diff options
author | Yorick Peterse <yorickpeterse@gmail.com> | 2017-11-21 15:57:02 +0100 |
---|---|---|
committer | Yorick Peterse <yorickpeterse@gmail.com> | 2017-11-21 17:48:52 +0100 |
commit | aafe5c123ef2f64ddf44a5655a6fc63ec4e571da (patch) | |
tree | 7c3f360dffb3222f476cad243b7301cbf35a3f3b /db/migrate | |
parent | 9024875e1f81a3aab3c0879d33a4cea912ce833d (diff) | |
download | gitlab-ce-aafe5c123ef2f64ddf44a5655a6fc63ec4e571da.tar.gz |
Update composite pipelines index to include "id"fix-ci-pipelines-index
This updates the composite index on ci_pipelines (project_id, ref,
status) to also include the "id" column at the end. Adding this column
to the index drastically improves the performance of queries used for
getting the latest pipeline for a particular branch. For example, on
project dashboards we'll run a query like the following:
SELECT ci_pipelines.*
FROM ci_pipelines
WHERE ci_pipelines.project_id = 13083
AND ci_pipelines.ref = 'master'
AND ci_pipelines.status = 'success'
ORDER BY ci_pipelines.id DESC
LIMIT 1;
Limit (cost=0.43..58.88 rows=1 width=224) (actual time=26.956..26.956 rows=1 loops=1)
Buffers: shared hit=6544 dirtied=16
-> Index Scan Backward using ci_pipelines_pkey on ci_pipelines (cost=0.43..830922.89 rows=14216 width=224) (actual time=26.954..26.954 rows=1 loops=1)
Filter: ((project_id = 13083) AND ((ref)::text = 'master'::text) AND ((status)::text = 'success'::text))
Rows Removed by Filter: 6476
Buffers: shared hit=6544 dirtied=16
Planning time: 1.484 ms
Execution time: 27.000 ms
Because of the lack of "id" in the index we end up scanning over the
primary key index, then applying a filter to filter out any remaining
rows. The more pipelines a GitLab instance has the slower this will get.
By adding "id" to the mentioned composite index we can change the above
plan into the following:
Limit (cost=0.56..2.01 rows=1 width=224) (actual time=0.034..0.034 rows=1 loops=1)
Buffers: shared hit=5
-> Index Scan Backward using yorick_test on ci_pipelines (cost=0.56..16326.37 rows=11243 width=224) (actual time=0.033..0.033 rows=1 loops=1)
Index Cond: ((project_id = 13083) AND ((ref)::text = 'master'::text) AND ((status)::text = 'success'::text))
Buffers: shared hit=5
Planning time: 0.695 ms
Execution time: 0.061 ms
This in turn leads to a best-case improvement of roughly 25
milliseconds, give or take a millisecond or two.
Diffstat (limited to 'db/migrate')
-rw-r--r-- | db/migrate/20171121144800_ci_pipelines_index_on_project_id_ref_status_id.rb | 35 |
1 files changed, 35 insertions, 0 deletions
diff --git a/db/migrate/20171121144800_ci_pipelines_index_on_project_id_ref_status_id.rb b/db/migrate/20171121144800_ci_pipelines_index_on_project_id_ref_status_id.rb new file mode 100644 index 00000000000..5a8ae6e4b57 --- /dev/null +++ b/db/migrate/20171121144800_ci_pipelines_index_on_project_id_ref_status_id.rb @@ -0,0 +1,35 @@ +# See http://doc.gitlab.com/ce/development/migration_style_guide.html +# for more information on how to write migrations for GitLab. + +class CiPipelinesIndexOnProjectIdRefStatusId < ActiveRecord::Migration + include Gitlab::Database::MigrationHelpers + + # Set this constant to true if this migration requires downtime. + DOWNTIME = false + + disable_ddl_transaction! + + TABLE = :ci_pipelines + OLD_COLUMNS = %i[project_id ref status].freeze + NEW_COLUMNS = %i[project_id ref status id].freeze + + def up + unless index_exists?(TABLE, NEW_COLUMNS) + add_concurrent_index(TABLE, NEW_COLUMNS) + end + + if index_exists?(TABLE, OLD_COLUMNS) + remove_concurrent_index(TABLE, OLD_COLUMNS) + end + end + + def down + unless index_exists?(TABLE, OLD_COLUMNS) + add_concurrent_index(TABLE, OLD_COLUMNS) + end + + if index_exists?(TABLE, NEW_COLUMNS) + remove_concurrent_index(TABLE, NEW_COLUMNS) + end + end +end |