- There is now one true way to start Ghost - you create a server, and then call start with an express app
- We may well expand this again to improve testing pathways in future, but it will be done with a bit more clarity about expectations
- In the old boot the server wasn't started til we were ready
- In new boot, we start the server immediately and send the old started event
- Then, when we are ready to accept some traffic, we send a ready event
- At the moment, ready isn't quite sent at the right time:
- It _should_ be when we're ready to serve real traffic, not just send 503s
- This is after the URL generation has finished
- But this requires more refactoring work :(
- So for now we send when everything else is ready
- This really needs some tests
- Notify is a more familiar name e.g. systemd has the sd_notify system which this is similar to
- We're actually announcing the server started, it's not actually ready for traffic (will serve 503s)
- use theme.ready for loading themes instead of server.start and properly clean this up
- remove server.start and server.stop as they are no longer used (only server.start was used, and only for themes)
- we're moving away from the pattern of using global events like this as they are hard to reason about
- This commit removes the old boot process and any files that are no longer needed as a result
- Remove the duplicate event for triggering inactive themes to load
- Tidied up a few other bits
closes https://github.com/TryGhost/Ghost#12562
- From NodeJS v13 `error.errno` returns error code instead of a string. Because of that use friendly "port is already in use" message did not work anymore.
- Changed to use `error.code` which acts the same way as `error.errno` in older NodeJS versions.
- Currently the new boot process breaks the loading of inactive themes because the theme loader event isn't wired until after the server.start event is fired
- This is a WIP fix, there needs to be some bigger refactoring of the boot process once the old process is gone
Background:
- Ghosts existing boot process is split across multiple files, has affordances for outdated ways of running Ghost and is generally non-linear making it nigh-impossible to follow
- The web of dependencies that are loaded on boot are also impossible to unpick, which makes it really hard to decouple Ghost
- With 4.0 we want to introduce a new, linear, simpler, clearer way to boot up Ghost to unlock decoupling Ghost into much smaller pieces
This commit:
- adds a new ghost.js file which switches between boot mode with `node index` or `node index old` so that if we find bugs we can work around them this week
- Note: the old boot process will go away very soon, but ghost.js will remain as the interface between the command to start Ghost and the application code
- reworks the database migration process into a standalone utility, so that the DB is handled as one simple step of the boot process, decoupled from everything else
- is missing tests for this new db utility
- leaves a lot of work to do around loading core code, services, express apps in a sensible order, as work to fix this would start to break the old boot process
- doesn't use the new maintenance app because we aren't restarting the server here, instead we have the concept of a "core app" that starts in maintenance mode - need to think about how apps will be decoupled in the near future
refs #12568
- When the server is in maintenance mode (e.g. during migration) it needs to serve a 503 page. Previously this role was delegated to the "frontend" which tightly coupled server bootup and frontend.
- With a dedicated HTTP application serving up 503 server is no longer coupled to the frontend during maintenance
refs #12568
- Allowing express app (frontend) to be booted after the migrations
decouples server maintenance mode from a need to have the "frontend"
loaded beforehand
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
- 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
- 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
- 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()
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
- 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
- Represents that logging is shared across all parts of Ghost at present
* moved core/server/lib/common/logging to core/shared/logging
* updated logging path for generic imports
* updated migration and schema imports of logging
* updated tests and index logging import
* 🔥 removed logging from common module
* fixed tests
* moved `server/config` to `shared/config`
* updated config import paths in server to use shared
* updated config import paths in frontend to use shared
* updated config import paths in test to use shared
* updated config import paths in root to use shared
* trigger regression tests
* of course the rebase broke tests
- Use array destructuring
- Use @tryghost/errors
- Part of the big move towards decoupling, this gives visibility on what's being used where
- Biting off manageable chunks / fixing bits of code I'm refactoring for other reasons
- All var declarations are now const or let as per ES6
- All comma-separated lists / chained declarations are now one declaration per line
- This is for clarity/readability but also made running the var-to-const/let switch smoother
- ESLint rules updated to match
How this was done:
- npm install -g jscodeshift
- git clone https://github.com/cpojer/js-codemod.git
- git clone git@github.com:TryGhost/Ghost.git shallow-ghost
- cd shallow-ghost
- jscodeshift -t ../js-codemod/transforms/unchain-variables.js . -v=2
- jscodeshift -t ../js-codemod/transforms/no-vars.js . -v=2
- yarn
- yarn test
- yarn lint / fix various lint errors (almost all indent) by opening files and saving in vscode
- grunt test-regression
- sorted!
no issue
- Updated Test & linting packages
- Updated use of hasOwnProperty
- Using Object.prototype.hasOwnProperty instead (ref. eslint.org/docs/rules/no-prototype-builtins)
- Removed already defined built-in global variable Intl
- Applied `--fix` with lint command on `core/test` folder
- The rules were broken because some of them were made stricter for `eslint: recommended` ruleset (ref. https://eslint.org/docs/user-guide/migrating-to-6.0.0#eslint-recommended-changes)
- Removed redundant global variable declarations to pass linting
closes#10773
- The refactoring is a substitute for `urlService.utils` used previously throughout the codebase and now extracted into the separate module in Ghost-SDK
- Added url-utils stubbing utility for test suites
- Some tests had to be refactored to avoid double mocks (when url's are being reset inside of rested 'describe' groups)
closes#9802
- we have to trigger both functions within Ghost core, otherwise people who are using Ghost as NPM module have to call these functions
- this is internal logic
- plus: this logic is conditional, because of our internal maintenance flag
- make it backwards compatible in case you call announceServerStart or announceServerStopped twice
- tested with "Ghost as NPM module" and with the CLI on production
no issue
- if you have a bootstrap socket configured, but the socket does not appear, Ghost would crash
- cases:
- your server bootstraps (Ghost-CLI automatically starts Ghost via systemd)
- you add a bootstrap config in your Ghost config manually, but you don't start a net server
- added handling to catch errors and retry 3 times
refs #9742, refs https://github.com/TryGhost/Ghost-CLI/issues/759
- required a reordering of Ghost's bootstrap file, because:
- we have to ensure that no database queries are executed within Ghost during the migrations
- make 3 sections: check if db needs initialisation, bootstrap Ghost with minimal components (db/models, express apps, load settings+theme)
- create a new `migrator` utility, which tells you which state your db is in and offers an API to execute knex-migrator based on this state
- ensure we still detect an incompatible db: you connect your 2.0 blog with a 0.11 database
- enable maintenance mode if migrations are missing
- if the migration have failed, knex-migrator roll auto rollback
- you can automatically switch to 1.0 again
- added socket communication for the CLI
