v0.0.11

Added

• deploy.<env>.port — SSH port for the deploy host (default 22)
• deploy.<env>.builder.build_port — SSH port for the build server (default 22)

Security

• SSH host keys verified trust-on-first-use against ~/.ssh/known_hosts; a known host whose key changed is rejected
• SFTP writes reject symlinked parent directories on the remote side
• .env.production written with mode 0600

v0.0.10

v0.0.10 (2026-08-30)

Fixed

• The s3 container ran SeaweedFS as weed mini, but the
  chrislusf/seaweedfs image's ENTRYPOINT already execs weed, so
  the container actually ran weed weed mini and exited with "unknown
  subcommand" — the S3 gateway never came up. The command is now mini
  (the ENTRYPOINT supplies the weed prefix), which starts the S3
  gateway on :8333 and pre-creates the bucket again.

Changed

• Bumped version constant to 0.0.10 (reflected in pier --version,
  cmd/pier/main_test.go, and the README status line).

v0.0.9

v0.0.9 (2026-08-30)

Added

• The s3 sidecar (SeaweedFS) now runs in weed mini mode, which
  starts the S3 gateway on port 8333 and pre-creates a bucket on
  startup. The bucket takes its name from S3_BUCKET, which pier
  interpolates from AWS_BUCKET (default app) so it stays in sync
  with the app.
• pier init now writes the full S3 configuration to .env (dev) and
  .env.production (prod) when s3 is in the services list:
  AWS_ENDPOINT=http://s3:8333, AWS_ACCESS_KEY_ID=somekey,
  AWS_SECRET_ACCESS_KEY=somesecret, AWS_DEFAULT_REGION=us-east-1,
  AWS_BUCKET=app, AWS_USE_PATH_STYLE_ENDPOINT=yes. Dev previously
  needed to add the AWS_* keys by hand.
• The prod compose now interpolates those AWS_* values
  (${AWS_ENDPOINT}, ${AWS_ACCESS_KEY_ID}, ...) into the app,
  queue, and scheduler containers, so they reach the app even though
  prod has no bind-mounted .env.

Fixed

• The s3 container ran SeaweedFS's default weed server, which does
  not start the S3 gateway unless -s3 is passed — so port 8333 never
  actually served S3 and the healthcheck could not pass. It now runs
  weed mini, which starts the S3 API on 8333.

Changed

• Bumped version constant to 0.0.9 (reflected in pier --version,
  cmd/pier/main_test.go, and the README status line).

v0.0.7-beta

v0.0.7-beta (2026-08-18)

Added

• Custom domains + HTTPS: the production webserver is now Caddy
  (caddy:2-alpine) with a pier-rendered docker/caddy/Caddyfile.
  A non-empty [deploy.<env>].domain enables HTTPS with automatic
  Let's Encrypt certificates; [deploy.<env>].redirect_domains
  (e.g. www) are served and redirected to the env's domain. Empty
  domain = plain HTTP by IP.
• Deploy DNS preflight: when a domain is set, pier deploy verifies
  it resolves to the deploy host before syncing and fails fast with
  an A-record hint otherwise; the health probe then checks
  https://<domain>/up end to end.
• pier init prompts for the production domain, written to
  [deploy.production].domain (blank = plain HTTP by IP).

Changed

• The deploy render phase re-writes docker/caddy/Caddyfile from
  pier.toml before syncing. Setting (or removing) a domain in
  [deploy.<env>] now takes effect on the next pier deploy; the
  file is pier-rendered, so manual edits are overwritten.
• The webserver service in the merged prod compose is now fully
  pier-managed: its ports follow pier.toml, so adding a domain
  publishes 443:443 (and removing one drops it) instead of keeping
  a stale [80:80] list that leaves Caddy unable to answer HTTPS.
• MergeEnvFile updates pier-derived .env.production keys (e.g.
  APP_URL from the env's domain) when the render changes, instead of
  preserving every existing line verbatim. User-owned keys are still
  never touched: secrets (APP_KEY, DB_PASSWORD, AWS credentials)
  and the ${...}-interpolated overrides (TRUSTED_PROXIES,
  CACHE_STORE, QUEUE_CONNECTION).
• Domains are per deploy env: [project].domain is removed and the
  old per-env extra-domains key is renamed to
  [deploy.<env>].redirect_domains. Existing configs keep loading
  (unknown keys are ignored), but envs no longer inherit a
  project-wide domain — set domain in each [deploy.<env>]
  section.
• Bumped version constant to 0.0.7-beta (reflected in
  pier --version, cmd/pier/main_test.go, and the README status
  line).

Removed

• [deploy.<env>].tls — HTTPS is now implied by domain presence.
  Existing configs keep loading (the key is ignored); delete it and
  set the domain instead.

v0.0.6-beta

v0.0.6-beta (2026-08-15)

Added

• queue_workers config ([stack] default 1, [deploy.<env>]
  override, max 32) runs that many queue:work processes in the
  queue container via supervisord numprocs, in dev and prod.
  pier init writes the explicit default.

Changed

• Bumped version constant to 0.0.6-beta (reflected in
  pier --version, cmd/pier/main_test.go, and the README status
  line).

v0.0.5-beta

Added

• pier init asks the full deploy setup — deploy host/user/path
  (branch defaulting to main) and the build machine (host_server /
  local_machine / build_server, with build host/user/path when
  build_server); --builder / --host / --user / --path /
  --build-host / --build-user / --build-path flags skip the
  prompts.
• [deploy.<env>].builder / build_host / build_user / build_path
  configuration for build server modes; pier bootstrap <env>
  provisions both machines when build_server is set.
• Real git SHA image tags (timestamp fallback) replace the hardcoded
  gitsha placeholder; docker tag wiring fixes pier rollback in
  every builder mode.
• Per-env sidecar services: [deploy.<env>].services overrides
  [stack].services for that env (absent = inherit, [] = none).
  pier init scaffolds [deploy.production] with the chosen
  services; pier service [env] replaces add/remove with a
  single init-style picker that edits dev or per-env lists.
• pier deploy <env> now re-renders docker-compose.prod.yml and
  .env.production from pier.toml before syncing (preserving
  hand-written compose edits and existing env values), so per-env
  services and [deploy.<env>].ports overrides take effect.
• Remote teardown: docker compose up runs with --remove-orphans,
  so sidecars removed from an env are stopped and removed on the
  server (named volumes are kept).

Removed

• pier buildmode <env> — the init flow now asks the build-machine
  question; change it later by editing pier.toml.

Changed

• The pier init deploy-host prompt now reads "Deploy host (SSH target
  Domain name or IP address, enter to skip)" so the expected input is
  clear.

Fixed

• The queue / scheduler app-image sidecars in image modes
  (local_machine, build_server) now reference the image tag the
  transfer phase actually ships (:current) instead of :latest:
  compose would otherwise try to pull :latest from the registry, and
  a first deploy failed with pull access denied.
• App-image sidecars now receive the same connection environment as the
  app (DB_*, REDIS_*, APP_KEY interpolated from
  .env.production): without it the worker authenticated with the dev
  .env baked into the image and crash-looped with password authentication failed.
• With redis in the stack, the rendered env files now default
  CACHE_STORE=redis and QUEUE_CONNECTION=redis (interpolated via
  ${CACHE_STORE} / ${QUEUE_CONNECTION} in the prod compose so the
  values can be overridden in .env.production; dev compose gets the
  same defaults). The database-driver defaults made queue/scheduler
  workers boot-check the cache/jobs tables — which only exist after
  after_deploy migrations — so a first deploy's up --wait could
  never pass, and a refused boot-time connection made queue:work exit
  0, which supervisord's default autorestart=unexpected never
  restarted, leaving the queue container unhealthy forever.
• The prod runtime's supervisord now always restarts the php program
  (autorestart=true), so a Laravel worker that exits 0 (queue: restart, a lost-connection stop in newer Laravel) comes back instead
  of staying dead.
• Bumped version constant to 0.0.5-beta (reflected in
  pier --version, cmd/pier/main_test.go, and the README status
  line).