- the route settings code was factored out into its own service, but the tests were not moved
- this moves them so it's clearer what tests are for what service
refs https://github.com/TryGhost/Team/issues/1497
- This enables us to keep track of which newsletter an email was sent to even if the related post is deleted.
- Adds the `newsletter_id` property to the email API
refs https://github.com/TryGhost/Team/issues/1500
The newsletter table schema has bunch of changes to go through for new and existing columns, this consolidates the schema changes into a single re-create table migration that drops and adds the newsletter table with correct schema. The table re-create migration needs to run before any of the tables using newsletter as foreign key. The changes include -
- new columns for design related fields
- new slug column for filtering
- unique constraint to `name` column
- remove `default` column (noops the existing default column migration)
- `sender_reply_to` has a default of newsletter and a validation of ['newsletter', 'support']
- updated default values for `subscribe_on_signup` and `recipient_filter`
* Added subscription_id to MRR events
refs https://github.com/TryGhost/Team/issues/1453
As part of 5.0 we want to have cancelled evetns affect MRR. We are going
to backfill and begin populating cancelled events _before_ 5.0 however,
adding a migration to set the MRR deltas as part of 5.0
This migration will need to associate the cancelled events with a
subscription, so we are adding the subscription_id now. This will allow
us to guarantee that all cancelled events will have a subscription_id.
The column is missing a NOT NULL and REFERENCES constraint because
we have not populated the missing values. These will be added in the future
once we have populated the column for all rows
refs https://github.com/TryGhost/Team/issues/1248
This is the underlying cause of the problems we've seen whilst handling
Stripe webhooks. A transaction ensures that the operations are atomic,
but not that they can run concurrently.
If you have some code which does this, concurrently:
1. Starts a transaction
2. Reads a value
3. Changes the values
4. Ends the transaction
Without applying the `FOR UPDATE` lock - you will have both sequenes
read the same value at step 2.
With the `FOR UPDATE` lock - one of the sequences will hang at step 2,
waiting for the other transaction to end, at which point it will resume
and read the _changed_ value.
Because the `edit` method explicitly does a read followed by a write, we
have also add the `FOR UPDATE` lock to this by default, to avoid any race
conditions. This does however require that `edit` is called within a
transaction. An issue here https://github.com/TryGhost/Team/issues/1503
considers running in a transaction by default.
refs https://github.com/TryGhost/Toolbox/issues/280
- The Accept-Version/Content-Version header handling is shared between all APIs and should not structurally belong to any of the admin/content/members folders
refs https://github.com/TryGhost/Toolbox/issues/280
- This change covers two use cases:
- The accept-version > current version + the request cannot be served: ERROR CASE 1
- The accept-version < current version + the request cannot be served: ERROR CASE 2
- Along with 406 status there's additional information about the probable cause and action to be taken by the Ghost site owner or an integration talking to the Ghost API.
- These errors is designed to allow introducing breaking API changes gradually and have meaningful information when the requests cannot be server any longer
refs https://github.com/TryGhost/Toolbox/issues/280
- In response to 'Accept-Version' header in the request headers, Ghost will always respond with a content-version header indicating the version of the Ghost install that is responding. This should signal to the client the content version that is bein g served
- This is a bare bones implementation and more logic with edge cases where `content-version` is served with a version value of "best format API could respond with" will be added later.
refs https://github.com/TryGhost/Team/issues/1470
Instead of counting the MRR by resolving all the deltas from the past until now, we should start with the current calculated MRR and resolve it until the first event. That would give a more accurate recent MRR (in exchange for a less accurate MRR for older data) and allows us to limit the amount of returned days in the future.
- Includes MRR stats service that can fetch the current MRR per currency
- The service can return a history of the MRR for every day and currency
- New admin API endpoint /stats/mrr that returns the MRR history
- Includes tests for these new service and endpoint
refs https://github.com/TryGhost/Team/issues/1248
refs https://github.com/TryGhost/Team/issues/1302
refs https://github.com/TryGhost/Ghost/pull/14433
This fixes multiple problems we've had with handling stripe webhooks where
linking the same subscription concurrently was not locking the database row, we
would have the same operation attempted multiple times, in the case of creating
subscriptions this causes unique constraint errors, and in the case of creating
cancellation events, it was causing multiple cancellation events to be created.
refs https://github.com/TryGhost/Team/issues/1474
- The `default` concept will be replaced by the first newsletter based on the `sort_order`
- This removes the `default` value from the newsletter API
- This simplifies the design to make the api and datastructure more maintainable
refs https://github.com/TryGhost/Team/issues/1471
- This is a many-to-one relation so that many posts can be linked to a specific newsletter
- The `newsletters` table had to come first in the schema file so that it's initialized before the `posts` table (because of the foreign key)
- Updated the model to make sure the new field doesn't leak in the API for now
- This migration isn't using the `createAddColumnMigration` util because of a performance issue. In MySQL, adding/dropping a column without `algorithm=copy` uses the INPLACE algorithm which was too slow on big posts tables (~3 minutes for 10k posts). Switching to the COPY algorithm fixed the issue (~3 seconds for 10k posts).
- SQLite isn't using the codepath where we run a raw SQL query because `knex` is doing multiple queries to add/remove a column
refs https://github.com/TryGhost/Team/issues/1469
Currently, all new members get auto subscribed to the default newsletter. This change adds same behavior with multiple newsletters by auto subscribing all available newsletters on site for new members(If flag is enabled).
Note: In future, this will also take into consideration the `subscribe_on_signup` flag for a newsletter to filter which newsletters should a member be auto-subscribed.
- adds newsletters service for working with newsletter data
- bumps `@tryghost/members-api` package which handles default subscription
- adds new test fixture/data for newsletters
- The url service was moved from frontend to server some time ago but the tests were forgotten
- This is only being done now because in 5.0 major changes are happening and it'll be annoying if the
files move on that branch
refs https://github.com/TryGhost/Team/issues/1372
- Added tests for Stripe webhooks that cancel a subscription (paid and complimentary)
- Tests for manually adding a complimentary member, and removing it again (the 'new' way)
- Added tests for creating new members (paid, complimentary)
- Includes tests for the `stripe_customer_id` property when creating new members (this is broken, fixed by https://github.com/TryGhost/Members/pull/378#issuecomment-1070835800)
- Deduplicated some nock code for Stripe
- Improved event assertions and interference between multiple tests in the giant members test file (by asserting only for the affected member id instead of all events).
no issue
- It was hard to plug in with additional code into current `.then` based chain. Refactoring to use a more modern syntax helps with readability and allows for easier edits
- This fits more closely, as this service is to so with rendering helpers and small parts
- Whereas we want to use "rendering" for things concerned with rendering pages
refs https://github.com/TryGhost/Team/issues/1469
- wires newsletter preferences for a member in Portal to view/edit real data
- allows members to control their subscription to multiple newsletters
refs https://github.com/TryGhost/Team/issues/1469
We have an after hook to update `sort_order` for pivot tables after update as base bookshelf plugin. Since new `members_newsletters` table doesn't has a `sort_order` column on it, this change disables the after hook for it to avoid errors.
refs https://github.com/TryGhost/Team/issues/1469
- allows Portal access to multiple newsletters on a site
- allows Portal to update member's newsletter subscription preference
refs https://github.com/TryGhost/Team/issues/1469
- updates member model to add relation to newsletter via pivot table
- updates member api serializer to include newsletter data
- updates tests
refs: https://github.com/TryGhost/Ghost/commit/11867ab43
- These checks live in the wrong place. They are mostly a frontend thing
- The only server place they were used was slack and that was fixed in 11867ab43
- Moving these to the frontend they fit neatly into the frontend data service
- This is the only piece of server code that relies on the schema.checks, the rest are all frontend
- IMO this code should actually check for the post properties that the slack message needs
- OR it should switch based on the event type
- either way there's no need to have a shared util for this simple use case
- especially becaue it's confusing the use case for it and creating cross-coupling between server and frontend
- Some of the helpers inside the routing service would be better suited to their own service
- These two helpers fetchData and entryLookup talk to the API to get data & so make a decent start for a data service
- The data service would be the single point of contact with the API for the frontend
- Doing this now cos I'm moving some files around ahead of deleting things for 5.0
- We had an incident where a migration file was misnamed.
- This is caused by using slimer with spaces rather than hyphens
- We didn't imagine this case when writing the regex for matching filenames
- However, now we know it, it's easy to tweak the regex to match this case
- This requires adding an override to the one badly named file
- Keeping overrides inside the file means the context of why a rule isn't firing is present inline when modifying a file
- This makes it easier to maintain files, at the cost of needing to search to find all overrides
refs 2fa3985d42
- Running tests with error logging set to "error" lever, produced a massive amounts of errors related to failed Stripe keys. Making it hard to look through the output.
- When Ghost is running in teste environment by default it is configured with an invalid Stripe key that looks like `sk_test***`. In this case the Members migrations runs creating requiest to Stripe, which fail.
