Phase 3 Store Baseline

Purpose

This note captures the first local benchmark baseline for the Phase 3 BEAM event store.

These numbers are not promises or production targets. They are a reproducible starting point for future comparisons.

Command

Run from the repo root:

cd beam
set MIX_ENV=test
mix dg.store.bench --traces 100 --events-per-trace 8 --batch-size 250 --payload-bytes 512

Equivalent workload:

• 100 traces
• 8 events per trace
• 800 total events
• 512 bytes of synthetic payload padding
• batch reads of 250 events

Environment

• host: local Windows development machine
• database: Docker-hosted PostgreSQL using the repo docker-compose.yml
• Mix environment: test
• repo state: Phase 3 event-store implementation with Postgres trigger-based sequence enforcement and advisory locks

Result

Measured output:

• append_ms: 6421.2
• append_events_per_second: 124.59
• batch_read_ms: 17.0
• batch_read_events_per_second: 47064.36

Interpretation

• Append throughput is currently correctness-first rather than optimized. Each write acquires a trace lock, validates parity rules, and commits its own transaction.
• Batch-read throughput is already strong enough for the first projector catch-up and replay workflows planned in Phase 4.
• Future benchmark comparisons should focus on whether we improve append throughput without weakening semantic guarantees.

Follow-Up Questions

• Should Phase 4 move more validation into fewer round trips while preserving the same error categories?
• Should Phase 4 or Phase 5 introduce write batching for producers that can preserve per-trace ordering?
• When we introduce projector workers, what event-volume target should become the first operator-facing SLO?