* upstream/master: (337 commits)
Update CHANGELOG for !4659
Center the header logo for all Devise emails
Add previews for all customized Devise emails
Customize the Devise `unlock_instructions` email
Customize the Devise `reset_password_instructions` email
Customize the Devise `password_change` emails
Use gitlab-git 10.2.0
Use Git cached counters on project show page
Fix indentation scss-lint errors
Added title attribute to enties in tree view Closes#18353
Banzai::Filter::ExternalLinkFilter use XPath
Reduce queries in IssueReferenceFilter
Use gitlab_git 10.1.4
Fixed ordering in Project.find_with_namespace
Fix images in emails
Banzai::Filter::UploadLinkFilter use XPath
Turn Group#owners into a has_many association
Make project_id nullable
CHANGELOG [ci skip]
CHANGELOG [ci skip]
...
Improve pipeline status in case that pipeline has no jobs
## What does this MR do?
This MR resolves problem with pipeline status when there are no build in pipeline.
This can happen when builds were skipped - for example - by using `only`/`except` keyword in `.gitlab-ci.yml`.
## What are the relevant issue numbers?
Closes#17977
See merge request !4403
Fair usage of Shared Runners
## What does this MR do?
Introduces a fair usage scheduler for shared runners.
It tries to assign builds to shared runner from projects that have the lowest number of builds currently running on shared runners.
**Example 1**:
```
We have following builds in queue:
build 1 for project 1
build 2 for project 1
build 3 for project 1
build 4 for project 2
build 5 for project 2
build 6 for project 3
With the new algorithm we will assign builds in following order:
- We choose build 1, because project 1 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We choose build 4, because project 2 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We choose build 6, because project 3 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We choose build 2, because project 1 as other it runs 1 build,
- We choose build 5, because project 2 runs 1 build, where project 1 runs 2 builds now,
- We choose build 3, because project 1 and runs 2 builds.
```
**Example 2**:
```
We have following builds in queue:
build 1 for project 1
build 2 for project 1
build 3 for project 1
build 4 for project 2
build 5 for project 2
build 6 for project 3
With the new algorithm we will assign builds in following order:
- We choose build 1, because project 1 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We finish build 1,
- We choose build 2, because project 1 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We choose build 4, because project 2 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We finish build 4,
- We choose build 5, because project 2 doesn't run currently any builds and has the lowest build number from projects that doesn't run builds,
- We choose build 6, because project 3 doesn't run currently any builds,
- We choose build 3, because project 1, 2 and 3 runs exactly one build now,
```
## Why was this MR needed?
Currently, we are scheduling builds using FIFO. This is catastrophic if there are projects that create a 100-300 jobs, this basically eats most of available shared runners.
## Performance
All this logic is implemented with the help of SQL queries, because this is the fastest way to process 1k-2k pending builds in queue.
It's not the fastest SQL query, because it sorts based on number of running_builds, and this forces to calculate a number of running builds for all dependent projects. However, since we have one/two shared runners that asks every few seconds for builds this should have minimal impact on DB performance.
```
explain analyze SELECT "ci_builds".* FROM "ci_builds" JOIN (SELECT "ci_builds"."gl_project_id", count(case when status = 'running' AND runner_id = (SELECT "ci_runners"."id" FROM "ci_runners" WHERE "ci_runners"."is_shared" = 't') then 1 end) as running_builds FROM "ci_builds" INNER JOIN "projects" ON "projects"."id" = "ci_builds"."gl_project_id" AND "projects"."pending_delete" = 'f' WHERE "ci_builds"."type" IN ('Ci::Build') AND "ci_builds"."status" IN ('running', 'pending') AND "projects"."builds_enabled" = 't' AND "projects"."shared_runners_enabled" = 't' GROUP BY "ci_builds"."gl_project_id") AS projects ON ci_builds.gl_project_id=projects.gl_project_id WHERE "ci_builds"."type" IN ('Ci::Build') AND "ci_builds"."status" = 'pending' AND "ci_builds"."runner_id" IS NULL ORDER BY projects.running_builds ASC, ci_builds.id ASC;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
-----------
Sort (cost=64777.28..64777.29 rows=1 width=1010) (actual time=301.794..302.535 rows=1537 loops=1)
Sort Key: (count(CASE WHEN (((public.ci_builds.status)::text = 'running'::text) AND (public.ci_builds.runner_id = $0)) THEN 1 ELSE NULL::integer END)), public.ci
_builds.id
Sort Method: quicksort Memory: 1423kB
-> Nested Loop (cost=63279.78..64777.27 rows=1 width=1010) (actual time=66.384..298.724 rows=1537 loops=1)
-> HashAggregate (cost=63177.15..63177.30 rows=15 width=15) (actual time=65.641..65.851 rows=187 loops=1)
InitPlan 1 (returns $0)
-> Seq Scan on ci_runners (cost=0.00..26963.66 rows=1 width=4) (actual time=1.145..34.381 rows=1 loops=1)
Filter: is_shared
Rows Removed by Filter: 6965
-> Nested Loop (cost=0.00..36186.34 rows=2715 width=15) (actual time=0.065..29.717 rows=1710 loops=1)
-> Index Scan using index_ci_builds_on_status on ci_builds (cost=0.00..8913.95 rows=3577 width=15) (actual time=0.051..12.012 rows=2583 loops
=1)
Index Cond: ((status)::text = ANY ('{running,pending}'::text[]))
Filter: ((type)::text = 'Ci::Build'::text)
Rows Removed by Filter: 1219
-> Index Scan using projects_pkey on projects (cost=0.00..7.61 rows=1 width=4) (actual time=0.003..0.004 rows=1 loops=2583)
Index Cond: (id = public.ci_builds.gl_project_id)
Filter: ((NOT pending_delete) AND builds_enabled AND shared_runners_enabled)
Rows Removed by Filter: 0
-> Bitmap Heap Scan on ci_builds (cost=102.63..106.64 rows=1 width=1002) (actual time=1.216..1.231 rows=8 loops=187)
Recheck Cond: ((gl_project_id = public.ci_builds.gl_project_id) AND ((status)::text = 'pending'::text))
Filter: ((runner_id IS NULL) AND ((type)::text = 'Ci::Build'::text))
-> BitmapAnd (cost=102.63..102.63 rows=1 width=0) (actual time=1.201..1.201 rows=0 loops=187)
-> Bitmap Index Scan on index_ci_builds_on_gl_project_id (cost=0.00..10.52 rows=241 width=0) (actual time=0.406..0.406 rows=1944 loops=187)
Index Cond: (gl_project_id = public.ci_builds.gl_project_id)
-> Bitmap Index Scan on index_ci_builds_on_status (cost=0.00..91.78 rows=3089 width=0) (actual time=0.652..0.652 rows=3362 loops=187)
Index Cond: ((status)::text = 'pending'::text)
Total runtime: 303.832 ms
```
## Specific runners
It doesn't affect the specific runners which still serve builds FIFO.
@stanhu @markpundsack @yorickpeterse What do you think?
See merge request !4634
* master: (198 commits)
Set inverse_of for Project/Services relation
Fix admin hooks spec
Prevent default disabled buttons and links.
Add index on `requested_at` to the `members` table
Rearrange order of tabs
Fix admin active tab tests
Show created_at in table column
Nest li elements directly under ul
Move builds tab to admin overview
Add monitoring link with subtabs
Add sub links to overview
Add counter for abuse reports
Remove admin layout-nav counters
Move admin nav to horizontal layout nav
Eager load project relations in IssueParser
Use validate and required for environment and project
Award Emoji can't be awarded on system notes backend
Get rid of Gitlab::ShellEnv
Update CHANGELOG.
Fix project star tooltip on the fly.
...
Conflicts:
app/services/ci/create_builds_service.rb
* master: (147 commits)
Update CHANGELOG
Remove deprecated issues_tracker and issues_tracker_id from project
Schema doesn’t reflect the changes of the last 3 migrations
Revert CHANGELOG
Also rename "find" in the specs
Change to new Notes styleguide
Add guide on changing a document's location
Change logs.md location in README
Move logs/logs.md to administration/logs.md
Make "four phase test"
Only show branches for revert / cherry-pick
Instrument all Banzai::ReferenceParser classes
Removed old comment from update_column_in_batches
Update columns in batches until no rows are left
Remove counters from Pipeline navigation
Handle NULL migration errors in migration helpers
Fix typo causing related branches to Error 500
Improved SVG sanitizer specs to include smoke tests for clean.
Refactored SVG sanitizer
Added SVG sanitizer fix to the changelog
...
* master: (538 commits)
Fix broken URI joining for `teamcity_url` with suffixes
Factorize duplicated code into a method in BambooService and update specs
Fix broken URI joining for `bamboo_url` with suffixes
Honor credentials on calling Bamboo CI trigger
Update CHANGELOG
Use Issue.visible_to_user in Notes.search to avoid query duplication
Project members with guest role can't access confidential issues
Allow users to create confidential issues in private projects
Update CHANGELOG
Remove deprecated issues_tracker and issues_tracker_id from project
Schema doesn’t reflect the changes of the last 3 migrations
Apply reviewer notes: update CHANGELOG, adjust code formatting
Move issue rendering tests into separate contexts
Move change description to proper release and fix typo
Add more information into RSS fead for issues
Revert CHANGELOG
Also rename "find" in the specs
Change to new Notes styleguide
Add guide on changing a document's location
Change logs.md location in README
...
Conflicts:
app/services/ci/create_builds_service.rb
app/services/ci/create_pipeline_service.rb
app/services/create_commit_builds_service.rb
spec/models/ci/commit_spec.rb
spec/services/ci/create_builds_service_spec.rb
spec/services/create_commit_builds_service_spec.rb
- Remove all view related methods from Ci::Build and CommitStatus
- Remove unused Ci::Commit and Ci::Build methods
- Use polymorphism to render different types of CommitStatus
Lin Jen-Shin
Rémy Coutable
Stan Hu
Grzegorz Bizon
Kamil Trzcinski
Jason Roehm