v0.3.0-rc.22

The twenty-second pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.22].

What an operator must do

• Every deployment: run paisans deployment apply. It rewrites each site's haproxy.cfg so the postgres health check reads Patroni's whole answer, which stops Patroni logging a Connection reset by peer traceback for checks that passed. No data or topology changes.
• Nothing to do for the heartbeat change: a push that recovers on retry no longer logs curl's errors.

Everything in [0.3.0-rc.1] to [0.3.0-rc.21] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #597
• Hold heartbeat curl errors until a push's retries are spent by @josephquigley in #598
• The postgres HAProxy check reads Patroni's whole answer by @josephquigley in #599
• Release 0.3.0-rc.22 by @josephquigley in #600

Full Changelog: v0.3.0-rc.21...v0.3.0-rc.22

v0.3.0-rc.21

The twenty-first pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.21].

What an operator must do

• An apply that stopped on a container restarting: run it again. The health gate now waits on a container Docker has restarted once or twice, and fails it at the third restart or when its five minute wait ends.
• A deployment with a pinned synapse, wikijs or writefreely: apply it. Core catalog v0.10.1 gives their postgres a healthcheck, and their services wait for it to answer, on up and on restart. The apply recreates those stacks once.
• Nothing to do for the rest. A finished one shot, such as the beszel hub's admin, no longer fails doctor or the check after an apply.

Everything in [0.3.0-rc.1] to [0.3.0-rc.20] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #592
• The health gate waits on a container restarted once or twice by @josephquigley in #593
• Doctor counts a finished one shot as done, not down by @josephquigley in #594
• Release 0.3.0-rc.21 by @josephquigley in #595

Full Changelog: v0.3.0-rc.20...v0.3.0-rc.21

v0.3.0-rc.20

The twentieth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.20].

What an operator must do

• An ops-notifier app: remove its hostname. The blueprint is now hostname: none: it serves no website, gets no DNS record, certificate or gateway route, and is checked by uptime over the mesh. validate refuses a hostname on it. Run paisans dns prune to remove a record an earlier apply made for it.
• A deployment with an engagement app: apply it. The digest's read only role is granted "user".application_status, entry_vote.created_at and post_vote.created_at, and its env carries the welcome settings. Welcoming new members also needs a core catalog whose mbin provides the forum role (paisans-blueprints #41) and an engagement service release that reads it; until then the forum link is empty and nothing changes for members.
• Catalog authors: blueprint format paisans.blueprint/v5 adds hostname: none. Toolkits before this release skip a catalog release using it.
• Nothing to do for the rest.

Everything in [0.3.0-rc.1] to [0.3.0-rc.19] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #588
• feat: the engagement blueprint renders what welcoming newcomers needs by @josephquigley in #585
• feat(blueprints): hostname: none for an app that serves no website by @josephquigley in #589
• Release 0.3.0-rc.20 by @josephquigley in #590

Full Changelog: v0.3.0-rc.19...v0.3.0-rc.20

v0.3.0-rc.19

The nineteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.19].

What an operator must do

• A deployment with a beszel app: apply it so the hub picks up the smtp block and its sender. validate now refuses a from_address that is not a bare email address (beszel-smtp-sender), and warns on starttls with no username (beszel-starttls-not-enforced). Beszel then emails the addresses each operator adds in its settings; it seeds no alert rules.
• Beszel alerts in Matrix are opt-in. To add them:
  • Declare an app with blueprint: ops-notifier and a hostname, pinned to the site the synapse app runs on. validate refuses it anywhere else (colocated-app-elsewhere), and without a homeserver or a beszel app (ops-notifier-without-homeserver, ops-notifier-without-hub).
  • Apply. The synapse stack is applied again with an admin client for the notifier, and the hub with a superuser of the notifier's own.
  • Its image track comes from the core catalog v0.10.0 (paisans-blueprints), which earlier toolkits skip, staying on v0.9.0.
• Catalog authors: blueprint format paisans.blueprint/v4 adds colocate: true on a consume. Toolkits before this release refuse a v4 blueprint, so a catalog release using it is skipped by them.
• Nothing to do for the rest.

Everything in [0.3.0-rc.1] to [0.3.0-rc.18] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #578
• feat(beszel): send alert email through the deployment's smtp block by @josephquigley in #579
• feat(blueprints): let a consume be colocated with its provider by @josephquigley in #581
• docs: specify ops-notifier, Beszel alerts in Matrix by @josephquigley in #582
• feat: add the ops-notifier blueprint, Beszel alerts in Matrix by @josephquigley in #583
• docs: design for welcoming new members by @josephquigley in #584
• Release 0.3.0-rc.19 by @josephquigley in #586

Full Changelog: v0.3.0-rc.18...v0.3.0-rc.19

v0.3.0-rc.18

The eighteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.18].

What an operator must do

• Nothing for an existing deployment. A site holding monitor may now run Beszel's hub with no uptime app beside it, Eg: a gateway given roles: [gateway, monitor] while the first monitor's external web server holds the uptime app alone. validate warns monitor shares a site there, as for any shared monitor.
• To use it, follow rc.17's steps for adding Beszel, pinning the beszel app to that site.

Everything in [0.3.0-rc.1] to [0.3.0-rc.17] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #574
• feat(validate): let a monitor site run Beszel's hub alone by @josephquigley in #575
• Release 0.3.0-rc.18 by @josephquigley in #576

Full Changelog: v0.3.0-rc.17...v0.3.0-rc.18

v0.3.0-rc.17

The seventeenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.17].

What an operator must do

• Server and container stats (Beszel) are opt-in. To add them:
  • Declare an app with blueprint: beszel pinned to the monitor site. paisans create offers it after a yes to the health page.
  • Run paisans site prepare on every site prepared by an earlier release, so the agent finds the deployment user's Docker socket.
  • Then apply.
  • The image tracks come from the core catalog v0.9.0 (paisans-blueprints), which earlier toolkits skip, staying on v0.8.5.
• paisans prune --site <site> is now paisans site prune <site>. It also removes the containers of services a stack no longer declares, keeping etcd, Patroni and Garage. paisans deployment prune runs it on every site.
• Nothing to do for the rest. deployment apply now lists those containers under left over, and removes nothing.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #571
• feat: server and container stats with Beszel by @josephquigley in #565
• site prune and deployment prune remove what a site no longer runs by @josephquigley in #567
• Release 0.3.0-rc.17 by @josephquigley in #572

Full Changelog: v0.3.0-rc.16...v0.3.0-rc.17

v0.3.0-rc.16

The sixteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.16].

What an operator must do

• Apply the deployment to pick up the digest's new name and the engagement service 0.2 image the core catalog now tracks (paisans-blueprints v0.8.5). The digest then names the community by community.name, e.g. "This week on Paisans", and Mbin keeps its own titles.
• Nothing to do for the rest. The engagement app's front page and /d/ with no date now redirect to the account app, or the community's domain without one.

🤖 Generated with Claude Code

What's Changed

• develop is 0.3.0-dev by @josephquigley in #564
• feat: send a visitor with no digest to read to the account app by @josephquigley in #566
• Name the community in the digest by community.name by @josephquigley in #568
• Release 0.3.0-rc.16 by @josephquigley in #569

Full Changelog: v0.3.0-rc.15...v0.3.0-rc.16

v0.3.0-rc.15

The fifteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.15].

What an operator must do

• Upgrade every workstation that changes this deployment. Every command that changes a site now refuses one last applied by a newer paisans, so once any operator applies with rc.15, an older paisans is refused on those sites until it upgrades. site admit and site evict refuse a host a newer paisans claimed in the same way. A build of develop still passes on a host a release candidate set up.
• Expect the deployment lease on more commands. With --execute, these commands now hold the lease and are refused while another command holds it: site add, site prepare, site remove, app remove, gateway move, mesh rekey, prune, storage init, storage add, storage rotate-key, secrets rotate, oidc client create and operator register. Each takes --break-lease, and --yes where it had none.
• Engagement app linked to a Matrix homeserver: set gated: true when you deploy paisans-engagement-service#15 or later, so the bot's "Read" links keep working. validate warns until it is set.
• Engagement app with bot_localpart left unset: the next apply moves the bot from @digest-bot to @community-bot, and the service starts new rooms from the new account. Set bot_localpart: digest-bot to keep the old account. The bot's display name is now bot_display_name, Community Bot by default.
• Nothing to do for the rest. deployment apply without --site leaves alone a site with nothing to do. site admit sets up relay forwarding only on the relay, and site evict removes it. dns prune removes a sender DMARC policy that nothing sends mail from any more.

Everything in [0.3.0-rc.1] to [0.3.0-rc.14] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #551
• fix: a relay dialled on its public_address forwards, and ufw lets it relay the mesh by @josephquigley in #552
• fix: forward only on the relay, so apply stops refusing other dialled sites by @josephquigley in #553
• fix: a build of develop meets requires_toolkit of its own pre-releases by @josephquigley in #554
• feat: name the engagement bot community-bot and let a deployment name it by @josephquigley in #555
• fix: dns prune removes a sender DMARC policy nothing sends mail from by @josephquigley in #557
• feat: deployment apply leaves alone a site with nothing to do by @josephquigley in #556
• feat: gate the engagement app's web digest, leaving unsubscribe open by @josephquigley in #559
• feat: commands that change a site refuse sites a newer paisans last applied by @josephquigley in #558
• feat: every command that changes sites holds the deployment lease by @josephquigley in #560
• fix: run the gate before an app's handles on a gated hostname by @josephquigley in #561
• Release 0.3.0-rc.15 by @josephquigley in #562

Full Changelog: v0.3.0-rc.14...v0.3.0-rc.15

v0.3.0-rc.14

The fourteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.14].

What an operator must do

• Run validate before the next apply. It now refuses an optional link with several providers and no setting, a link sharing a secret with a provider outside the deployment, and an engagement bot_localpart that is not a Matrix localpart; fix whichever it names.
• Run deployment apply on a deployment with an engagement app and a matrix-homeserver. The engagement app then runs the Digest Bot, synapse's Matrix Authentication Service gets an admin client for it on a mesh-only listener, and the gateway forwards the web digests at /d/*. The synapse side is in core catalog 0.8.3, which requires this release; an older toolkit keeps 0.8.2.
• Nothing to do for the rest: the next deployment apply keeps DOCKER_HOST in the deployment user's ~/.bashrc and links ~/deployment to /srv/paisans/<token>, and secrets rotate hands a shared secret to its provider before the consumer.

Everything in [0.3.0-rc.1] to [0.3.0-rc.13] still applies.

🤖 Generated with Claude Code

What's Changed

• Develop is 0.3.0-dev by @josephquigley in #547
• feat: keep DOCKER_HOST and ~/deployment in the user's shell by @josephquigley in #548
• Blueprints link optionally and share secrets; the engagement app runs the Digest Bot on Matrix by @josephquigley in #543
• Release 0.3.0-rc.14 by @josephquigley in #549

Full Changelog: v0.3.0-rc.13...v0.3.0-rc.14

v0.3.0-rc.13

The thirteenth pre-release of 0.3.0, published as a GitHub pre-release and not marked latest. See CHANGELOG.md, [0.3.0-rc.13].

What an operator must do

• Upgrade every operator's toolkit on a deployment together. deployment apply and deployment publish now take a lease on every site, but an older build neither takes nor honours it, so a run from one can still overlap a run from this one.
• Nothing to apply for the rest: site ssh <site> is new, and deployment log and doctor now say what each site last applied.

Everything in [0.3.0-rc.1] to [0.3.0-rc.12] still applies.

🤖 Generated with Claude Code

What's Changed

• develop is 0.3.0-dev after rc.12 by @josephquigley in #540
• feat: say what each site applied, apart from what is published by @josephquigley in #541
• feat: hold a deployment lease while apply or publish runs by @josephquigley in #542
• feat: add site ssh, a session on a site as its ssh user by @josephquigley in #544
• Release 0.3.0-rc.13 by @josephquigley in #545

Full Changelog: v0.3.0-rc.12...v0.3.0-rc.13