no issue
- job registration was checking for submitted emails in it's email count but the job registration method is called as soon as an email is created meaning the email has a status of 'pending' which prevented the analytics job from being started until a second email was sent
no issue
- email analytics may be desirable to fully switch off in certain circumstances, when that happens we want to prevent related background jobs from running and expose the feature flag via the config endpoint in the Admin API so that clients can adjust accordingly
no issue
- if emails are older than 30 days we wouldn't be able to fetch any analytics for them and if a site used emails in the past but is no longer using them it doesn't make sense to keep potentially expensive background worker threads spinning up
no issue
- recurring jobs spin up worker threads which can be quite CPU intensive even when not performing much processing, this can be problematic in environments where there are many Ghost instances running
- updated the email job scheduling to be skipped on bootup when there are no emails in the database and to be started when the first email is created as long as we're not in testing env
- increase analytics job schedule from every 2 minutes to every 5 minutes to help spread the load further across instances
no issue
- typically cron/later schedules will schedule for :00 on the minute which would create API spikes with every members-email-using Ghost site hitting the API at the same time
- adjusted the scheduling to use cron syntax with job runs every 2 minutes on 1,3,5... or 2,4,6... and a random seconds value to smooth usage across sites
no issue
- it's possible background jobs may cause unintended side-effects so it's useful to have a kill-switch to disable them individually to keep sites working
- users imported from CSV with no created_at date where having their created_at date being stored as an int rather than a datetime.
- this was causing parsing issues with the graph so this commit fixes the formatting
closes#12083
- fixes a parsing issue where negative offset values were incorrectly having the + sign added regardless of actual offset for sqlite databases.
- for mysql databases absolute values of offset were taken with sign applied where appropriate to stop issues where both hours and minutes could be negative which would cause both an issue with offsets that could present as -2:30 and by the look of the code also trigger extra padding to result in -2:-030 rather than the expected -2:30
no issue
- the 4.2.0 version of `sqlite3` that we're using is not compatible with `worker_threads`
- 5.0.0 should add support this but there are other errors
- 5.0.1 is released but not published (https://github.com/mapbox/node-sqlite3/issues/1386)
no issue
- added `EmailAnalyticsService`
- `.fetchAll()` grabs and processes all available events
- `.fetchLatest()` grabs and processes all events since the last seen event timestamp
- `EventProcessor` passed event objects and updates `email_recipients` or `members` records depending on the event being analytics or list hygiene
- always returns a `EventProcessingResult` instance so that progress can be tracked and merged across individual events, batches (pages of events), and total runs
- adds email_id and member_id to the returned result where appropriate so that the stats aggregator can limit processing to data that has changed
- sets `email_recipients.{delivered_at, opened_at, failed_at}` for analytics events
- sets `members.subscribed = false` for permanent failure/unsubscribed/complained list hygiene events
- `StatsAggregator` takes an `EventProcessingResult`-like object containing arrays of email ids and member ids on which to aggregate statistics.
- jobs for `fetch-latest` and `fetch-all` ready for use with the JobsService
- added `initialiseRecurringJobs()` function to Ghost bootup procedure that schedules the email analytics "fetch latest" job to run every minute
no issue
- cleans up unused tables `emails.{meta,stats}`
- adds timestamp columns `email_recipients.{delivered_at,opened_at,failed_at}` that can be used for event timelines and basic stats aggregation
- indexed because we want to sort by these columns to find the "latest event" when limiting Mailgun events API requests
- adds aggregated stats columns `emails.{delivered_count,opened_count,failed_count}`
- adds a composite index on `email_recipients.[email_id,member_email]` to dramatically speed up `email_recipient` update queries when processing events
- modifies the db initialisation to support an `'@@INDEXES@@'` key in table schema definition for composite indexes
closes https://github.com/TryGhost/Ghost/issues/12416
This fixes compatibility for the `send_email_when_published` option for
the Posts API.
The model layer only allows setting the `email_recipient_filter` column
when the `status` is being changed. Because of this we need to withhold
the `status` change until after we have determined the
`email_recipient_filter`.
- this helps simplify the code and gets rid of Promise chaining
- apparently I can't easily use an async function within filter, so I've
left it for now
refs https://github.com/TryGhost/Ghost/issues/12256
We no longer want to filter out cancelled subscriptions, so we are able
to remove the whereIn clause of the relation.
* Fixed paid flag on member
* Fixed content gating for members
Now that the subscriptions for a member include all of them, we must
explicitly check that the member has an active subscription in order to
consider them "paid"
- this helps bring all the code together so we can extract it in the
future
- turning it into a class also lets us easily inject the i18n instance
and store it locally
no issue
- the `'bulk-email`' tag was only being added to bulk emails if another more specific tag was set up via config
- we always want the `'bulk-email'` tag to be present for better event filtering
refs #12402
- With bumped version of job-manager it offloads job procesing into separate worker thread. Having jobs run out of main Ghost process even loop allows for safe job execution, which does not block Ghost from serving requests or performing other functions without a delay
- Added experimental data access to 'testmode' jobs. This should serve as an illustration of how to access data from the job layer
refs https://github.com/TryGhost/Ghost/issues/10354
- The serializer wasn't hooked up properly during the implementation (1a4497fc9a). It is not possible to hook it back in now as that would introduce breaking change to now stable v2 API.
- v3 (canary) has is properly hooked in, so there should not be a problem going forward
refs #12366
This implements redirection based on the settings for successful member sign up!
- Removes support for redirecting to `req.path` afterwards, this was never used and
we now have a more configurable implementation.
- Retains redirection to the homepage for unsuccessful sign up (invalid/expired token)
no-issue
This logic would assume that the option was always passed at the point
of publishing the post, which is not the case for scheduled posts.
Instead of setting the property to 'none' when the option is not
present, we take the approach of ONLY setting the propery when
1. It is present and not 'none'
2. The post is being published or scheduled
This means that scheduled posts will have the property set correctly,
and any future publishing will leave the it in the original state
no refs
- Migrations were using previous default for badge which was set to `false`
- Default config for badge was `false` which caused migration to switch off the badge, updated to `true`
- these helpers remove a lot of the duplicated code that we had when doing up/down column migrations, and provides a much shorter way of doing this in the future
no issue
Updated newsletter badge to be shown by default which was previously set to be hidden. Users can still control the visibility via Email settings in Ghost-Admin
no-issue
We want to give users to ability to customise the content of their newsletter, and the first step
toward that is a setting in which we can store text or html to embed in the template
no-issue
* Handled send_email_when_published in Posts API
This restores backwards compatibility of the Posts API allowing existing
clients to continue to use the `send_email_when_published` flag. This
change uses two edits, which is unfortunate. The reason being is that
this is an API compatibility issue, not a model issue, so we shouldn't
introduce code to the model layer to handle it. The visibility property
of the model is used to determine how to fall back, and because it can
be left out of the API request, and relies on a default in the settings,
we require that the model decide on the `visibility` before we run our
fallback logic (or we duplicate the `visibility` default at the cost of
maintenance in the future)
* Dropped send_email_when_published column from posts
Since this column is not used any more, we can drop it from the table.
We include an extra migration to repopulate the column in the event of
a rollback
* Updated importer to handle send_email_when_published
Because we currently export this value from Ghost, we should correctly
import it. This follows the same logic as the migrations for this value.
* Included send_email_when_published in API response
As our v3 API documentation includes `send_email_when_published` we must
retain backward compatibility by calculating the property.
* Fixed fields filter with send_email_when_published
* Added safety checks to frame properties
Some parts of the code pass a manually created "frame" which is missing
lots of properties, so we check for the existence of all of them before
using them.
* Fixed 3.1 migration to include columnDefinition
We require that migrations have all the information they need contained
within them as they run in an unknown state of the codebase, which could
be from the commit they are introduced, to any future commit. In this
case the column definition is removed from the schema in 3.38 and the
migration would fail when run in this version or later.
refs https://github.com/TryGhost/Ghost/issues/12355
- Adds new default settings for newsletter customisations - `newsletter_show_badge`, `newsletter_show_header` and `newsletter_body_font_category`
- Adds migrations to update group for new settings
- Add migration to update settings based on existing config value for newsletter settings
- Passes new newsletter settings to newsletter template and updates design based on them
- Fix tests
no issue
- These changes should allow easy testing of scheduled jobs and provide an implementation example for future jobs within Ghost
- Added experimental support to run jobs based on functions defined in modules
- Extendeded testmode "schedule" endpoint with an optional job name parameter which picks up job from "jobs" folder
- Adjusted scheduleJob method use and passed "schedule" as first parameter
- Bumped job-manager version to allow for all new functionality
no issue
- Living in separate folder will allow adding more testmode specific modules (for example test jobs)
- Mimicks the folder structure of how v2/canary routes are structured better
no-issue
* Used email_recipient_filter in MEGA
This officially decouples the newsletter recipients from the post
visibility allowing us to send emails to free members only
* Supported enum for send_email_when_published in model
This allows us to migrate from the previously used boolean to an enum
when we eventually rename the email_recipient_filter column to
send_email_when_published
* Updated the posts API to handle email_recipient_filter
We now no longer rely on the send_email_when_published property to send
newsletters, meaning we can remove the column and start cleaning up the
new columns name
* Handled draft status changes when emails not sent
We want to reset any concept of sending an email when a post is
transition to the draft status, if and only if, and email has not
already been sent. If an email has been sent, we should leave the email
related fields as they were.
* Removed send_email_when_published from add method
This is not supported at the model layer
* Removed email_recipient_filter from v2&Content API
This should not be exposed on previous api versions, or publicly
* Removed reference to send_email_when_published
This allows us to move completely to the email_recipient_filter
property, keeping the code clean and allowing us to delete the
send_email_when_published column in the database. We plan to then
migrate _back_ to the send_email_when_published name at both the
database and api level.
no issue
- set `emails.track_opens` to `true` when the `enableDeveloperExperiments` flag is set
- update mailgun bulk-email provider to pass the open-tracking header to Mailgun when the email's `track_opens` flag is set
- this check should never really be hit unless the DB has missing data,
but the code tries to read from an undefined variable and crashes, which we
can protect against
- this can be triggered by doing a `knex-migrator init` followed by a
`knex-migrator rollback --v 3.0 --force`
- 504509bb6 removed the global override for Promise
- there are a bunch of places in code that use Bluebird Promise methods,
but Bluebird wasn't being imported in these places
- this would have thrown errors all over the place
no issue
- tracking of bulk email opens can be enabled/disabled over time, if we're calculating analytics for emails we don't want emails which didn't have tracking enabled skewing the results so we need a record of whether tracking was enabled for each email
no issue
- We need an ability to early test scheduled jobs to refine the new feature and it's API. Should be used with caution
- To schedule an example scheduled job every 30 seconds run following request: `curl http://localhost:2368/ghost/api/schedule/every%2030%20seconds`
refs #11085
- Incorrect usage error was logged to the output when there was no recirecst configuration file present in the system. Previously an empty string was returned in such situation, resulting in "ENOENT" error, which was ignored through special handling.
- The fix resembles logic in redirects async getter function where empty array is returned when the config file does not exits.
- Attempting to read unexistent config should not ever happen and will be handled on the config service layer, this is why special "ENOENT" handling has been removed
no-issue
This column will allow us to store the canonical recipient filter on the
email resource giving us a detailed log of which members an email was
intended for
no-issue
This column will allow us to decouple the recipients of newsletter from
the `visibility` of a post, allowing us to send emails to specifically
free members, or to send paid posts as newsletters to all members.
closes#11085
- Ghost has been using YAML format for other configurations (e.g. routes). The plan is to move to this format for all user-edited settings files. By default JSON format is still used in Ghost Admin API v2/v3, but will be changed to YAML in API v4. Check referenced issue for more context.
- New format supports all the features available before. The main noticeable change is the structure of config file. It is now grouped by redirect HTTP code instead of specifying `"permanent": true | false` attribute for each config property. Example format for YAML config:
```
302:
/from-url/: /to-url/
301:
/category/([a-z0-9\-]+)/i: /tag/$1/
/v([0-9\.]+)/docs/([a-z0-9\-]+)/i: /docs/$2/
```
- Added 2 new endpoints: `POST redirects/upload` and `GET redirects/download`. These serve as an alias to current GET/POST `/redirects/json. "upload/download" naming pattern is introduced to match the convention with other resources that can be uploaded and downloaded (images, themes etc.). `/redirects/json` endpoints will be removed in Admin API v4
- The parsing code from `custom-redirects.js` has been moved to `frontend/services/redirects/settings.js`. This location is more appropriate for this logic and eventually `custom-redirects.js` middlewear might be moved into "frontend" as this middlewear plays a role mostly effecting that area.
no refs
[Portal](https://github.com/TryGhost/Portal) is a new drop-in script to make the bulk of Ghost membership features work on any theme out of the box, which was under a developer flag so far. This release removes the flag for Portal and makes it included as default for any members-enabled Ghost site. The Portal script is backward compatible with old public members script and existing Members-enabled themes should notice no change.
- Removes Portal config flag as Portal is now enabled by default
- Removes old members script as Portal is backward compatible with it
- Changes `{{content}}` helper to show default CTA in case of restricted content access
- `accent_color` setting is no more behind the dev experiment flag and included by default
- Adds migration to switch off Portal button setting for all existing sites which don't have Portal enabled in beta
closes#12263
- Express parses repeated query parameters as an array (req.query properties). Because there is no clear reason on why not to support this behavior extended order parameter parsing logic to handle arrays. This follows the rule of "liberal inputs, conservative outputs"
- Example supported query string for ordering can now look like: `?order=featured&order=published_at asc`, the priority of the order stays the same with the most significant appearing first and least significant last
refs #11572
- Filtering by fields coming from posts_meta table did not work for post resources. This was due to lack of support for these types of operations on NQL layer. The approach taken here is using same way filtering was done for many:many relations and generates a `WHERE IN` filtering clause. In the future we could look into adding preloading of 1:1 relations which should allow getting rid of `WHERE IN` in favor of `JOIN` and filtering directly by field names.
- Changed structure of `EXPANSIONS` filter configuration. Current approach was based on "bag of all the things". Such structure will become problematic as more fields are added. For example, adding all the fields from 1:1 relation posts:posts_meta might collide with any other relations that would have similar naming like meta_description from tags table (if it were was added).
- Bumped nql version to 0.5.0. This adds filtering support to 1:1 relations
- Added filter expansions which can be unique per model Previous approach with single global expansions lookup wasn't working in case different models would need to declare expansion for same field names. Having a `filterExpansion` method per model works in a similar convention other filter related model methods do (e.g. enforcedFilters, defaultFilters)
no issue
- adding user variables via the mailgun API when sending emails means that events related to email have those variables attached to them
- adding the `email.id` value to user variables means we can easily associate mailgun events to emails, otherwise we'd have look up the batch via `email_batches.provider_id` then use a join to get back to the associated email
- currently if Ghost has a version of something like
`3.37.0-pre.abc.def`, Ghost will return `3.37.0-pre.abc` as the full
version
- this hides parts of the version which are useful for debugging
- this commit updates the logic to join together all prerelease elements
so we keep the full string
no issue
The email change verification template was using the same as for `subscribe`, which did not have the right messaging that needs to be communicated about the action thats happening in this case. Updates the email template to same as what we use for email verification for support/newsletter address
closes https://github.com/TryGhost/Ghost/issues/12253
- Allows using custom action param for requests from Portal by using a new `requestSrc` option that is passed down when a request for magic link is made via Portal
refs https://github.com/TryGhost/Ghost/issues/12256 , https://github.com/TryGhost/Ghost/issues/12255
Currently when listing subscriptions for Members in both the Admin and the Theme, we only show the subscriptions which have a status of trialing or active.
Based on discussion, the `unpaid` and `past_due` states on Stripe also represent owner's intention of considering a subscription as active instead of `cancelled`, so we allow any subscriptions under these 2 states to be also listed for a member and consider them as `paid`.
- Subscriptions will go into a past_due state if the payment is missed, this should be considered a grace period where the member still has access.
- After this the subscriptions will either go to the unpaid or the cancelled state - this can be configured on an account by account basis in the Stripe dashboard. `unpaid` is considered as an intention to keep the subscription to allow for re-activation later.
no refs.
- Members settings were moved to a new page [here](https://github.com/TryGhost/Ghost-Admin/pull/1736)
- updated members related redirect for from/newsletter email address change to point to new page
no refs
- The token generation logic for single use token was replacing only the first instance of + or / to make the token URL safe, instead of replacing all instances which caused a bug where token was not validated properly in case it included multiple + or / in it.
- The fix ensures replacing all the + or / in the token with URL safe _ or - so it can be properly validated via magic link
closes https://github.com/TryGhost/Ghost/issues/11548
- bumps @tryghost/kg-default-cards
- introduces a breaking change where `imageOptimization` is passed as an options argument instead of separating out individual config options
- fixes width/height attributes for images in gallery cards being larger than the max default width used when resizing images
no issue
- centralises definition of max width and allows customisation if needed
- allows for passing of the config value through to rendering libraries
no issue
- standard browse/read/add/edit/destroy API endpoints for snippets resource
- updates `@tryghost/admin-api-schema` dependency to version that includes snippet definition and schemas
no issue
- all staff users can browse/read snippets so they're usable in the editor for everyone
- only administrators, editors, and admin integrations are able to create/edit/delete snippets
no issue
- minimal table structure required for the first iteration of content snippets
- snippets are stored pieces of re-usable content that could effectively be entire posts so the `mobiledoc` field length matches the `posts.mobiledoc` field length
no issue
- Member email settings - newsletter and support - was previously being set to null for all cases, which can lead to `null` values being set for those addresses on a successful import whereas the expected behavior is to retain the old values.
- Adds the 2 email settings to ignored settings list so they are not changed during import process
no issue
- In a case where support address is null, member's site data won't load as there was no fallback.
- The fix adds default `noreply` as the support address for site data
no-issue
The Action model loops through all registered models when the file is
loaded, by loading the model last, we ensure that it can read all
models, rather than an arbitrary selection which come before it.
closes#12273
- `comped` field has been allowed when editing a member or importing from a CSV. There has been a usecase (Zapier Integration) for API client to create a member with "Complimentary" plan, which made this change necessary
- Previously the logic for comped field was to skip and continue member record creation if Stripe was not connected. Now we throw an error - same as the one we have been throwing before when stripe_customer_id field was passed in. The implication of this change is that we won't be creating any record now if comped === true and Stripe is disabled.
- Bumped admin-api-schema-package. Contains `comped` schema change so this field gets passed through to controller
no issue
When scheduling a post to publish+send the "view online" link was pointing at https://site.com/404/ rather than the published post's url.
The problem occurred because the `/schedules/` endpoint wraps it's post read+edit calls in a transaction. Context:
- when a post is published with with the "send email" option the email record is immediately generated and added to the API response, as part of the email record generation we render the email content including fetching the url for the "view online" link
- urls for all resources are handled by our `url` service, that service updates it's internal cache based upon model events such as the "edited" event triggered when a post is published
- if the posts API controller is given a transaction, the email record is also generated inside of that transaction however at this point the `url` service will not have been updated because the post record hasn't been committed meaning it has no available url for the post
Fix:
- removed the `models.Base.transaction()` wrapper around the post read+update in the `/schedules/` API controllers
- we don't need a transaction here. It was added as protection against another write request coming in between the `/schedules/` controller reading a post and publishing a post but we already have protection against that in the form of collision detection - if a write request comes in and commits between the schedules controller reading the post and updating it, the scheduler's update call will fail with a collision error at which point the scheduler itself should retry the request which could then publish the post successfully if everything else is in order
closes https://github.com/TryGhost/Ghost/issues/12260
- if a card type was not explicitly chosen (i.e. a url was pasted into the editor) then abort fetching the oembed endpoint if we detect it's a `wp-json` oembed and return a bookmark card payload instead
- cleaned up an unused argument in the internal `fetchBookmarkData()` method
closes https://github.com/TryGhost/Ghost/issues/12257
- there was a destructuring problem introduced in the recent email refactor which meant the correct replacement data was not being passed over to the Mailgun provider when sending email
closes https://github.com/TryGhost/Ghost/issues/12259
- adds a `DISTINCT` to the query used to fetch member rows when generating an email recipient list
- this increases query time 2.7s vs 1.6s locally with ~94k paid members but once the `members.paid` column is implemented this slow query can be removed
closes https://github.com/TryGhost/Ghost/issues/12247
- Internal preview controller was lacking "mapping" call to post object which handled not only missing meta attribute information but lots of other mappings (e.g. users, tags, etc.)
- Have added a regression test to catch issues like this in the future
no issue
- fixed passing of errors up through send/processBatch/processEmail
- fixed errant overwrite of email status with a "submitted" status after a failure had occurred
no issue
- wrap email batch/recipient record creation in a transaction so if an error occurs during creation we're not left with a partially created batch/recipient set in the database
no issue
- if an error occurred whilst creating email batch/recipient records the email status was never updated and was left in the 'pending' status
- adjusted the error handling to update the email status and record the error message if such a scenario occurs
no issue
- the paid-member SQL query that is obtained using `models.Member.getFilteredCollectionQuery({paid: true})` can return multiple columns with the same name (eg, `email`, `name`), when that happens the last column with duplicate names "wins" and it's value is used in the resulting knex row instance
- in the `mega` service when fetching email recipient rows we ran into this problem, to avoid it we adjust the query to explicitly select only the data from the `members` table
no issue
- we've had an issue with emails failing due to unexpectedly missing data when inserting email recipient rows
- added a validation check before adding recipient details along with a log so that invalid data can be investigated
no issue
- We had previously allowed accent_color setting for member site settings behind portal flag, but Ghost Admin also needs the public site setting with accent color to correctly reflect the accent color when flag is switched on
- Removes deletion of accent color setting when behind the Portal flag OR dev experiment flag
no issue
- there was an typo in the recent email sending refactor that resulted in `Error: 'to' parameter is missing` errors when sending email previews and bulk emails
refs #12249
This was incorrectly assuming the presence of the data-members-name
element in the document. By guarding against it and defaulting to
undefined, we fallback to the existing behaviour when the element is not
present.
closes#12244
As per RFC 6454 the Origin header MUST be set to the string 'null' when
in a "privacy-sensitive" context. We were not handling this string and
this was causing errors. This commit updates all checks of the 'Origin'
header to treat the value 'null' as if the header was not present.
ref: https://tools.ietf.org/html/rfc6454#section-7.3
refs #12055
As part of the work in TryGhost/Members#206 we load the stripeCustomers relation on the member model, and we do not want this to be part of the API response. The changes here include a refactor but the main thing is that the serialized object is explicit and does not include unexpected or unknown fields.
* Moved mapMember out of mapper file
This cleans up the serializer a bit by keeping it's functionality all in
one place, rather than a shared mapper file
* Refactored members controller to return models
Previously the controller was calling toJSON, which is serialization,
this updates the controller to only deal with models, leaving all of the
serialization to the serializer!
* Refactored members serializer
This adds typings to all of the methods/functions in the serializer, as
well as making the serializating explicit, rather than returning the
result of toJSON, we explicitly set the properties we expect to be on
the output object. This protects us against accidental API changes in
the future.
closes https://github.com/TryGhost/Ghost/issues/10628
- JSON Schemas were extracted into a separate module to allow other clients to reuse them (for example documentation). Having them in a separate package also slims down the amount of code needed to be maintained in the core.
- Updated canary API input validators to use admin-api-schema module
- Removed canary schemas that moved into admin-api-schema package
- Updated v2 API input validators to use admin-api-schema package
- Removed v2 schemas that moved into admin-api-schema package
- Updated tests to contain needed information in apiConfig to pick up correct validation
- Added @tryghost/admin-api-schema package dependency
refs https://github.com/TryGhost/Ghost/pull/12232
When viewing sent emails in Ghost's admin area, it displays the `html` field directly from the `email` relation loaded with the post. Since the `mega` refactor we now store raw content in that field rather than sanitized "preview" content so it's necessary to modify the API output to match the old behaviour.
- use the API output serializers to parse replacements in email content and replace with the desired fallback or empty string
no issue
- store raw content in email record
- keep any replacement strings in the html/plaintext content so that it can be used when sending email rather than needing to re-serialize the post content which may have changed
- split post email serializer into separate serialization and replacement parsing functions
- serialization now returns any email content that is derived from the post content (subject/html/plaintext) rather than post content plus replacements
- `parseReplacements` has been split out so that it can be run against email content rather than a post, this allows mega and the email preview service to work with the stored email content
- move mailgun-specific functionality into the mailgun provider
- previously mailgun-specific behaviour was spread across the post email serializer, mega, and bulk-email service
- the per-batch `send` functionality was moved from the `bulk-email` service to the mailgun provider and updated to take email content, recipient info, and replacement info so that all mailgun-specific headers and replacement formatting can be handled in one place
- exposes the `BATCH_SIZE` constant because batch sizes are limited to what the provider allows
- `bulk-email` service split into three methods
- `send` responsible for taking email content and recipients, parsing replacement info from the email content and using that to collate a recipient data object, and finally coordinating the send via the mailgun provider. Usable directly for use-cases such as test emails
- `processEmail` takes an email ID, loads it and coordinates sending related batches concurrently
- `processEmailBatch` takes an email_batch ID, loads it along with associated email_recipient records and passes the data through to the `send` function, updating the batch status as it's processed
- `processEmail` and `processEmailBatch` take IDs rather than objects ready for future use by job-queues, it's best to keep job parameters as minimal as possible
- refactored `mega` service
- modified `getEmailData` to collate email content (from/reply-to/subject/html/plaintext) rather than being responsible for dealing with replacements and mailgun-specific replacement formats
- used for generating email content before storing in the email table, and when sending test emails
- from/reply-to calculation moved out of the post-email-serializer into mega and extracted into separate functions used by `getEmailData`
- `sendTestEmail` updated to generate `EmailRecipient`-like objects for each email address so that appropriate data can be supplied to the updated `bulk-email.send` method
- `sendEmailJob` updated to create `email_batches` and associated `email_recipients` records then hand over processing to the `bulk-email` service
- member row fetching extracted into a separate function and used by `createEmailBatches`
- moved updating of email status from `mega` to the `bulk-email` service, keeps concept of Successful/FailedBatch internal to the `bulk-email` service
requires https://github.com/TryGhost/Ghost/pull/12192
- added initial `EmailBatch` and `EmailRecipient` model definitions with defaults and relationships
- added missing `post` relationship function to email model
- fetch member list without bookshelf
- bookshelf can add around 3x overhead when fetching the members list for an email
- we don't need full members at this point, only having the data is fine
- if we need full models later on we can push the model hydration into background jobs where recipient batches are fetched ready for an email to be sent
- bookshelf model instantiation of many models blocks the event loop, using knex directly keeps concurrent requests fast
- adds `getFilteredCollectionQuery` method to base model to facilitate getting a knex query based on our normal model filters along with transaction/forUpdate applied
- store recipient list before sending email
- chunk already-fetched members list into batches and insert records into the `email_recipients` table via knex
- chunked into batches of 1000 to match the number of emails that Mailgun accepts in a single API request but this may not be the absolute fastest batch size for recipient insertion:
| Batch size | Batch time | Total time |
| ---------- | ---------- | ---------- |
| 500 | 20ms | 4142ms |
| 1000 | 50ms | 4651ms |
| 5000 | 170ms | 3540ms |
| 10000 | 370ms | 3684ms |
- create an email_batch record before inserting recipient rows so we can effeciently fetch recipients by batch and store the overall batch status
no issue
- The accent_color setting was being removed from members site data when behind portal flag
- The accent color setting is now allowed in members site data for all cases as it doesn't make any sense to remove it specifically from here where we already have all the other Portal settings included which is a dev/portal flag feature anyways
no-issue
By using the "email" validation, we were validating emails in CSV
imports using a different validator to the rest of the API. AJV's built
in email validation was failing on emails with "special" characters,
such as letters with an umlaut above them.
This commit brings the validation for CSV imports in line with the rest
of the API.
no issue
The email table should be a reference for all data that was used when sending an email. From and Reply-to addresses can change over time and we don't have any other reference for their value at the time of sending an email so we should store them alongside the email content.
- schema updated with `from` and `reply_to` columns
- both are set to `nullable` because we don't have historic data (can be populated and changed in later migrations if needed)
- neither `from` or `reply_to` have `isEmail` validations because they can have name+email in an email-specific format
- will help keep concerns separated in the future. `mega` service can deal with all of the email contents/properties, and the `bulk-email` service's concerns are then only email sending and any provider-specific needs
refs #12033
- Allowing to change parent integration opens up possible security holes and has no clear usecase at the moment. After a webhook record is created it should not be possible to change parent integration.
- Had do partially duplicate JSON schema definition from webhooks definition as there is no proper composition technique available in current version of JSON Schema.
refs https://github.com/TryGhost/Ghost/issues/12033
refs https://github.com/TryGhost/Ghost/issues/10567
- Creating a webhook without valid parent integration leads to orphaned webhook records, which shoult not ever happen
- This scenario is only possible for non-integration authentication,
because in case of integration being authenticated it's id is
automatically assigned to creatd webhook
refs #11729
- When ordering is done by fields from a relation (like post's `meta_title` that comes form `posts_meta` table), Bookshelf does not include those relations in the original query which caused errors. To support this usecase added a mechanism to detect fields from a relation and load those relations into query.
- Extended ordering to include table name in ordered field name. The information about the table name is needed to avoid using `tableName` within pagination plugin and gives path to having other than original table ordering fields (e.g. order by posts_meta table fields)
- Added test case to check ordering on posts_meta fields
- Added support for "eager loading" relations. Allows to extend query builder object with joins to related tables,
which could be used in ordering (possibly in filtering later). Bookshelf does not support ordering/filtering by proprieties coming from relations, that's why this kind of plugin and query expansion is needed
- Added note about lack of support for child relations with same property names.
closes#12038
Previously we were emitting changed events for _all_ settings which would
cause any listeners for those to be triggered, this ensures that listeners are
only triggered if the corresponding setting, _did_ in fact change.
no-issue
This change allows themes to collect member names on signup with use of the
`data-members-name` attribute on an input element in the `data-members-form`
refs #11729
- Having a plugin allows to reason about ordering better and gives way
to further extraction away form the core.
- Just like with filtering, ordering falls into similar category of having effect on knex'es query builder object extension through additional statements - ORDER BY in this case.
- Because there were bugs associated with use of permittedAttributes inside of `parseOrderOption` method, introduced separate `orderAttributes` function which returns only those fields which are orderable (https://github.com/TryGhost/Ghost/issues/11729#issuecomment-685740836)
refs #11878
- When password reset link is invalid previous messaging left the user
without clear information about why the reset failed and what they could do about it.
- Updated messaging around password reset tokens including detection of
when password token has invalid structure, has expired or has already
been used
This reverts commit 80af56b530.
- reverting temporarily so that all associated functionality can be merged in a single release
- creating email batch/recipient records without using them would cause inconsistent data
no-issue
* Added SingleUseTokenProvider to members service
This implements the TokenProvider interface required by members-api to
generate magic links. It handles checking if the token is expired and
pulls out any associated data.
Future improvments may include the email in the error for expired
tokens, which would make resending a token simpler.
* Passed SingleUseTokenProvider to members-api
This sets up the members-api module to use the new single use tokens
* Installed @tryghost/members-api@0.30.0
This includes the change to allow us to pass a token provider to the members-api
no-issue
This is a model for the tokens table, which handles the single use
aspect by customising the `findOne` method to automatically destroy the
model after reading from it
no-issue
refs: https://github.com/TryGhost/Members/commit/63942f03
The above commit updated all magic-link methods to be async, this change
ensures that we will handle a Promise return when it's updated in Ghost.
`await` has no effect on non Promise return values so it's safe to add
this now.
no-issue
After discussion with Matt, we decided that 192 bits for the token is a
good number, as it has no padding when base64 encoded and is more secure
than 128 bits, whilst still a managable size.
no-issue
This is a table to store single use tokens for use in magic links, the
columns are as simple as possible at the moment and are designed as:
id - standard ObjectID like all of our tables
token - 128bit base64 encoded string
data - arbitrary data to store against the token
created_at - timestamp to allow for expiry to be implemented for tokens
no issue
- In a recent change to ownership verification email flow, we changed the FROM address of ownership verification mails to use the same email as the one we are verifying, aka TO address.
- Email clients like Gmail flags off such emails as possible spam
- Fix updates the `FROM` address to `noreply@domain.com` where domain.com is domain for TO address
- In case the TO is already noreply@domain.com, we use no-reply@domain.com to bypass the same address restriction.
no issue
- The new Portal config flag allows switching on Portal conditionally with config
- The dev experiment flag still works for enabling Portal
- The flag currently defaults to `false` as Portal is still a beta feature and switched off by default
- We expose it on the admin api config endpoint so that the Ghost-Admin client can use it to conditionally render Portal settings
no issue
- Members site data was not appending blog domain for default support address which is `noreply`
- The change allows Portal to use default support address correctly
requires https://github.com/TryGhost/Ghost/pull/12192
- added initial `EmailBatch` and `EmailRecipient` model definitions with defaults and relationships
- added missing `post` relationship function to email model
- fetch member list without bookshelf
- bookshelf can add around 3x overhead when fetching the members list for an email
- we don't need full members at this point, only having the data is fine
- if we need full models later on we can push the model hydration into background jobs where recipient batches are fetched ready for an email to be sent
- bookshelf model instantiation of many models blocks the event loop, using knex directly keeps concurrent requests fast
- store recipient list before sending email
- chunk already-fetched members list into batches and insert records into the `email_recipients` table via knex
- chunked into batches of 1000 to match the number of emails that Mailgun accepts in a single API request but this may not be the absolute fastest batch size for recipient insertion:
| Batch size | Batch time | Total time |
| ---------- | ---------- | ---------- |
| 500 | 20ms | 4142ms |
| 1000 | 50ms | 4651ms |
| 5000 | 170ms | 3540ms |
| 10000 | 370ms | 3684ms |
- create an email_batch record before inserting recipient rows so we can effeciently fetch recipients by batch and store the overall batch status
no issue
We want to store a list of recipients for each bulk email so that we have a consistent set of data that background processing/sending jobs can work from without worrying about moving large data sets around or member data changing mid-send.
- `email_batches` table acts as a join table with status for email<->email_recipient
- stores a provider-specific ID that we get back when submitting a batch for sending to the bulk email provider
- `status` allows for batch-specific status updates and picking up where we left off when submitting batches if needed
- explicitly tying a list of email recipients to a batch allows for partial retries
- `email_recipients` table acts as a join table for email<->member
- `member_id` does not have a foreign key constraint because members can be deleted but does have an index so that we can efficiently query which emails a member has received
- stores static copies of the member info present at the time of sending an email for consistency in background jobs and auditing/historical data
refs #2635
- Adds 'Location' header to endpoints which create new resources and have corresponding `GET` endpoint as speced in JSON API - https://jsonapi.org/format/#crud-creating-responses-201. Specifically:
/posts/
/pages/
/integrations/
/tags/
/members/
/labels/
/notifications/
/invites/
- Adding the header should allow for better resource discoverability and improved logging readability
- Added `url` property to the frame constructor. Data in `url` should give enough information to later build up the `Location` header URL for created resource.
- Added Location header to headers handler. The Location value is built up from a combination of request URL and the id that is present in the response for the resource. The header is automatically added to requests coming to `add` controller methods which return `id` property in the frame result
- Excluded Webhooks API as there is no "GET" endpoint available to fetch the resource
closes#12045
- When member's email is updated to an already existing email of different member it caused table's unique constraint error, which was not handled properly.
- Added handling for this error similar to one in members `add` method.
closes#11999
- When the routes.yaml file changes (manually or through API) we need
to store a checksum to be able to optimize routes reloads in the future
- Added mechanism to detect differences between stored and current routes.yaml hash value
- Added routes.yaml sync on server boot
- Added routes.yaml handling in controllers
- Added routes hash synchronization method in core settings. It lives in core settings
as it needs access to model layer. To avoid coupling with the frontend settings it accepts
a function which has to resolve to a routes hash
- Added note about settings validation side-effect. It mutates input!
- Added async check for currently loaded routes hash
- Extended frontend settings loader with async loader. The default behavior of the loader is
to load settings syncronously for reasons spelled in 0ac19dcf84
To avoid blocking the eventloop added async loading method
- Refactored frontend setting loader for reusability of settings file path
- Added integrity check test for routes.yaml file
closes#12060
- A 500 error what happening when invited user provided an email that is associated with an existing user
- Additional validation for existing email address was added to prevent invalid data hitting db constraint error
refs #11999
- The `routes_hash` setting will be used during the boot process to update the hash
of currently loaded routes.yaml file in case it's different from last restart
no issue
- The async/await syntax makes it easier to reason about the code. Because adding 'Location' header is in the works it's a prep-work in a sense
closes https://github.com/TryGhost/members.js/issues/94
- The member-api package was recently updated to work directly with models and needs explicit `withRelated` options to attack relations
- Without options, the endpoint was returning the default member data without subscriptions attached, which in Portal showed paid member as free
- Fix updates the middleware for updating member data to correctly pass the relations needed to populate the member
no issue
- By default for new sites, support address is set same as from address to `noreply` , with full email address using the domain for `@`
- For newsletter emails, the support address was missing the default site domain to be added to address if its `noreply`
- Fix updates the support address to use the same format as from address and add relevant domain for default case
no refs
- The `update` method in members-api package was edited to return Model object instead of JSON directly - TryGhost/Members@a28bcc5
- This unsubscribe handler was returning the raw member object returned from `update` method, which is now a model object and not able to access `member.email`
- Fix updates the unsubscribe request handler to return the member JSON again
no issue
- members who have trial subscriptions added directly via Stripe will have a status of `"trialed"` in their Ghost subscription
- the `paid: true` filter was not taking that into account meaning trial users were not receiving newsletters sent to paid members even though they have a "paid" subscription
no issue
- We used existing "from" address as sender for mails sent to new email address for verification, but that breaks the flow to update if the current "from" address has DMARC policy set.
- This updates the flow to always send the ownership verification email TO the new address and FROM the new address which both verifies the email deliverability for new address and ownership
no issue
- The newsletter emails are sent out with `from` address as sender
- The new `members_reply_address` setting is now used to set reply-to address for emails, which can be either newsletter or support address
no issue
- Member auth emails were previously using the `from` address as sender
- New `members_support_address` was introduced with default as original "from" address
- Auth emails use the new support address as sender
no issue
- Updated magic link generation and validation methods for email update API to handle new support address
- Updated importer to ignore the new support address as it can only be updated via verification
- Updated members service to listen on settings edit for new support/reply address fields as well
- Updated tests to include the new settings
no issue
- Added default settings for the two new setting fields - `members_support_address` and `members_reply_address`
- Added migrations for setting group for new email settings
- Migration sets current from address as new support address default
- Added migration to set new support address same as from address
- Updated tests for new settings
- `members_support_address` - How members can reach for help with their account, public setting
- `members_reply_address` - Where you receive responses to newsletters
closes#12167
- Tags API v2 was ignoring `count.posts` include parameter.
- Regression was introduced with a3f693b472
- Introduced regression tests across all Content API versions to avoid similar bug in the future
no issue
- Added default settings for the two new setting fields - `members_support_address` and `members_reply_address`
- Added migrations for setting group for new email settings
- Migration sets current from address as new support address default
- Added migration to set new support address same as from address
- Updated tests for new settings
- `members_support_address` - How members can reach for help with their account
- `members_reply_address` - Where you receive responses to newsletters
no issue
- When an import was done and there were no "global labels" present Ghost created generic `import-[data]` label which later helped to find a specific batch of imported data
- It did not make sense to create such generic label when user provided their own unique label
- The rules that work now are:
1. When there is no global provided Ghost generates on and removes it in case there are no imported records
2. When there is a unique new global label provided no new label is generated, but the label stays even if there are no imported records
no issue
- tested performance between knex raw, knex `count()` and bookshelf `count()` and found no difference over 1000 iterations of each (each ~19,500ms +- 500ms for 104k members locally)
- switched to using bookshelf as the code is the simplest
no issue
- extract filtering of an collection into a separate function
- use extracted function in `findAll()` so that it's query behaviour matches `findPage()`
no issue
- for large result sets or complex queries the count query itself can be quite time consuming
- when `limit: 'all'` is passed as an option there's no need to perform a separate count query because we can determine the pagination data from the final result set
- skipped count query when `limit: 'all'` option is present
- re-ordered comments to be closer to the code they reference (ie, why we have our own count query instead of Bookshelf's `.count()`
no issue
- This is handled on input sanitization layer with date
format check in JSON schema validation, so there's no need to do this
check again in the importer.
no issue
- When bulk insert fails there is no transactional logic to revert
related records form being inserted. Also, previously there were no
attempts to "retry" the insert.
- To avoid complex retry logic, an iterative one-by-one insert retry
approach was taken. If this becomes a bottleneck in the future, the
retry algorithm could be improved.
- To avoid a lot of code duplication refactored model's `bulkAdd` & `bulkDestroy`
methods to use 'bulk-operations' module.
- Updated error handling and logging for bulk delete operations. It's very
unlikely for error to happen here, but still need to make sure there is
a proper logging in place to trace back the failure.
- Added debug logs. This should improve debugging experience and
performance measurements.
- Added handling for unrecognized errors. Handling inspired by current unrecognized
error handling by ghost importer -10e5d5f3d4/core/server/data/importer/importers/data/base.js (L148-L154)
no issue
- When no members are succesfully imported through CSV import process the import label should not be created. Otherwise after multiple failed attempts to import there are orphaned labels in the system
no issue
- updates `@tryghost/kg-default-cards` which contains two fixes
- removes email-specific output being added to post html (had no visual impact due to use of conditional comments but keeps rendered html smaller+cleaner)
- adds a background-url style to the thumbnail container to give two options for styling
- updates member email template styling to hide the `<img>` element in bookmark cards and use a background image instead to get consistent rendering across email clients
no issue
- adds a `members:emailTemplate` config object
- `showSiteHeader` - defaults to `true`, shows the site title and icon in member emails
- `showPoweredBy` - defaults to `false`, adds a "Publish with Ghost" button to member email footer
- updates member newsletter email template with hideable site header and "powered by" badge
closes https://github.com/TryGhost/members.js/issues/87
- The `update` method in members-api package was edited to return Model object instead of JSON directly [here](a28bcc5b2a)
- This caused the update member API on member endpoint to return partial response only as most properties couldn't be fetched
- Fix updates the middleware to correctly call `toJSON` before formatting response
* 3.30.2:
Bumped @tryghost/members-api to 0.28.1 in lockfile
Bumbed @tryghost/members-api to 0.28.1
🐛 Fixed unable to delete member when stripe is connected
refs #12127
- Adds new `editSubscription` endpoint for members admin API which allows updating individual subscription for a member - `PUT /members/:id/subscriptions/:subscription_id/`
- `editSubscription` has same permissions as member's `edit` endpoint
- Currently allows toggling of cancellation at period end for an active subscription
no issue
- Similar handling to one introduced in 8418c829de
- Having granular tracking for failed to remove id's would make it possible to return more specific errors to the client
no issue
- When batch insert fails handling should be more granular and aim to retry and insert as many records from the batch as possible.
- Added retry logic for failed member's batch inserts. It's a sequential insert for each record in the batch. This implementation was chosen to keep it as simple as possible
- Added filtering of "toCreate" records when member fails to insert. We should not try inserting related members_labels/members_stripe_customers/members_stripe_customer_subscriptions records because they would definitely fail insertion without associated member record
no issue
- When stripe is disconnected and there are Stripe-connected records present in imported set they should not be processed and proper error should be thrown
closes https://github.com/TryGhost/Ghost/issues/12139
- once the email content has been rendered in the post serializer, perform some whole-content transformation of `figure` and `figcaption` to `div` using cheerio
- juiced will have already inlined the elements styles so there's no need to adjust the template's stylesheet
closes#12078
- Root cause was that pseudo class .kg-bookmark-author:after was not getting inlined to email newsletter or its preview.
- That is where the margin and bullet-point content are added between author and publisher.
- Dependency juice supports an option inlinePseudoElements which is false by default.
- Fix was to set inlinePseudoElements to true when serializing post email.
Co-authored-by: Jeremy Davidson <jeremy@crossingcontour.com>
- testmode allows us to test Ghost's behaviour with long running requests and jobs
- moved all the testmode logging to only happen if Ghost starts successfully to make it clearer what is happening
- added a (currently very dirty) bit of code to fetch the jobService and output the queue status giving us a similar
output for both open connections and jobs
- all of this allows for debugging exactly what Ghost is doing whilst it is processing a shutdown request
closes https://github.com/TryGhost/Ghost/issues/11536
- bumps `@tryghost/kg-default-cards`
- image and gallery cards now output `width/height` attributes on `img` elements with a max width of 600px
- uses resized images where possible to keep email weight down
- adds `height: auto` style to image card images so that the `height` attribute does not cause distortion at smaller screen widths
no-issue
* Added bulkAdd method to Member,Customer&Subscription model
This allows us to keep the db access in the model layer
* Updated @tryghost/members-api to 0.27.2
This includes fixes for rate-limiting of requests, and exposes necessary
Stripe methods for creating customers and complimentary subscriptions,
without affecting the database.
* Refactored importer to parallelise tasks where possible
By parallelising our tasks we are able to improve the speed at which the
entire import completes.
refs #12126
- Adds migration to add impersonation permission to administrators
- Adds default permission fixture to allow administrators to read member impersonation urls
- Allows administrators to create member impersonation magic links
closes https://github.com/TryGhost/Ghost/issues/12129
Migrations were failing when upgrading from a version that didn't include stripe member subscriptions table. The `up` migration was checking for the existence of a unique index that was added to the schema in 3.29 but it was using the wrong variable name which meant that it would never return true for an existing index. For most migrations this worked because the index existed but when through 2.34 the table was created from scratch and did included the index at that point.
- fixed variable name and re-ordered variable assignment for better code locality that would have made the typo more visible
no issue
- Additional validation is needed for imported data because in case of bulk insertions (through knex) we bypass model layer validation - this could lead to invalid data in the database, which would be hard to fix.
- Chose validation method we use for other endpoints - through JSON Schema. It proved to be very performant (200ms overhead for 50k records). When comparing it with iterative method (validating each record separately) this was adding about 17s of overhead.
- Refactored returned values from "sanitizeInput" method to encapsulate more logic so that the caller doesn't have to calculate amount of invalid records and deal with error types
- Whole sanitizeInput method could now be easily extracted into separate module (somewhere close to members importer)
- Bumped members-csv package. It is meant to handle empty string values - '' and null, which should allow validating member records more consistently!
- upon further testing, this started to not work again so I'm removing
this horrible hack
- it could be Bunyan playing around with us, and the server is actually
shutting down
- I have no idea why this is required and I really don't like it
- this change looks superfluous but I had problems with Ghost not
displaying shutting down messages properly and this is the only
thing that got it working
- the async-awaits were getting stuck in `await this._cleanup()`,
and then wouldn't enter into the `finally` block
- oddly, this was only reproducible when a job was running, it would
shut down normally otherwise
- if I find the solution in the mean time, I'll happily get rid of this
no issue
- Updated sanitization logic to be self contained and return sanitieze input along with error stats
- This should give a nice place for validations to fit in
no issue
- There were many failed import records due to rate-limit errors. With concurrency of 9 imports go through with 100% success
- Would need to verify these limits with live API to make the most of it
no issue
- we output the post excerpt in a hidden div in the email template so that email clients pick it up as the "preview" text when listing emails
- when no custom excerpt is provided the preview text is grabbed from post.excerpt which is the first 500 chars of the post.plaintext value
- post.plaintext formats links as "Link [http://url/]" which is unwanted in html email previews
- add a basic replacement to the post email serializer to remove any `[http://url/]` occurrences from the post excerpt before rendering the email content
refs https://github.com/TryGhost/Ghost/issues/11536
- Outlook supports `'` as a special char for apostrophes but not `&#apos;` which is what cheerio/juiced render
- adds a basic string placement to the email serializer to switch to the older style of special char
no issue
- moves the meat of `pendingEmailHandler()` code into a new function `sendEmailJob()` that is passed over to the new job service
- lets the server keep processing email generation and sending when it receives a shutdown request rather than halting processing mid-send and ending up in a partial state
no-issue
* Added stripeSubscriptions relation to member model
This allows us to fetch the subscriptions for a member via standard
model usage, e.g. `withRelated: ['stripeSubscriptions']` rather than
offloading to loops and `decorateWithSubscriptions` functions, this is
more performant and less non-standard than the existing method.
* Updated serialize methods to match existing format
The current usage of `decorateWithSubscriptions` and the usage of
members throughout the codebase has a subscriptions array on a stripe
object on the member, this ensures that when we serialize members to
JSON that we are using the same format.
There is definitely room to change this in future, but this is an
attempt to create as few breaking changes as possible.
* Installed @tryghost/members-api@0.26.0
This includes the required API changes so that everywhere can use
members-api directly rather than models and/or helper methods
no issue
- The code in controller was becoming hard to reason about.
- Having a single module shows exactly how many dependencies are there to do an import for single batch.
- Having a separate module would make it easier to extract into it's own package in Members monorepo
no issue
- Member's labels have to have sort_order assigned when added/edited. This was lacking from batched importer.
- Implementation is based on logic used in model's base - e484709e73/core/server/models/base/index.js (L81-L86)
- Bottom line - we need to manage shutting down gracefully when doing long-running tasks
- To achieve that, we're going to use job queues
In this commit:
- added new @tryghost/job-manager dependency
- added a minimal job service, that handles in passing things like logging and (maybe later) config
- job service is wired up to server shutdown, so that the queue finishes before the server exits
- also added a new job endpoint to testmode so that it's easy to test job behaviour without needing to do real work
- Using consistent patterns for shutdown and stop
- Make it clear what each method does
- Use async/await to make the code more readable and simple
- This lays groundwork for having more cleanup tasks in stop than just server.stop()
- deleted files under `core/server/lib/promise` and related test files
- added `@tryghost/promise` as a dependency
- fixed all local requires to point to the new package
closes#12118
- server.start was mistakenly removed in 71f02d25e9
- it is used for loading themes (and other things) and is critical
- added tests to prevent this regressing again in future
no issue
- New Member API batched import is meant to be a substitution to current import
with improved performance while keeping same behaviore. Current
import processes 1 record at a time using internal API calls and times
out consistently when large number of members has to be imported (~10k
records without Stripe).
- New import's aim is to improve performance and process >50K
records without timing out both with and without Stripe connected
members
- Batched import can be conceptually devided into 3 stages which have
their own ways to improve performance:
1. labels - can be at current performance as number of
labels is usually small, but could also be improved through batching
2. member records + member<->labels relations - these could
be performed as batched inserts into the database
3. Stripe connections - most challanging bottleneck to solve because
API request are slow by it's nature and have to deal with rate limits of
Stripe's API itself
- It's a heavy WIP, with lots of known pitfalls which are marked with
TODOs. Will be solved iteratively through time untill the method can be
declared stable
- The new batched import method will be hidden behind 'enableDeveloperExperiments' flag to
allow early testing
- A simple way to test behaviours without having to do complex interactions to e.g. generate errors or slow requests
- Makes it easier to test the new shutdown behaviour, among other things
- there can now be quite a big delay between SIGINT/TERM being received and shutdown finishing
- add an extra log message to acknowledge the SIGINT/TERM to facilitate debugging and just be clear
- stopppable is a dependency that handles closing connections properly, which server.close does not
- active connections are allowed to complete what they are doing
- idle connections are closed
- no new connections are allowed
- we call stoppable in stop() instead of server.close so that idle connections don't hold the server open
- calling await stop() from shutdown then ensures that we have a consistent experience of stop
- all together this allows ghost to shutdown gracefully when there are long-running requests
- @TODO: handle graceful shutdown of long-running processes
- @TODO: consider do we need to send 503s whilst the server is shutting down?
- changed method to logStopMessages, as we use start and stop, not start and shutdown
- changed logStopMesasges to output the "proper" messages and use this method consistently - the closing connections message isn't really useful
- changed uptime message to always be output cos I can't see a case where there isn't interesting/useful
- Connection handling is legacy code added in 1438278ce4
- Although we were tracking all connections in memory, we weren't actually closing any because stop isn't called
- This (and restart) were both added as part of the now long-deprecated system for using Ghost directly as an npm module
- If we want to close connections cleanly, we should use a tool to do this
- the announce functions exist for the purpose of communicating with Ghost CLI
- the functions were called announceServerStart and announceServerStopped, which implies they tell Ghost CLI when the server starts and stops
- however, the true intention / purpose of these functions is to:
- either tell Ghost CLI when Ghost has successfully booted (e.g. is ready to serve requests)
- or tell Ghost CLI when the server failed to boot, and report the error so that Ghost CLI can communicate it to the user
- therefore, I've refactored the old functions into 1 function to make it clearer they do the same job, but with 2 different states
- also added some tests :D
refs 022a433e56
- noticed I missed adding debug info in one place (not that it's used, just for consistency)
- also made the SIGINT/TERM code slightly more readable
no-issue
- switch from `membersService.api.members.list` to using bookshelf `Member.findPage()` with the `{paid: true}` filter to avoid per-member queries (N+1) to decorate members with subscriptions and a heavy post-fetch filter via `contentGating`
- add concurrency to the Mailgun API requests in `bulk-email` service to reduce overall time submitting API requests
- add debug statements with timing output for easier measurements
no issue
For large numbers of members we're unable to perform queries for free/paid members in a reasonable time without indexes. Bulk delete is also very slow when looping through bookshelf models and having bookshelf manage deletion of child records, adding `ON DELETE CASCADE` to the foreign key indexes moves child deletion to the DB level and allows for performant bulk delete queries.
- migrations only run on mysql because sqlite does not support altering tables
- adds foreign key indexes and constraints with 'ON DELETE CASCADE' for members_labels, members_stripe_customers, and members_stripe_customers_subscriptions
refs #12100
For performance reasons we want to add foreign key and unique constraints
to the members_stripe_* tables so we can utilised cascading deletes and
joins across the tables when querying.
In order to do this we must first ensure that:
- There are no duplicate entries in the `subscription_id` or `customer_id` columns
- There are no orphaned rows in the subscription or customers tables
If the first is not true, the unique constraint will fail, and if the second is not true,
the foreign key constraint will fail.
As we are only adding the indexes to existing MySQL databases at this point, the
cleanup migrations will also only be done for existing MySQL databases too.
The migrations for removing orphaned rows splits the deletion into a `SELECT`
followed by a `WHERE IN` to avoid the database "optimising" the query into a
`JOIN` which ends up taking much longer due to the lack of indexes.
no-issue
We are in the process of creating migrations to add foreign key constraints
and cascading deletes to the members_stripe_* tables to make listing members
and deleting members faster. As well as the migrations we need to update the
database schema so that new installations have the correct indexes and constraints.
Co-authored-by: Kevin Ansfield <kevin@lookingsideways.co.uk>
