refs https://github.com/TryGhost/Toolbox/issues/152
refs 5bea089dfe
refs 16ad5f73c4
- The refactor is heavily inspired by the referenced commit from @ErisDS
- This refactor is meant to serve as a template for further refactors and unification of the abstraction over "request". Should allow us to be more agnostic towards the library that's powering the request thing. For example, mock-express style of request handling could substitute or get substututed easily if all tests are behind similar "get(Authenticated)Agent" interface
refs https://github.com/TryGhost/Toolbox/issues/152
- This is a continuation of an experiment to switch over to serverless boot in regression tests. Just a proofe of concept that authentication scenarios would also work and the expectations don't collapse unexpectedly.
- Nothing too crazy so far, easy and straight forward substitution of the config "url" with an express app passed to the supertest agent
refs https://github.com/TryGhost/Toolbox/issues/152
- Being able to set up test suites without blocking a port opens a door to new ways to run tests - for example this has been one of the blockers for running mocha tests in parallel
- Additional benefit is lighter statrup, which reducec the test execution time slightly. Doesn't seem like much but these things stack up!
no issue
- we've seen some instances of SQLite saying "database disk image is malformed"
- I think this happens because we copy the file whilst we are still connected to
the old DB
- this commit destroys the connection before copying the clean file to the live
DB file
refs https://github.com/TryGhost/Toolbox/issues/136
- we nuke and reinitialize the DB many times between tests
- this forms a good portion of the time taken and we shouldn't be spending
so much time on just resetting the DB back to a known state
- this commit switches out the knex-migrator reset + init calls for SQLite to
a function which keeps a clean copy of the DB and copies the file back
when we "reset"
- for MySQL, existing functionality is kept
- this massively speeds up tests because it saves ~700ms+ for every reset
- whilst this change seems to work, it's just the start so there's a
lot more refactoring needed. this change is currently gated to CI until
it's deemed safe/sane enough to run on local machines without blowing
up
refs https://github.com/TryGhost/Toolbox/issues/152
- Have skimmed through the test suites in hopes to find some quick performance wins to bring the runtime speed closer to the one in "main". Haven't been successful to identify major wins, cleaned up a couple of small bits.
- We'll have to live with a tradeoff between maintainability/unified boot VS cost of mainitaining a fake boot process. Imo extra couple seconds of runtime is worth it.
refs 3c7a8dead4
- The tests needed adjustments with the native boot mechanism.
- The number of returned posts changed in the test resutls because duing native boot we also insert fixtures which add to the number of initial posts
- The vhost regression suite is still failing and I had no strength to figure out why. The redirect it fails with makes no sense, the clue here is that the test doesn't fail when running in isolation, so probably has to do with some leftover overrides from the previous test cases.
refs 3c7a8dead4
- The boot process has been using an asyc method to load the routes file, which is the case now for these tests since the switch to raw boot method instead of mimicking it manually
refs 3c7a8dead4
- Simplifies the state initialization code significantly and reuses native boot mechanism instead of mimicking it (it was a headache to maintain with all the internal services moving around)
refs ec28478435
- the method signature changed in
9fa8800b9d
and the usage wasn't updated in Ghost
- this commit updates the tests to reflect this internal change
refs https://github.com/TryGhost/Team/issues/960
- Character like "%%" or "%80" would crash our current url escaping behavior. We consider they aren't valid URLs as the percentages haven't been properly escaped.
refs https://github.com/TryGhost/Team/issues/1200
- The leading/trailing whitespaces are trimmed by `new URL()` but are considered invalid in metascraper. Trimming solves this edge case.
refs: TryGhost/Toolbox#147
* Replaces all references to isIgnitionError with isGhostError
* Switches use of GhostError to InternalServerError - as GhostError is no longer public
There are places where InternalServerError is not the valid error, and new errors should be added to the @tryghost/errors package to ensure that we can use semantically correct errors in those cases.
refs https://github.com/TryGhost/Team/issues/1237#issuecomment-981770688
- API key names for external services now follow a standard pattern
- top-level key of the service name
- public/private and read/write perms inside the name, eg. `publicReadOnlyApiKey`
- updated test to match expected API key name
closes https://github.com/TryGhost/Toolbox/issues/148
- These regression tests introduce very little additional value in exchange for an expensive time to run them. Because we mostly care about stable support for the latest stable API version, the older API version test for "internal" API can go away
- In case there are bugs found we do care about in the v2/v3 APIs we can always revert some of these tests.
https://github.com/TryGhost/Toolbox/issues/140
- This test was bloating the regression/site suite and was using hacks (calling the Admin API) to create a custom redirects state
- It suits way better in e2e frontend test suite with less hacky approach to start the Ghsot instance - using custom routes file path to initialize the instance
https://github.com/TryGhost/Toolbox/issues/140
- Allows to fully start an instance with a custom routes.yaml file without a need to do workarounds - e.g. we used to call an internal Admin API from a regression test suite, which is not a good practice
- Providing "routesFilePath" to startGhost method copies that file to the default settings location and start the instance with that configuration. No need to do API calls or check if the routing service "isFinished"
refs https://github.com/TryGhost/Toolbox/issues/140
- Same/similar RSS tests were present in regression and e2e tests suites. It made sense to move missing cases from regression to e2e. This saves us time bootstraping db state in multiple places and keeps all test cases regarding single feature in same place
refs https://github.com/TryGhost/Toolbox/issues/139
- With ec2aed5ce8 the DynamiRedirectsManager has reached 100% test coverage and most of the tests present in the removed regression suite have been ported to unit tests
- No need to keep slow tests around! :)
refs https://github.com/TryGhost/Toolbox/issues/139
- The regression test suite for redirects functionality for way too big. And each restart was causing massive overhead. It's enough to have a single exhaustive test using multiple input files
- The tests testing API endpoints should've been e2e tests to start with
- The rest is covered in the unit tests for redirects api service
refs https://github.com/TryGhost/Toolbox/issues/139
- Having tight coupling with backup file path calculation for redirects makes it extremely hard to test. In addition, having it injected will make it easier to swap this dependency to the mechanism similar to one used for routes files
refs 91efa4605c
- Referenced commit introduced a double json-stringification to uploaded redirects.json files.
- The endpoint has no stability index of any sort and is meant to be dropped in Ghost v5. It's best to rework the redirects to the yaml format as descirbe here - https://ghost.org/docs/tutorials/implementing-redirects/#file-structure
- moving this middleware because we're about to add a second piece of middleware
- it's easier to see what we have when each middleware is in its own file rather than in one big middleware.js file
no issue
- this commit adds a counter for the number of boots we do in tests
- which therefore allows us to calculate the average boot time we
experience
- only useful for debugging test performance
refs https://github.com/TryGhost/Toolbox/issues/139
- The v2 and v3 redirects APIs are unofficial and should not be used by anyone in production. There's no good reason to maintain expensive to run test suites for old unofficial APIs.
- The test cases in canary suite covers the functionality of redirects enough to be sure they work as expeted
refs https://github.com/TryGhost/Toolbox/issues/138
- These tests still rely on the frontend to be present. Needs further investigation to remove "frontend: true" flag - it slows down test runs!
refs https://github.com/TryGhost/Toolbox/issues/138
- Final batch of the refactor to async/await syntax. Doing these refactors before modifying "testUtils.startGhost" everywhere to boot only with the backend
refs https://github.com/TryGhost/Toolbox/issues/138
- This is a continuation of a bigger refactor to use async/await syntax before migrating "startGhost" methods to only use backend boot
- Removed a little bit of dead code (like admin user creation) which should speed up test execution too!
- Refactored user variables to be declared closer to their usecases instead of being high up in a global scope - variables shoul not live that far apart from the code that uses them
refs https://github.com/TryGhost/Toolbox/issues/138
- First batch of the refactor to async/await syntax. Next one will cover the rest. Doing these refactors before modifying "testUtils.startGhost" everywhere to boot only with the backend
refs https://github.com/TryGhost/Toolbox/issues/138
- There is no good reason to keep this extra variable around just call "stop" in couple very specific cases. Even for those cases, there's `testUtils.stopGhost` method which achieves the same without additional variable to track.
refs https://github.com/TryGhost/Toolbox/issues/138
- Having the "ghost" alias only added cognitive load when reading through the test code and didn't provide any additional value. Removed the pattern to keep things simpler and more explicit
refs https://github.com/TryGhost/Toolbox/issues/138
- Using asycn/await syntax is way more readable and allows to identify further reusable patterns in test initialization. This refactor also served as an exploreation around how the code looks like at this point
refs: 2af9e2e12
- This new HTMLErrorRenderer is borrowed heavily from finalHandler
- This is the module that express uses to render errors if there is no custom errorhandler
- It just renders a really simple html page wrapping err.stack in a <pre>
- This results in a nicely formatted, but unstyled error page
- I also updated BasicErrorRenderer to use the same res.statusCode + err.stack pattern rather than err.message
Note: This error renderer is _only_ used for renderering errors on the `/ghost/` route
- In almost all cases, errors here are rendered by Ember
- The only error that can be rendered here is a missing template error see: 2af9e2e12
- Card asset reloading was incorrectly only happening if the API version changed 🙈
- In addition, having an init function was redundant, as theme activation happens on boot
- This meant that the card assets were being generated twice on boot
- Instead, we now only generate them on theme activation, which covers the boot case and simplifies all the logic
- Currently it's assumed that public files are 100% static
- With card assets, we're using it for files that are partially static, but can change between reboots and theme changes
- We already have a system for managing cache busting across theme changes and restarts - the ?v= key that is added via the asset helper
- This was already in place and used, but servePublicFile's internal cache didn't honor this key, and cached for the lifetime of boot
- This small change means that if a ?v= query param is present on a request for a public file, that we pay attention to it. Else we cache as before
refs https://github.com/TryGhost/Toolbox/issues/135
- Looking closer into the reason why the test was failing without "forceStart" revealed that the server only start was overoptimized - "initServicesForFrontend" should be a part of a backend as those are generic theme services.
no issue
- Having rewire here doens't do any difference and should not be used if absolutely needed. Usually using rewire gives a code "smell" so there's some sort of coupling that's going on and probably has to be addressed first
refs https://github.com/TryGhost/Toolbox/issues/135
- The reason the test **settings** test was failing when the force start flag was removed in the **custom themes** was the bridge! The bridge was trying to execute function on the frontend when the boot was done without initializing the frontend. The setting test was changing locale and the timezone which triggered events calling up on frontend components - we clearly don't want to do this when the instance is booted without the frontend
- To make event initialization conditional moved it to the "init". This way the event listeners are only set up when we boot with the "frontend" flag set to true
refs https://github.com/TryGhost/Toolbox/issues/135
- Without sensible defaults the web app was not initializing either the backend nor the frontned parts of the application. Fixed the defaults so the problem doesn't happen again and optimized mock-express-style initialization to only initialize the frontend routing
refs https://github.com/TryGhost/Toolbox/issues/135
- This optimization is expected to play a role in more consistent "backend-only" boot where the previous test state might have left over a different theme version which might cause in unwanted URL Services reainitializations.
- What has been happening here is the themes.test.js suite was uploading a theme with a v4 api and when the users api test suite loaded up it switched back to a default v2 theme, which caused routing reinitialization
- The root problem here is the themese suite is leaving a mess behind so a "restartModeGhostStart" is not really possible anymore - this should be cleaned up separately
refs https://github.com/TryGhost/Toolbox/issues/135
- Allows to turn off overwriting urls/resources JSON file caches on testing environment. This is needed to have predictable state when running multiple test suites that stop the Ghost process and try to persiste URL cache.
refs https://github.com/TryGhost/Toolbox/issues/135
- This extracts the file storage knowledge out of the URL Service an allows to have optional features based on the environment - for example turning off writing cache for when running tests
refs https://github.com/TryGhost/Toolbox/issues/135
- The global default should stay the same as it used to be and we can introduce an override for "withFrontend:false" on casa-by-case or area-by-area bases
refs https://github.com/TryGhost/Toolbox/issues/135
- Going though local utils allows to have the "withFrontend: false" flag applied only to the e2e-api test suites. This way we can gradually introduce the no-frontend change across all test suites
refs https://github.com/TryGhost/Toolbox/issues/135
- Having the proxy in local utils allows to disable frontend gradually in test suites instead of all at once (the latter approach didn't work out as too many other suites still depend on full frontend initialization)
refs https://github.com/TryGhost/Toolbox/issues/135
- When running e2e-API test in most cases there's no need to boot Ghost instance with full frontend. This should improve the boot time which should reflect on the speed of running test suites
- The tests where the "forceStart" and "withFrontend" are used together indicate that there's still some work to do to fully separate frontend/backend boot line. The force start is also unnecessary, but was needed to reinitialize all services properly - should be investigated!
refs https://github.com/TryGhost/Toolbox/issues/135
- To be able to reliably start ghost instance without a frontend the process needs access to urls/resources caches
- Storing the configuration in "paths" for now as there's no better place for it untill we are able to mock the content folder in pre-boot
refs https://github.com/TryGhost/Toolbox/issues/135
- These fixture had to be updated due to hardcoded ids introduced in the fixtures.json for test
- This allows to boot up a Ghost instance without recalculating urls for most common database states
refs https://github.com/TryGhost/Toolbox/issues/135
- These fixtures needs id's to be able to form a stable urls and resources cache that would be used across all e2e tests when booting without a frontend
refs https://github.com/TryGhost/Toolbox/issues/135
- This way the test is performed on a "full" set of posts that will become unified within all e2e tests for performance optimization
closes: https://github.com/TryGhost/Ghost/issues/13739
- Ghost cannot write to the core folder in correctly configured production installations
- Built assets therefore need to be written to the content directory
- Ghost does not overwrite anything in the content folder as part of an upgrade, therefore static files that are provided by Ghost
must still live inside /core
- So as a result, we now have core/frontend/public and content/public
- These two things are meant to improve performance at the cost of reliability.
- Perfect for testing, however I think they make a minimal impact on modern SSDs :(
- Still worth a shot to see if it helps with CI
refs: https://github.com/TryGhost/Toolbox/issues/130
The API version stays at v2 unless we stub the getFrontendApiVersion method. But stubbing the method doesn't get picked up unless we actually restart Ghost.
TODO: Maybe change the default here so we don't need to restart Ghost just to test the current version's API
- These are simple functions that get data from config in a specific format
- They are also used by the topmost part of the application
- Config helpers seems like a reasonable fit to get them out of the web folder
- Functions have also been renamed to try to get them to make more sense
refs https://github.com/TryGhost/Team/issues/1217
- moved top-level `tenorApiKey` to `tenor:apiKey` and added `tenor:contentFilter`
- added base config to `defaults.json`
- updated `public-config.js` and API output serializer to use the new top-level `tenor` key
https://github.com/TryGhost/Toolbox/issues/130
The transaction no longer commits in the promise chain, which wasn't
valid logic for a transaction, since it is commited automatically when
the promise chain resolves, and rollsback automatically when the
promise chain rejects.
This makes code which fails during the transaction error in the right
place, instead of getting stuck here. (Especially good for writing
tests).
The tests for this code can now live in the integration folder.
- This is a minor bugbare, but it will affect some configuration I'm about to do for c8
- I've been wanting to do it for ages, middleware is plural all on it's own so it's an odd affectation in our codebase
- This also only exists in 2 places, everywhere else we use "middleware"
- Sadly it did result in a lot of churn as I did a full find and replace, but consistency is king!
refs https://github.com/TryGhost/Toolbox/issues/116
- Allows to have deterministic routing on each start of the instance - can use cache when starting tests
- To make use of the new fixtures need to place them in the content/data
folder when the test suite is initialized, then the cache detection will
kick in and boot up the UrlService in a constant time intead of having
to check "urlServiceUtils.isFinished"
- this keeps production and test fixtures separate, so that changing the prod fixtures doesn't change the shape of our tests.
- we may still want to test that the production fixtures do what we expect, but that can be handled in a separate integration test, by specifically setting the fixture path
refs: https://github.com/TryGhost/Toolbox/issues/133
- instead of just a collection of utils, we now have a class that manages fixtures
- this should allow us to change the path to fixtures, e.g. between prod/dev and test, so that different fixtures can be loaded by default
- also makes it easier to test the fixture manager code itself
refs 042618fe93
- There's no longer "getValue" method passed around in UrlService, so the were unnecessary and did a wrong thing here :) Yey, less code!
refs https://github.com/TryGhost/Toolbox/issues/127
- Passing around whole instance of a frontend router was an overkill when there are only 3 static pieces of information that needed to be loaded. Extracting the router out makes the UrlGenerator way more readable, tests slimer, and the memory footpring of the process should be slightly lighter
- The toString overloading didn't make sense at the time of this refactor, maybe if there's a concrete usecase we could resurect it in a form of passing in a router's name or something.
refs https://github.com/TryGhost/Toolbox/issues/127
- This is an effor t to define a precise set of data needed for the UrlGenerator to function, which should help with decoupling it from the frontend routes
- This is almost the last piece to free us up from the massive "router" object that has been passed around
refs https://github.com/TryGhost/Toolbox/issues/127
- This is an effor t to define a precise set of data needed for the UrlGenerator to function, which should help with decoupling it from the frontend routes
refs https://github.com/TryGhost/Toolbox/issues/127
- This is an effor t to define a precise set of data needed for the UrlGenerator to function, which should help with decoupling it from the frontend routes
refs https://github.com/TryGhost/Team/issues/1217
- add `tenorApiKey` to `publicConfig.config()
- update canary config endpoint output serializer to include `tenorApiKey` when the `gifsCard` labs flag is enabled
