docs.coopcloud.tech/docs/abra/upgrade.md

title

title
Upgrade

Release channels

Stable

abra upgrade

Release candidate

abra upgrade --rc

Migration guides

General release notes are here

0.7.x-beta -> 0.8.x-beta

• We now have an --offline flag instead of relying on internal logic to try to decide when offline/online works best. It's up to you! A lot of abra operations require network access, so it is not really truly "offline". The logic prefers local filesystem access when this flag is passed. E.g. if there is a local copy of the catalogue, then don't git pull.

• There is more --chaos! There is more consistent flag handling for manually overriding when to update the local recipe or leave it alone when hacking on changes.

• Secrets are now only generated by reading the recipe config, not the env vars. This should hopefully not affect you. If you're seeing weird behaviour, please see #464.

• There is a new linting rule for catching invalid tags in recipe versions. This is an seemingly unavoidable issue that requires some maintenance work. If you run into the error, here's some docs to help work through it.

• ~/.abra/catalogue is now only updated via git pull. You may need to cd ~/.abra/catalogue && git checkout . to get abra to stop complaining about unstaged changes.

• abra record ... & abra server new have been removed! Following a usage poll, these features were not being relied on. They were also alpha prototypes which we feel can be reconsidered once other more critical parts of Abra are more stable.

0.6.x-beta -> 0.7.x-beta

• ALERTA, ALERTA, security related issue: all $domain.env env vars are now exposed to the deployment via the app service container. Each FOO=BAR is exported within the context of the container. If you have any privately committed secrets in your .env files, please migrate them to the secrets: ... configuration in the recipe. This change was made to facilitate tooling which can support auto-upgrading of apps in a deployment.

• abra can no longer install Docker, initialise swarm mode and the proxy network. It will check if a Docker install exists and is in swarm mode or not and error out accordingly. We leave the provisioning to tools that are designed for that and reduce the command-line surface that we have to maintain going forward.

• abra server add <host> <args> 👉 abra server add <host>. We have finally removed the custom SSH handling code and now solely rely on invoking /usr/bin/ssh directly and reading from the ~/.ssh/config. The <host> argument should correspond to a Host <host> entry in your ~/.ssh/config or in an Include <file> statement (hosts are retrieved via ssh -G <host>). This means "how does abra interact with SSH is 1) do you have an ~/.ssh/config entry for <host> 2) can you ssh <host> successfully? 3) there is no 3. It's an easier mental model and also the way abra-bash works, hence, less weird obscure errors. <host> being public a domain name is still required.

• abra no longer tries to do the TOFU host key verification prompt. We follow the praxis of the Docker CLI and just give up when host keys are not validated. We leave it to folks to SSH in and verify themselves.

• Digests have been removed from the catalogue generation. They are not being used elsewhere and were significantly slowing down generation.

0.5.x-beta -> 0.6.x-beta

• Using {{ .Domain }} in recipe .envrc.sample files went away because it was portable enough. We revert to replacing e.g gitea.example.com with the domain. See 8fad34e for more.

• If your abra.sh scripts depend on /bin/sh and /bin/bash is available in the container then /bin/bash will be used from now on. /bin/sh is only now used if /bin/bash is not available. See 7f745ff for more.

v0.4.x -> v0.5.x

• The only breaking change was making abra understand that the JSON dump for the recipes listing lives on recipes.coopcloud.tech/recipes.json instead of recipes.coopcloud.tech.

v0.3.x -> v0.4.x

Make sure to back up your ~/.abra/servers configurations first for safety.

• Please run mv ~/.abra/apps ~/.abra/recipes.

• "app name" as a concept went away, abra now uses the domain name of an app as the identifier. However, we don't expect to see breaking behaviour if you have .env files like ~/.abra/servers/foo.com/mycoolapp.env and you still want to run abra app ps mycoolapp. abra still reads the filename to figure out the identifier. When running abra app new <recipe>, abra will now take the domain name as the name of the .env file.

• abra has a new SSH implementation which enforces SSH host key checking. You may run into connection issues as a result of this code churn, please see this entry for help navigating a fix.

• CLI flag/args handling has been made more flexible. We're working within the constraints of an upstream library issue but have hopefully made it easier to mange passing flags to commands with abra. See this troubleshooting entry for the full review.

• A number of short style flags have been re-mapped and/or added. This is again related to an issue with the upstream CLI library which sometimes understands short style flags as long style flags. E.g. --ch instead of -ch for --chaos. As a concrete example, --ch is now -C on abra app deploy.

• abra app backup & abra app restore are ready for alpha-testing! See this entry and that entry for more.

• abra server add --traefik went away, it was too unreliable & hard to maintain.