now those exceptions are being returned in GraphQL,
not raised

When using the new graphql interpreter, we need to be using the updated
types, otherwise validation fails.  They are the same types, so external
interfaces won't change.

When using the new graphql interpreter, we need to be using the updated
types, otherwise validation fails.  They are the same types, so external
interfaces won't change.

When using the new graphql interpreter, we need to be using the updated
types, otherwise validation fails.  They are the same types, so external
interfaces won't change.

When using the new graphql interpreter, we need to be using the updated
types, otherwise validation fails.  They are the same types, so external
interfaces won't change.

enabled

ported from the original RecursionAnalyzer

ported from the original LoggerAnalyzer

for the deprecated version

Enable runner_detailed_view_vue_ui by default

See merge request !65948

Enable runner_list_view_vue_ui by default

See merge request !65901

Create branch from Jira issue MVC 1: add supporting Vue components

See merge request !66034

Denormalize shared runners to CI::PendingBuild

See merge request !64644

Partition manager identifies partitions to detach

See merge request !65093

Release the Helm charts registry

See merge request !65768

Steve Abrams authored 3 years ago and David Fernandez committed 3 years ago

- Remove the helm_charts feature flag
- Update helm related documentation

Changelog: added

Fix GET requests from graphql client when using relative url

See merge request !66053

Add information about The Art Of PostgreSQL training

See merge request !65272

Improve inline error message for elapsed time field in policy model

See merge request !65775

Docs for recording and assignment tracking shared example

See merge request !66011

Update Gitaly version

See merge request !66068

Migrate the deprecated "urlParamsToObject" to "queryToObject"

See merge request !65898

Simon Tomlinson authored 3 years ago and Andreas Brandl committed 3 years ago

By creating and detaching partitions using the same lease, we can
easily skip the partition sync code if another node is currently
executing it. This is safe because we have six months of headroom for
future partitions and the job to sync partitions runs every six hours

Simon Tomlinson authored 3 years ago and Andreas Brandl committed 3 years ago

Adds the partition_pruning_dry_run feature flag.

Even though the partition manager does not actually detach any
partitions yet, putting the logic for determining the set of partitions
to detach behind a flag makes rolling out this code safer, because the
logic changed touches the logic for creating new partitions.

Feature flag issue: #335315

Simon Tomlinson authored 3 years ago and Andreas Brandl committed 3 years ago

The previous code was using try_obtain once and two seperate locks,
so concurrent code would skip the critical section.

Switching to using in_lock and a single lock ensures that all concurrent
runners get to execute the code, and that partition creation and
partition detaching do not interfere with each other.

Simon Tomlinson authored 3 years ago and Andreas Brandl committed 3 years ago

This worker runs `PartitionManager.detach_partitions` on a schedule to
trigger the detach logic for old database partitions that are no longer
needed