Support Restricting Monitoring to a Single Cluster #160
Conversation
These replace the GetScheduledBuilds query
| @@ -1,7 +1,5 @@ | |||
| package api | |||
|
|
|||
| //go:generate go run github.com/Khan/genqlient | |||
|
|
|||
There was a problem hiding this comment.
Moved to api/client.go as it is what uses the generated client.
api/schema.graphql
Outdated
There was a problem hiding this comment.
Copied from the bk/bk source code. I have no idea why it is so different. The diff only needs to be the new arguments to the job field, allowing filtering by cluster.
triarius
changed the title
internal/monitor/monitor.go
Outdated
| for _, job := range jobsResp.Organization.Jobs.Edges { | ||
| jobs = append(jobs, job.Node.(*api.JobJobTypeCommand)) | ||
| } | ||
| } else { |
There was a problem hiding this comment.
[Optional] Probably worth considering dividing the the method into two sub-methods, one for each of the branches.
There was a problem hiding this comment.
I think this is a symptom of how the generated methods return 2 struct types, GetScheduledJobsReponse and GetScheduledJobsClusteredResponse that have the same structure and need to be handled identically, but the type system won't let us write only one block of code to handle both.
I decided to rectify this using an interface to abstract over the parts of the structs we need. There's still a bit of duplication, but it is limited to identical implementations of the interface methods that only do "one thing". If go had a more expressive syntax, both of these methods could be one-liners. In any case, this should be earlier to maintain that larger duplicated blocks that do multiple things.
triarius
force-pushed
the
pdp-825-query-by-cluster-in-the-controllers
branch
from
df0db7f
to
fded53f
Compare
triarius
force-pushed
the
pdp-825-query-by-cluster-in-the-controllers
branch
from
0f62938
to
188931c
Compare
triarius
force-pushed
the
pdp-825-query-by-cluster-in-the-controllers
branch
from
188931c
to
22c830f
Compare
Previously, the controller would poll the GraphQL API for all jobs in an organization that matched certain tags. However, if the controller were only equipped with a Cluster token, the buildkite agent in the resulting pods would not be able to acquire the Buildkite Job from a different (or no) cluster.
In this PR, we change the GraphQL query to either restrict jobs to a configured Cluster UUID, or if no Cluster UUID is in the config, to jobs NOT belonging to a cluster.
There is also some improvements to how the GraphQL client is generated. In particular, there is a recipe for downloading the schema file that may be used by those with access to the buildkite repositories.