Why Did systemd OOM-Kill My Service (and How to Stop It)
Your service vanished with no crash log because the kernel OOM killer took it. Confirm it from journalctl and dmesg, read the real memory pressure, then bound it with cgroup limits.
Pillar: Pillar 14 - Scalability & Performance Headroom
Pillar 14 · Evolve
sysctl Tuning That Earns Its Keep: Network and Kernel Settings for a Busy Server
Most sysctl optimisation lists are cargo cult. Here is a defensible /etc/sysctl.d drop-in for a busy Linux server, with why each setting exists and how to confirm it took effect.
Spot GPUs Without Losing Work: Preemption-Safe Training and Batch Inference
Preemptible GPUs cut training and batch-inference cost by 60-90%, but only if jobs checkpoint and resume cleanly. How to make GPU workloads preemption-safe.
Instant Payments Fraud Controls: Real-Time Risk Without Breaking the 10-Second SLA
Instant means irrevocable, so fraud scoring must run inside the ten-second settlement window. A reference architecture for real-time risk that fits the latency budget.
Verification of Payee at Scale: Architecture for Sub-Five-Second Name Matching
Verification of Payee has to answer inside the instant-payments clock, which makes name matching an architecture problem. A reference design for the matching, caching and directory-lookup path.
Charging Parity and Reachability for SEPA Instant: Re-Pricing and Capacity Planning
The Instant Payments Regulation bans charging more for instant than standard credit transfers and demands full reachability. Both are pricing and capacity decisions, not compliance line items.
Cutting CI Time: Dependency Caching in GitHub Actions That Actually Hits
Most GitHub Actions caches never hit. Here is how to key them on lockfile hashes with restore-keys fallbacks, cache uv and Docker layers, and prove the cache is working.
Automating Allocations and Confirmations for T+1: The Latency Budget
T+1 removes the overnight window manual allocation and confirmation flows depend on. Automate same-day affirmation and measure every post-trade hop against the shortened clock.
Designing Filter, Sort and Field-Selection Query Params Without Reinventing SQL
A safe, bounded grammar for filter, sort and field-selection query params that maps to indexed columns only — injection-proof, in FastAPI and SQLAlchemy 2.
Why Did My GraphQL Query Fire 500 SQL Statements? Killing N+1 With DataLoader
One nested author field turned a single GraphQL query into 501 database hits. How to spot the N+1 in logs and APM, then batch it away with DataLoader in Apollo Server 4.