it will decide the method for disable statement_timeout upon
per transaction or per session, based on how it's called.
When calling with a block, block will be executed and it will use
session based statement_timeout, otherwise will default to existing
behavior.
By default statement_timeout will only be enabled during transaction
lifetime, therefore not leaking outside of it.
With `transaction: false` it will set for entire session, but requires
a block to passed. It yields control and cleans up session after block
finishes, also preventing leaking outside of it.
This changes the BackgroundMigration worker so it checks for the health
of the DB before performing a background migration. This in turn allows
us to reduce the minimum interval, without having to worry about blowing
things up if we schedule too many migrations.
In this setup, the BackgroundMigration worker will reschedule jobs as
long as the database is considered to be in an unhealthy state. Once the
database has recovered, the migration can be performed.
To determine if the database is in a healthy state, we look at the
replication lag of any replication slots defined on the primary. If the
lag is deemed to great (100 MB by default) for too many slots, the
migration is rescheduled for a later point in time.
The health checking code is hidden behind a feature flag, allowing us to
disable it if necessary.
This works the same way as change_column_type_using_background_migration, but
for renaming a column. It takes a table, not a relation, to match its concurrent
counterpart.
Also, generalise the cleanup migrations to reduce code duplication.
In Arel 7.0.0 (Arel 7.1.4 is used in Rails 5.0) the `engine` parameter
of `Arel::UpdateManager#initializer` was removed.
This commit makes the gitlab database helpers work both in rails 4 and
rails 5.
Fixes errors like this one:
```
1) Gitlab::Database::MigrationHelpers#update_column_in_batches when running outside of a transaction updates all the rows in a table
Failure/Error:
update_arel = Arel::UpdateManager.new(ActiveRecord::Base)
.table(table)
.set([[table[column], value]])
.where(table[:id].gteq(start_id))
ArgumentError:
wrong number of arguments (given 1, expected 0)
# ./lib/gitlab/database/migration_helpers.rb:317:in `new'
# ./lib/gitlab/database/migration_helpers.rb:317:in `block in update_column_in_batches'
# ./lib/gitlab/database/migration_helpers.rb:307:in `loop'
# ./lib/gitlab/database/migration_helpers.rb:307:in `update_column_in_batches'
# ./spec/lib/gitlab/database/migration_helpers_spec.rb:367:in `block (4 levels) in <top (required)>'
```
[10.6] Prevent notes on confidential issues from being sent to chat
See merge request gitlab/gitlabhq!2366
# Conflicts:
# app/helpers/services_helper.rb
Index creation does not have an effect if the index is present already.
Index removal does not have an affect if the index is not present.
This helps to avoid patterns like this in migrations:
```
if index_exists?(...)
remove_concurrent_index(...)
end
```
The concurrency issue originates from inserts on
`user_interacted_projects` from the app while running the post-deploy
migration.
This change comes with a strategy to lock the table while removing
duplicates and creating the unique index (and similar for FK
constraints).
Also, we'll have a non-unique index until the post-deploy migration is
finished to speed up queries during that time.
Closes#44205.
Prior to this commit we would essentially update all rows in a table,
even those where the source column (e.g. `issues.closed_at`) was NULL.
This in turn could lead to statement timeouts when using the default
batch size of 10 000 rows per job.
To work around this we don't schedule jobs for rows where the source
value is NULL. We also don't update rows where the source column is NULL
(as an extra precaution) or the target column already has a non-NULL
value. Using this approach it should be possible to update 10 000 rows
in the "issues" table in about 7.5 - 8 seconds.
Fixes https://gitlab.com/gitlab-org/gitlab-ce/issues/42158
This adds a minimum interval to BackgroundMigrationWorker, ensuring
background migrations of the same class only run once every 5 minutes.
This prevents a thundering herd problem where scheduled migrations all
run at once due to their delays having been expired (e.g. as the result
of a queue being paused for a long time).
If a job was recently executed it's rescheduled with a delay that equals
the remaining time of the job's lease. This means that if the lease
expires in two minutes we only need to wait two minutes, instead of
five.
Fixes https://gitlab.com/gitlab-org/gitlab-ce/issues/41624
In a previous attempt (rolled back in
https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/16021) we tried
to migrate `issues.closed_at` from timestamp to timestamptz using a
regular migration. This has a bad impact on GitLab.com and as such was
rolled back.
This commit re-implements the original migrations using generic
background migrations, allowing us to still migrate the data in a single
release but without a negative impact on availability.
To ensure the database schema is up to date the background migrations
are performed inline in development and test environments. We also make
sure to not migrate that that doesn't need migrating in the first place
or has already been migrated.
This adds a bunch of checks to migrations that may create or drop
triggers. Dropping triggers/functions is done using "IF EXISTS" so we
don't throw an error if the object in question has already been dropped.
We now also raise a custom error (message) when the user does not have
TRIGGER privileges. This should prevent the schema from entering an
inconsistent state while also providing the user with enough information
on how to solve the problem.
The recommendation of using SUPERUSER permissions is a bit extreme but
we require this anyway (Omnibus also configures users with this
permission).
Fixes https://gitlab.com/gitlab-org/gitlab-ce/issues/36633
Due to a missing `on_delete: :cascade`, users would hit the error that
looked like:
```
PG::ForeignKeyViolation: ERROR: update or delete on table "protected_tags"
violates foreign key constraint "fk_rails_f7dfda8c51" on table
"protected_tag_create_access_levels" DETAIL: Key (id)=(1385) is still
referenced from table "protected_tag_create_access_levels". : DELETE FROM
"protected_tags" WHERE "protected_tags"."id" = 1385
```
Closes#36013
* master: (56 commits)
File view buttons
Don't reset the session when the example failed, because we need capybara-screenshot to have access to it
Resolve "MR comment + system note highlight don't have the same width"
Add feature spec for dashboard state filter tabs
Wording of Mysql support.
a new feature checklist and more elaborate documentation requirements
Filter archived project in API v3 only if param present
Revert to using links instead of buttons in Issuable Index tabs.
Do not run the codeclimate job on docs-only changes
Only show gray footer space if environment actions exist
Migrate Gitlab::Git::Blob.find to Gitaly
Backport filtered search lazy token consistent state fix
Add a comment explaining how the branch clean up happens
Fix Github::Representation::PullRequest#source_branch_exists?
Add CHANGELOG
Fix GitHub importer performance on branch existence check
Rebuild the dynamic path before validating it
Rename stage ref migration specs to match a class name
Enable Style/DotPosition Rubocop 👮
Revert "Merge branch 'winh-merge-request-related-issues' into 'master'"
...
Conflicts:
db/post_migrate/20170526185921_migrate_build_stage_reference.rb
When using update_column_in_batches the upper limit on the batch size is
now 1000. This ensures that for very large tables we don't lock tens of
thousands of rows during the update. This in turn should reduce the
likelyhood of running into deadlocks.
MySQL doesn't allow us to create a trigger for a column that doesn't
exist yet. Failing with this error:
```
Mysql2::Error: Unknown column 'build_events' in 'NEW': CREATE TRIGGER trigger_6a80c097c862_insert
BEFORE INSERT
ON `services`
FOR EACH ROW
SET NEW.`build_events` = NEW.`job_events`
```
Creating the new column before creating the trigger avoids this.
Starting with GitLab 9.1.0 we will no longer allow downtime migrations
unless absolutely necessary. This commit updates the various developer
guides and adds code that is necessary to make zero downtime migrations
less painful.
This was initially not implemented simply because I forgot about the
size limit of constraint names in PostgreSQL (63 bytes). Using the old
technique we can't add foreign keys for certain tables. For example,
adding a foreign key on
protected_branch_merge_access_levels.protected_branch_id would lead to
the following key name:
fk_protected_branch_merge_access_levels_protected_branches_protected_branch_id
This key is 78 bytes long, thus violating the PostgreSQL size
requirements.
The hashing strategy is copied from Rails' foreign_key_name() method,
which unfortunately is private and subject to change without notice.
This method allows one to create foreign keys without blocking access to
the source table, but only on PostgreSQL.
When creating a regular foreign key the "ALTER TABLE" statement used for
this won't return until all data has been validated. This statement in
turn will acquire a lock on the source table. As a result this lock can
be held for quite a long amount of time, depending on the number of rows
and system load.
By breaking up the foreign key creation process in two steps (creation,
and validation) we can reduce the amount of locking to a minimum.
Locking is still necessary for the "ALTER TABLE" statement that adds the
constraint, but this is a fast process and so will only block access for
a few milliseconds.
Long-running migrations may take more than the timeout allowed by
the database. Disable the session's statement timeout to ensure
migrations don't get killed prematurely.
This changes update_column_in_batches to ensure it always updates all
rows now. These changes also allow for an extra SELECT query to be
removed, nor does it use the row count for determining offsets and the
likes; instead it's only used to determine the batch size.
This ensures that whatever locks are acquired aren't held onto until the
end of the transaction (= after _all_ rows have been updated). Timing
wise there's also no difference between using a transaction and not
using one.
By passing a block to update_column_in_batches() one can now customize
the queries executed. This in turn can be used to only update a specific
set of rows instead of simply all the rows in the table.
Instead of updating a fixed number of rows (based on the amount of rows
available at the start of the update) the method
"update_column_in_batches" will now continue updating rows until it runs
out of rows to process.
For a table with a high rate of inserts this may result in the migration
taking quite some time. However, the alternative is not all rows being
updated or the "change_column_null" method raising an error due to there
being NULL values.
Ryan Hefner
Gabriel Mazetto
Yorick Peterse
Sean McGivern
blackst0ne
Douwe Maan
Bob Van Landuyt
Andreas Brandl
🙈 jacopo beschi 🙉
Douwe Maan
Michael Kozono
Grzegorz Bizon
Stan Hu
Grzegorz Bizon
Nick Thomas
Bob Van Landuyt
Tiago Botelho
Drew Blessing
Felipe Artur