Articles

ClickHouse acknowledges an ALTER ... UPDATE before the data is rewritten. ch-migrate 0.7 waits for the mutation and records the migration once it's done.

Per-laptop database logins for an AI agent don’t scale. How I replaced them with one hosted MCP server in production, with permissions outside the model.

ClickHouse doesn't roll back a failed migration, so every statement must be safe to run twice. ch-migrate 0.6 checks that first, and makes SET apply.

An LLM labeling service on Gemini and Claude, versioned training data, shadow serving and a promotion gate: the infrastructure behind a production vision model.

ClickHouse migrations in plain SQL: ch-migrate 0.5 drops the Python glue each schema change used to need. Write an upgrade file and a downgrade file.

Identity Center consumes your assertion, then the connection drops before the token arrives. How to handle that failure, and the API AWS should add.

25 coding-agent sessions on a 64 GB workstation hit a scaling problem, not a RAM shortage. How I shared MCP and model servers, workflow unchanged.

IoT device lifecycle management breaks when history follows the hardware. Why I skipped the obvious cleanup, which would have misrouted about 900 live devices.

A ClickHouse backfill that writes into the live table has no rollback. How I rebuilt one around resumable extraction, a shadow load and an atomic swap.

When ten agent sessions share one VPN identity, nothing errors. Requests just vanish like a network blip. How to give each session its own identity.

“Too many parts” in ClickHouse usually comes down to batching. A worked example: a pipeline change that removed it, and a fix that keeps retries safe.

A materialized view cutover cut per-fire memory from 140 MiB to 7.4 MiB. Resident memory didn't move. The SQL pattern, and the metric that predicted OOMs.

You wrote a Claude Code skill yesterday. Three teammates want it by Friday. Here's how to package skills as plugins and roll them out by version.

Every AI dev tool ships with the same demo. The hard question is what compounds. Why I picked Claude Code over Codex, and the surface that earned it.

Railway gauge isn't really two Roman horses wide. The true part of the myth, path dependence, is why the right fix is often to build around an old choice.

AI removed the bottleneck on shipping code, not on understanding it. That gap causes a new class of failures, and a 40-year-old paper predicted why.

Lambda is simple until you hit its sharp edges: cold starts, memory-CPU coupling, timeout chains and the 75 GB storage cliff. The fix for each.

If transformers mean chatbots to you, you're missing the concept. What attention does, and why its infrastructure problems are ones you already know.

ch-migrate is an open-source ClickHouse migration tool built on Alembic: versioned schema changes for every environment, with bootstrap and rollbacks.

Runbooks go stale and nobody reads them. Encode your Terraform, migration and deploy procedures as Claude Code skills the agent follows step by step.

45-second dashboard queries during winter storms. How I moved 1.3 billion IoT rows from SQL Server to ClickHouse Cloud and got them under 2 seconds.

RDS Proxy defaults: IdleClientTimeout 30 minutes, ConnectionBorrowTimeout 120 seconds. When to change each for Lambda, and how pinning caps concurrency.

Install Python packages from AWS CodeArtifact with uv: the keyrings.codeartifact plugin, keyring-provider set to subprocess, and one variable.

The platform crawled during winter storms, just when snow removal crews needed it. How I led a three-person team to fix it without leaving SQL Server.