Long-form notes on software, distributed systems, and the craft of building. Shipping one a week.
11 posts
A 0-click query for `dbos ontime` showed up in my Search Console last week. The reader is not asking about DBOS — they are asking how to run a job every minute, exactly once, across a fleet. From my own notes, an etcd lease, the `concurrency.Election` package, and a fencing token cover that case in under 100 lines of Go, without pulling in a workflow engine.
When the log is the source of truth, every schema change is permanent. A Kotlin/Avro walkthrough of the rename that passed the Schema Registry check and silently corrupted every old event, plus the Protobuf and Avro invariants I now keep pinned above my desk.
Notes from two weekends of digging into DSPy. I stopped treating prompts as the source of truth and started treating them as compiled output from a typed signature, a metric, and an optimizer. Here is the smallest end-to-end program I kept, how MIPROv2 actually searches, and where the approach breaks down in practice.
Spring AI's server-side A2A integration is stable enough to put in production, but the protocol is most useful at organizational boundaries, not as an internal RPC replacement. This post walks through what actually changes in a Spring AI codebase, where the sharp edges still are, and a practical decision framework for A2A vs MCP vs plain REST.
DBOS reuses Postgres as the durability layer for workflows, while Temporal runs a dedicated cluster. The right choice depends on team size, workload shape, and where you want your operational budget to go. This is a practical rubric for picking between them.
Cell-based architecture contains blast radius, but it is not free. A look at what Slack, DoorDash, and Roblox actually paid for cells in production — and a checklist for the cheaper fault-isolation patterns most teams should reach for first.
JetBrains Tracy is a Kotlin library that wires LLM-aware tracing into your app on top of OpenTelemetry. This post walks through how I integrated it in a Spring Boot service, the design decisions that matter, and the failure modes teams hit once LLM calls become the hottest path in their system.
Fully autonomous agents are hard to bound, hard to test, and expensive to operate. A deterministic backbone with narrow agent steps gives you the control flow back while keeping the intelligence where it matters. Here is how to design, test, and migrate toward it.
Most AI memory benchmarks grade on recall and stop there. That hides the real failure mode: stale facts quietly poisoning the context window. Here is a lifecycle-based evaluation framework that tests recall, revision, and controlled forgetting across the change points every long-lived project goes through.
The transactional outbox is a ledger, not a queue. Treating it like one is what breaks Postgres under load. This post walks through the specific failure modes — autovacuum stalls, xmin horizon drift, replication slot lag, poison pills — and the operational rules that actually keep it working in production.