ClickHouse's ops-heavy week meets dbt's parser transition

Tinybird's Classic deadline arrives

The long-announced Tinybird Classic sunset landed this week: remaining Free Workspaces are deleted outright, Developer Workspaces lose UI access and can only be migrated through support, and Enterprise plans are untouched. If you're still on Classic, the window for self-service migration closed. The same changelog notes two operational traps for Forward users — the CLI now demands Python 3.11 or later, and Tinybird Local's arm64 image refuses to start on hosts with 16 KiB or 64 KiB kernel pages, a class of machine that includes some recent Apple silicon and ARM server setups.

dbt moves its defaults forward

The week's biggest developer-facing shift is in dbt-core's 1.13 beta, which flips every flag introduced in 1.9 and 1.10 to default true — a breaking change for projects that never opted in. It also bundles the opt-in dbt-state plugin, adds a dbt login command for platform authentication, and introduces a --use-v2-parser flag that hands parsing to Fusion, with parser fixes for duplicate-metric partial parsing and a circular-selector recursion crash on Windows. The 1.12 and 1.11 maintenance lines picked up the smaller pieces of that work: wildcard FQN selectors match nested models regardless of subfolder depth, the v2 parser's logs now relay cleanly through --log-format otel (fixing a latent bug where a stray stderr line could abort a --warn-error run), and package tarball downloads are validated to stay inside the dbt downloads directory. On the v2 side, catalog.json comes back for dbt build --write-catalog after a preview regression silently skipped it, spurious TableNotFound errors during analyze are gone, and the PyPI wheels now support Python 3.10.

ClickHouse's quiet background hum

ClickHouse's 25 releases across four channel lines this week are routine patch churn — the stable and LTS branches advancing in lockstep, with nothing flagged as a headline change. Worth knowing they exist if you pin versions, but nothing here should change how you build. The substance came from a native TCP client for .NET that promises faster reads and inserts with fewer allocations, columnar APIs, POCO support, and a migration path off HTTP. Alongside it, ClickStack added built-in z-score anomaly detection on existing charts plus a sharded scheduler sized for tens of thousands of alerts migrated from tools like Grafana, and the team published a head-to-head CostBench result against Redshift Serverless. Two posts drill into the plumbing: how Managed Postgres and ClickHouse work as one stack, with a live WalShadow demo replicating Postgres in under a second, and a careful hunt for a 32 ms tail-latency spike traced to unbatched Native protocol metadata packets overflowing a reset congestion window — a good read if you've ever blamed the query planner for network physics.