted · 200 ms
8. micro-batch · 1 s trigger
8. micro-batch · 1 s trigger
9. enrich, window, watermark
9. enrich, window, watermark
10. streaming ingest · dedup tag
10. streaming ingest · dedup tag
11. poison event to DLQ
11. poison event to DLQ
12. materialized view update
12. materialized view update
13. KQL over view
13. KQL over view
14. result · 350 ms p95
14. result · 350 ms p95
Event to Insight — Critical Path
Event to Insight — Critical Path
Budget: 200 ms accept, 1.5 s trigger, 1.2 s process, 1.5 s ingest visibility. 4.4 s of a 5 s p95 target.
Budget: 200 ms accept, 1.5 s trigger, 1.2 s process, 1.5 s ingest visibility. 4.4 s of a 5 s p95 target.
v 1.0 · owner Data & AI Architecture · date 2026-08
v 1.0 · owner Data & AI Architecture · date 2026-08Text is not SVG - cannot display

Where the time goes

  • Accept 200 ms · trigger wait 1.5 s · enrich and window 1.2 s · ingest visibility 1.5 s
  • 4.4 s at p95 against a 5 s target; the trigger interval is the largest single lever
  • Query side: 60 ms gateway, 15 ms on cache hit, 350 ms on a materialized view

Producer contract

  • 202 Accepted means committed to the durable log, not processed — this is stated in the SDK
  • Producers batch up to 500 events or 1 second, whichever comes first
  • Retry is the producer's responsibility; event_id makes a retry safe to accept twice

Not shown

  • Token acquisition and tenant claim validation are drawn in the identity flow
  • Cache population and follower routing are drawn in the query and serving view
  • The dead-letter branch is drawn in full in the stream processing view
◀  Windowing and Late-Arriving Events All views Query and Serving Architecture  ▶
Solution Architecture v1.0 · Azure · Data & AI Architecture · 2026-0813-event-to-insight-sequence