Marianne R.

Hiring a QA engineer who can build a regression suite from scratch is fundamentally different from hiring someone to run an existing one. The first is a systems thinker - someone who looks at your app, maps the critical paths, picks the right framework, and builds coverage that actually catches regressions before prod. The second is an executor - valuable, but a different role entirely. When you're early-stage, you need the builder. When you're scaling, you need both. #QA #Hiring #Engineering

500

42 Comments

Most people use Claude Code wrong. 🤖 They treat it like ChatGPT with a terminal. But the real power is in context control. If you’re not configuring your CLAUDE.md properly, you’re wasting tokens 💸 and getting generic output. Your CLAUDE.md should clearly define: • 🏗 Project architecture • 🛠 Tech stack • 📁 Folder structure • 📐 Coding standards • 🏷 Naming conventions • 📜 “How we build here” rules Think of it as onboarding Claude like a senior developer joining your team 👨‍💻 The better the context, the smarter the output. AI coding tools aren’t magic ✨ They’re leverage. And leverage depends on configuration. How are you structuring your CLAUDE.md? 👇 #AI #Claude #DeveloperTools #BuildInPublic #StartupLife #SaaS #TechFounder #Productivity #Engineering

10

1 Comment

🚀 Grafana Mimir vs VictoriaMetrics — Architecture, Performance & Real-World Trade-offs When evaluating a time-series database for Kubernetes observability, the conversation often revolves around scale numbers: 👉 Billion series 👉 Millions of samples/sec 👉 Massive cluster sizes But the real question is: How do these systems behave under realistic production workloads on comparable infrastructure? I recently published a deep dive comparing Grafana Mimir and VictoriaMetrics Cluster under a 24-hour benchmark on GKE with identical hardware and ingestion patterns. Here’s what stood out 👇 🔍 Workload Profile • ~5.5M active time series • 360K samples/sec ingestion • 31B samples over 24h • Continuous churn simulation • Instant alert-style queries Both systems remained stable and handled ingestion without errors. But efficiency told a different story. ⚙️ Key Findings 📉 CPU Usage VictoriaMetrics used ~1.7× less CPU under the same workload. 🧠 Memory Usage The biggest differentiator. VictoriaMetrics used ~5× less memory. Mimir’s ingesters became memory-heavy under churn. 💾 Storage Footprint (24h) VictoriaMetrics consumed significantly less storage compared to Mimir’s combined local + object storage usage. ⏱️ Query Latency • p50: Similar • p99: Lower tail latency in VictoriaMetrics during this test Under 5× increased load, Mimir hit memory limits while VictoriaMetrics saturated CPU but remained operational — highlighting very different architectural bottlenecks. 🏗 Architectural Philosophy Matters Grafana Mimir • Modular microservices (Distributor, Ingester, Querier, Store Gateway, Compactor) • Strong object storage integration • Designed for multi-tenant, extreme-scale environments • Higher memory footprint due to in-memory ingestion model VictoriaMetrics • Leaner cluster model (vminsert, vmstorage, vmselect) • Disk-first storage approach • Fewer moving parts • Optimized for resource efficiency Both are production-grade systems. But they optimize for different priorities. 🎯 What This Means for Platform Teams If your priority is: ✅ Cost efficiency ✅ Lower memory footprint ✅ Simpler operations VictoriaMetrics looks very compelling. If your priority is: ✅ Massive multi-tenant scale ✅ Cloud object storage integration ✅ Modular distributed design Mimir offers strong enterprise capabilities. The biggest takeaway? There is no universal winner. Architecture decisions shape performance characteristics. Your workload pattern should drive your choice. If you’re running large-scale observability on Kubernetes, I’d strongly recommend benchmarking against your own ingestion + query profile before deciding. 📖 Full deep-dive blog: https://lnkd.in/gc3yuFCP #Kubernetes #Observability #SRE #Prometheus #CloudNative #Grafana #VictoriaMetrics #PlatformEngineering

12

One of the riskiest moves teams make... Adopting Terraform AFTER infrastructure already exists. I imported a live production AWS environment into Terraform with: Zero downtime Zero resource changes Full state reconciliation ⚙️ Why this matters: IaC migrations fail more often than greenfield builds Drift and forced rebuilds are common Done right, Terraform becomes boring — and that’s a compliment. #CloudComputing #AWS #TechCareers #CloudEngineer #Terraform

2