$ the-wire · showcase
Redacted credentials in etl output, bounded shutdown drain
By RepoJournal · Filed · About Supabase · Composed from the cited sources · methodology
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
12 people shipped this
bnjjj
@bnjjj
3 cited
joshenlim
@joshenlim
1 cited
farazdagi
@farazdagi
1 cited
Colum Ferry
@Coly010
supabase-evals-releaser
@supabase-evals-releaser
Barry Roodt
@barryroodt
github-actions[bot]
@github-actions[bot]
Illia Basalaiev
@Ellba
Miranda Limonczenko
@czenko
jbergstroem
@jbergstroem
Joey Lei
@leizerbeam
iambriccardo
@iambriccardo
The day's work is mostly hazard removal: credentials stop leaking into test and CLI output in supabase/etl, pipeline shutdown gets a hard deadline, and Studio's unit test suite drops from about 14 minutes to about 3.
fix: redact credentials from test and CLI output supabase/etl
Credentials could surface in destination test failures, CLI help, malformed credential input and example launch commands; raw DuckLake catalog and ClickHouse URLs no longer print, and the existing DuckLake, ClickHouse and Iceberg sanitizers are now reused in test helpers and direct SDK calls. BigQuery service-account JSON is parsed without echoing the rejected string, and Snowflake token assert...
Audit existing unit tests + Remove redundant ones supabase/supabase
A cleanup of redundant and overly specific unit tests, touching only test files plus studio-unit-tests.yml, cuts the per-PR GHA run from roughly 14 minutes to roughly 3. Coverage was the stated reason the suite had grown; the redundancy was the cost.
fix: bound pipeline shutdown and preserve safe replay supabase/etl
Pipeline shutdown could spend the entire pod termination grace period draining a large backlog; a configurable 30-second drain deadline now triggers worker cancellation and bounded DuckLake cleanup, and durable checkpoints are preserved for safe replay after restart.
build(ducklake): upgrade to stable DuckDB 2.0.0 supabase/etl
DuckLake moves to stable DuckDB 2.0.0 with matching extensions, and automatic migration is disabled so existing 1.0 catalogs keep their format while new catalogs are created as 1.1. Existing deployments are not rewritten in place.
fix(ducklake): rewrite heavily deleted small tables supabase/etl
DuckLake rewrites are now requested and selected when more than 95% of file-backed rows are deleted, including tables below the file-count threshold, with the policy shared between the replicator and the maintenance runner and the trigger reason reported. In supabase/mcp, the query splitter no longer breaks an UPDATE on a semicolon inside a dollar-quoted value, so a statement like a $t$keep it;...
