Files

master f5b5f24d95 Add StellaOps.Workflow engine: 14 libraries, WebService, 8 test projects

Extract product-agnostic workflow engine from Ablera.Serdica.Workflow into
standalone StellaOps.Workflow.* libraries targeting net10.0.

Libraries (14):
- Contracts, Abstractions (compiler, decompiler, expression runtime)
- Engine (execution, signaling, scheduling, projections, hosted services)
- ElkSharp (generic graph layout algorithm)
- Renderer.ElkSharp, Renderer.ElkJs, Renderer.Msagl, Renderer.Svg
- Signaling.Redis, Signaling.OracleAq
- DataStore.MongoDB, DataStore.PostgreSQL, DataStore.Oracle

WebService: ASP.NET Core Minimal API with 22 endpoints

Tests (8 projects, 109 tests pass):
- Engine.Tests (105 pass), WebService.Tests (4 E2E pass)
- Renderer.Tests, DataStore.MongoDB/Oracle/PostgreSQL.Tests
- Signaling.Redis.Tests, IntegrationTests.Shared

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-03-20 19:14:44 +02:00

3.3 KiB

Raw Blame History

Signal Driver / Backend Matrix 2026-03-17

Purpose

This snapshot records the current six-profile synthetic signal round-trip comparison:

Oracle
PostgreSQL
Mongo
Oracle+Redis
PostgreSQL+Redis
Mongo+Redis

The matrix is artifact-driven. Every value comes from measured JSON artifacts under TestResults/workflow-performance/. No hand-entered metric values are used.

The exact generated matrix artifact is:

WorkflowPerfComparison/20260317T210643496-workflow-backend-signal-roundtrip-six-profile-matrix.md
WorkflowPerfComparison/20260317T210643496-workflow-backend-signal-roundtrip-six-profile-matrix.json

Serial Latency

Primary comparison rows in this section are:

Signal to first completion avg
Drain-to-idle overhang avg

Signal to completion avg is a mixed number. It includes both real resume work and the benchmark drain policy.

Metric	Unit	Oracle	PostgreSQL	Mongo	Oracle+Redis	PostgreSQL+Redis	Mongo+Redis
End-to-end avg	ms	3091.73	3101.57	151.36	3223.22	3073.70	3099.51
End-to-end p95	ms	3492.73	3143.39	308.90	3644.66	3090.75	3162.04
Start avg	ms	105.88	16.35	38.39	110.03	8.32	21.77
Signal publish avg	ms	23.39	11.47	14.30	23.90	7.55	10.43
Signal to first completion avg	ms	76.15	37.56	55.06	81.46	31.77	40.88
Signal to completion avg	ms	2985.81	3085.21	112.92	3113.11	3065.38	3077.73
Drain-to-idle overhang avg	ms	2909.65	3047.65	57.86	3031.66	3033.61	3036.85

Parallel Throughput

Metric	Unit	Oracle	PostgreSQL	Mongo	Oracle+Redis	PostgreSQL+Redis	Mongo+Redis
Throughput	ops/s	24.17	26.28	119.51	21.88	25.51	25.14
End-to-end avg	ms	3740.84	3546.11	688.57	4147.82	3643.70	3701.72
End-to-end p95	ms	3841.33	3554.13	701.92	4243.83	3675.15	3721.14
Start avg	ms	47.44	11.32	17.89	55.82	17.07	17.04
Signal publish avg	ms	15.62	15.11	10.85	23.80	10.53	12.27
Signal to completion avg	ms	3525.84	3469.46	590.78	3872.54	3564.43	3598.14

Integrity

The comparison test required all six columns to pass these checks on both the serial-latency and parallel-throughput source artifacts:

Failures = 0
DeadLetteredSignals = 0
RuntimeConflicts = 0
StuckInstances = 0
WorkflowsStarted = OperationCount
SignalsPublished = OperationCount
SignalsProcessed = OperationCount

All six columns passed.

Interpretation

The main conclusions from this six-profile matrix are:

Native Mongo is still the fastest measured profile for the synthetic signal round-trip.
Native PostgreSQL remains the best-performing relational profile.
Oracle+Redis is slower than native Oracle in this benchmark.
PostgreSQL+Redis is very close to native PostgreSQL, but not clearly better.
Mongo+Redis is dramatically worse than native Mongo because the Redis path reintroduces the empty-wait overhang that native change streams avoid.

The most useful row for actual resume speed is Signal to first completion avg, not the mixed Signal to completion avg, because the latter still includes drain-to-idle policy.

3.3 KiB Raw Blame History