## Summary
At the moment, our package generator creates all packages with the type
`shared-common`. This means that we cannot enforce boundaries between
server-side-only code and the browser, and vice-versa.
- [x] I started fixing `packages/core/*`
- [x] It took me to fixing `src/core/` type to be identified by the
`plugin` pattern (`public` and `server` directories) vs. a package
(either common, or single-scoped)
- [x] Unsurprisingly, this extended to packages importing core packages
hitting the boundaries eslint rules. And other packages importing the
latter.
- [x] Also a bunch of `common` logic that shouldn't be so _common_ 🙃
### For maintainers
- [x] This was checked for breaking API changes and was [labeled
appropriately](https://www.elastic.co/guide/en/kibana/master/contributing.html#kibana-release-notes-process)
---------
Co-authored-by: kibanamachine <42973632+kibanamachine@users.noreply.github.com>
## Summary
The spread operator is costly and put pressure on GC. It should be
avoided when possible, especially in loops.
This PR adapts a lot of `reduce` calls in the codebase to remove the
usages of the diabolic spread operator, when possible.
Note: the PR is not fully exhaustive. I focused on the server-side, as
we're more directly impacted than on browser-side code regarding
performances.
## Removing `...` usages in `kittens.reduce()`
For `reduce` loops, the spread operator can usually easily be replaced:
#### - setting a value on the accum object and returning it
#### BAD
```ts
return this.toArray().reduce(
(acc, renderer) => ({
...acc,
[renderer.name]: renderer,
}),
{} as Record<string, ExpressionRenderer>
);
```
#### GOOD
```ts
return this.toArray().reduce((acc, renderer) => {
acc[renderer.name] = renderer;
return acc;
}, {} as Record<string, ExpressionRenderer>);
```
#### - assigning values to the accum object and returning it
#### BAD
```ts
const allAggs: Record<string, any> = fieldAggRequests.reduce(
(aggs: Record<string, any>, fieldAggRequest: unknown | null) => {
return fieldAggRequest ? { ...aggs, ...(fieldAggRequest as Record<string, any>) } : aggs;
},
{}
);
```
#### GOOD
```ts
const allAggs = fieldAggRequests.reduce<Record<string, any>>(
(aggs: Record<string, any>, fieldAggRequest: unknown | null) => {
if (fieldAggRequest) {
Object.assign(aggs, fieldAggRequest);
}
return aggs;
},
{}
);
```
#### - pushing items to the accum list and returning it
#### BAD
```ts
const charsFound = charToArray.reduce(
(acc, char) => (value.includes(char) ? [...acc, char] : acc),
[] as string[]
);
```
#### GOOD
```ts
const charsFound = charToArray.reduce((acc, char) => {
if (value.includes(char)) {
acc.push(char);
}
return acc;
}, [] as string[]);
```
## Questions
#### Are you sure all the changes in this are strictly better for
runtime performances?
Yes, yes I am.
#### How much better?
Likely not much.
#### Are you planning on analyzing the perf gain?
Nope.
#### So why did you do it?
I got tired of seeing badly used spread operators in my team's owned
code, and I had some extra time during on-week, so I spent a few hours
adapting the usages in all our runtime/production codebase.
#### Was it fun?
Take your best guess.
---------
Co-authored-by: Kibana Machine <42973632+kibanamachine@users.noreply.github.com>
## Summary
This implements the behaviour for the `migrator` role introduced in
https://github.com/elastic/kibana/pull/151978.
The approach taken is kept simple:
1. Start Kibana as per usual, letting the `Root` and `Server` be
instantiated by the bootstrap logic
2. Inside the `Server`, just after the SO migrations are completed we
throw a special `CriticalError` (named `CriticalException` with code
`MigrationOnlyNode`)
3. The `Root` then handles this exception as it would handle any other
start up exception (like invalid config)
## To reviewers
* We can make this more sophisticated. I suspect this can quickly become
a tricky flow to implement/maintain since there are lots of moving parts
during "normal" Kibana startup. Using the same flow means our "migrator"
node will always behave the same way up until SOs have been migrated.
* Are we missing anything? Specifically, we want this "migrator" node to
exit with `0` since shutting down is expected. But is throwing during
`start` phase going to cover all cases we can think of?
Closes https://github.com/elastic/kibana/issues/150295
---------
Co-authored-by: kibanamachine <42973632+kibanamachine@users.noreply.github.com>
## Summary
Adds the `migrator` special role to the node roles config:
```yml
# 👇🏻 new
node.roles: ['migrator']
# or
node.roles: ['background_tasks', 'ui']
# or
node.roles: ['*'] # this one is slightly weird now because it actually excludes 'migrator' so it is not truly "all roles", but "all combinable roles"...
```
## How to test
Start Kibana locally and add `node.roles: ['migrator']` to the
`kibana.dev.yml`. Kibana should start normally and log:
```
[2023-02-23T12:08:54.123+01:00][INFO ][node] Kibana process configured with roles: [migrator]
```
Note: this role currently does not do anything. This PR just adds the
ability to configure it.
Partially addresses https://github.com/elastic/kibana/issues/150295
## Slight improvement to error messages
When specifying known, accepted values but combining with either
`migrator` or `*` you will get a message like:
```
[config validation of [node].roles]: wildcard ("*") cannot be used with other roles or specified more than once
```
---------
Co-authored-by: Luke Elmers <lukeelmers@gmail.com>
## Dearest Reviewers 👋
I've been working on this branch with @mistic and @tylersmalley and
we're really confident in these changes. Additionally, this changes code
in nearly every package in the repo so we don't plan to wait for reviews
to get in before merging this. If you'd like to have a concern
addressed, please feel free to leave a review, but assuming that nobody
raises a blocker in the next 24 hours we plan to merge this EOD pacific
tomorrow, 12/22.
We'll be paying close attention to any issues this causes after merging
and work on getting those fixed ASAP. 🚀
---
The operations team is not confident that we'll have the time to achieve
what we originally set out to accomplish by moving to Bazel with the
time and resources we have available. We have also bought ourselves some
headroom with improvements to babel-register, optimizer caching, and
typescript project structure.
In order to make sure we deliver packages as quickly as possible (many
teams really want them), with a usable and familiar developer
experience, this PR removes Bazel for building packages in favor of
using the same JIT transpilation we use for plugins.
Additionally, packages now use `kbn_references` (again, just copying the
dx from plugins to packages).
Because of the complex relationships between packages/plugins and in
order to prepare ourselves for automatic dependency detection tools we
plan to use in the future, this PR also introduces a "TS Project Linter"
which will validate that every tsconfig.json file meets a few
requirements:
1. the chain of base config files extended by each config includes
`tsconfig.base.json` and not `tsconfig.json`
1. the `include` config is used, and not `files`
2. the `exclude` config includes `target/**/*`
3. the `outDir` compiler option is specified as `target/types`
1. none of these compiler options are specified: `declaration`,
`declarationMap`, `emitDeclarationOnly`, `skipLibCheck`, `target`,
`paths`
4. all references to other packages/plugins use their pkg id, ie:
```js
// valid
{
"kbn_references": ["@kbn/core"]
}
// not valid
{
"kbn_references": [{ "path": "../../../src/core/tsconfig.json" }]
}
```
5. only packages/plugins which are imported somewhere in the ts code are
listed in `kbn_references`
This linter is not only validating all of the tsconfig.json files, but
it also will fix these config files to deal with just about any
violation that can be produced. Just run `node scripts/ts_project_linter
--fix` locally to apply these fixes, or let CI take care of
automatically fixing things and pushing the changes to your PR.
> **Example:** [`64e93e5`
(#146212)](64e93e5806)
When I merged main into my PR it included a change which removed the
`@kbn/core-injected-metadata-browser` package. After resolving the
conflicts I missed a few tsconfig files which included references to the
now removed package. The TS Project Linter identified that these
references were removed from the code and pushed a change to the PR to
remove them from the tsconfig.json files.
## No bazel? Does that mean no packages??
Nope! We're still doing packages but we're pretty sure now that we won't
be using Bazel to accomplish the 'distributed caching' and 'change-based
tasks' portions of the packages project.
This PR actually makes packages much easier to work with and will be
followed up with the bundling benefits described by the original
packages RFC. Then we'll work on documentation and advocacy for using
packages for any and all new code.
We're pretty confident that implementing distributed caching and
change-based tasks will be necessary in the future, but because of
recent improvements in the repo we think we can live without them for
**at least** a year.
## Wait, there are still BUILD.bazel files in the repo
Yes, there are still three webpack bundles which are built by Bazel: the
`@kbn/ui-shared-deps-npm` DLL, `@kbn/ui-shared-deps-src` externals, and
the `@kbn/monaco` workers. These three webpack bundles are still created
during bootstrap and remotely cached using bazel. The next phase of this
project is to figure out how to get the package bundling features
described in the RFC with the current optimizer, and we expect these
bundles to go away then. Until then any package that is used in those
three bundles still needs to have a BUILD.bazel file so that they can be
referenced by the remaining webpack builds.
Co-authored-by: kibanamachine <42973632+kibanamachine@users.noreply.github.com>
* [packages] add kibana.jsonc files
* auto-migrate to kibana.jsonc
* support interactive pkg id selection too
* remove old codeowners entry
* skip codeowners generation when .github/CODEOWNERS doesn't exist
* fall back to format validation if user is offline
* update question style
* [CI] Auto-commit changed files from 'node scripts/eslint --no-cache --fix'
Co-authored-by: kibanamachine <42973632+kibanamachine@users.noreply.github.com>
* refact(NA): apply root_input_dir=src to each already created pkg
* refact(NA): update package generator
* fix(NA): correctly use rootDir
* fix(NA): use root input dir on latest introduced pkgs for jsts_transpiler macro
* chore(NA): merge with main
Co-authored-by: Kibana Machine <42973632+kibanamachine@users.noreply.github.com>
