Soltheris
News and analysis from the world of production systems.
Practical Notes on Postgres Connection Pooling
May 28, 2026
Handling PostgreSQL backends imposes substantial overhead because each active client spawns a dedicated process consuming significant RAM, making intermediary connection management mandatory at scale. While session-level pooling preserves compatibility seamlessly, transaction pooling aggressively reassigns connections between queries, breaking features like LISTEN channels, advisory locks, and runtime configuration parameters.
Operational headaches with connection proxies usually stem from poor connection hygiene: forgotten checkouts left dangling in application code, cursors maintained across idle sleep loops, and large migrations hogging pooled slots. Enforcing transaction pooling surfaces these bad patterns immediately by terminating unsupported stateful operations.
HTTP/3 and QUIC: What Changed for Operators
September 7, 2026
Implementing transport logic inside user space alongside encrypted protocol framing gives QUIC superior privacy while frustrating network operators relying on packet sniffers. Diagnostic data such as congestion state and window allocations are obscured behind cryptographic boundaries.…
Reading Latency Percentiles Without Fooling Yourself
April 25, 2026
Arithmetic averages hide the extreme tail latencies that percentiles make evident. Even if an endpoint posts an average duration of fifty milliseconds, one out of twenty calls might experience a grueling two-second delay; customers subjected to cold paths and overloaded database shards are the ones …
Zero-Downtime Deployments Without the Drama
August 20, 2026
Narratives surrounding uninterrupted deployments obsess over flashy infrastructure switching and instant green-blue cutovers. Real-world stability relies on mundane mechanics: deep readiness endpoints that confirm functional dependencies, complete socket draining, and backward-compatible persistence…
Why Edge Caching Still Matters in 2026
July 21, 2026
Every few years someone declares the edge cache obsolete: bandwidth is cheap, compute is fast, so why bother? Yet p99 latency keeps telling a different story. The round trip from a user in Sao Paulo to an origin in Frankfurt costs around 200 ms on a good day, and no amount of application optimisatio…
More reading
- Data Residency Basics for Global Teams — Compliance, August 31, 2026
- Managing Secrets Without Losing Sleep — Security, August 2, 2026
- When to Choose a Queue Over a Request — Engineering, June 11, 2026
About us
We cover the unglamorous middle of software: queues that back up, caches that lie, and DNS at 3 a.m. Everything is written from real operational experience.