The previous method did not include the service type in the user-auth-service hook, which meant it was kinda guess/check as to whether a datastore was filtered correctly for the service in question.
Additionally, we now handle newlines correctly, ensuring that when there _is_ filtering, we treat each datastore as a distinct one.
If a user implements the user-auth-service trigger in a plugin and that plugin does not echo the passed in app(s) on stdout, the app is assumed to not exist. This mirrors the functionality for applications in regards to auth filtering.
This may still need auditing to ensure it covers everything and doesn't cause issues, but local testing implies that everything is working as expected.
Previously they would just 'disappear', resulting in broken applications until the app was relinked to the service in question.
Closesdokku/dokku-redis#142
This will help ensure that users upgrading to a new plugin version who stop/start databases will always get the same version. This is particularly important for datastores such as elasticsearch and postgres that have more involved upgraded processes.
This is a bit more clear as the links aren't always idempotent and may have errors - this allows users to reconcile at what point a link may have failed so that they can take further action.
If the ID file contained an incorrect value, calling :start would say the service is started but :info would still show the container as missing.
Also fix an issue where docker inspect leaked stderr when the container was missing.
Refs dokku/dokku-redis#133
This PR allows dashes in service names, while still sanitizing them before they are used as database names. If the datastore is pre-existing, the datatabase name is assumed to be the same as the service name, and returned appropriately.
This pull request switches testing to use an actual docker daemon, vs mocking everything out.
It may also catch actual breaking issues in our tests, which is great!
We previously allowed a wide range of service names. As the service name is sometimes used to name databases, the name was actually more restricted than any character, resulting in services that wouldn't start. Going forward, only alphanumeric and underscore characters are allowed.
This only impacts service creation. Any services with invalid names should be migrated to a new service, with the data exported and imported as normal.
Closesdokku/dokku-redis#99Closesdokku/dokku-mysql#47Closesdokku/dokku-mongo#86Closesdokku/dokku-redis#81
This should actually be refactored to avoid the grep call completely, but the current fix will correct the issue for now.
Refs dokku/dokku-postgres#131
