68 lines (51 loc) · 2.08 KB

k6 Load Test Results

Date: 2026-03-07 Environment: Local (single machine), PostgreSQL 16 (Docker), Node.js v24.11.1 k6 version: v1.6.1

Scenario

Subscribers: Ramp 0 → 1,000 WebSocket clients over 30s, hold 60s, ramp down 10s
Inserters: 50 INSERTs/sec for 90s (4,500 total inserts)
CDC engine: PostgresProvider (WAL logical replication via pgoutput)
Batching: EventBatcher (50 events or 16ms flush)

Results

Metric	Value
Peak concurrent WebSocket clients	1,000
Total events delivered	3,976,800
Event throughput	~37,872 events/sec
Total inserts	4,500 (0% failure)
Insert rate	~43/sec

Event Latency (DB write → client receive)

Percentile	Latency
p50	41ms
p90	84ms
p95	96ms
max	165ms

WebSocket Connect Time

Percentile	Latency
p50	6ms
p90	126ms
p95	272ms
max	579ms

HTTP Insert Latency

Percentile	Latency
p50	61ms
p90	97ms
p95	106ms

Thresholds

Threshold	Target	Actual	Status
Event latency p95	< 200ms	96ms	PASS
WS connect time p95	< 500ms	272ms	PASS
Events received	> 1,000	3,976,800	PASS

Network

Data received: 1.4 GB (14 MB/s)
Data sent: 710 KB (6.8 KB/s)

Key Takeaways

p95 event latency of 96ms — well under the 100ms target for most percentiles
1,000 concurrent clients with zero connection failures
Fan-out: each of the 4,500 inserts was delivered to ~883 average connected clients (3,976,800 / 4,500)
EventBatcher keeps throughput high while maintaining low latency