New Deploy section 'Supporting services live in your compose, not the
fleet': Redis is declared per-repo, never shared fleet-wide — a shared
instance would re-couple services the per-project network deliberately
isolates.
- ephemeral vs persistent as a conscious choice, both compose snippets in
content tabs (--save "" + no volume, vs --appendonly yes + ${MOUNTS_DIR})
- connect by compose service name (redis://redis:6379), never a host port
- caveats given weight: appendfsync everysec ~1s loss window, and mounts
are not backed up — the dividing line between queue state and Postgres
- SQLite noted as the no-container alternative for a single drainer
- the don'ts, plus one line on why ACLs aren't the answer here
- cross-links the redis lib on the Libraries page
Signed-off-by: disqualifier <dev@disqualifier.me>
dsql
a248879451