GraphQL Admin API: Support export, draining, shutdown and setting lru_mb operations

[pawanrawal]

This PR adds support for doing export, shutdown and draining using the GraphQL Admin API. The mutations are shown below. The current /admin HTTP endpoints have been kept as before.

Export

mutation {
  export(input: {format: "json"}) {
    response {
      code
      message
    }
  }
}

Draining

mutation {
  draining(input: {enable: false}) {
    response {
      code
      message
    }
  }
}

Shutdown

mutation {
  shutdown {
    response {
      code
      message
    }
  }
}

Config(lru_mb)

mutation {
  config(input: {lruMb: 1025}) {
    response {
      code
      message
    }
  }
}

---

This change is 

[martinmr]

Just a minor comment but the changes in Dgraph look good. I didn't review the graphQL part since I am not very familiar with it.

Reviewed 4 of 9 files at r1.
Reviewable status: 4 of 9 files reviewed, 2 unresolved discussions (waiting on @manishrjain, @MichaelJCompton, and @pawanrawal)

---

worker/groups.go, line 1084 at r1 (raw file):

			if err != nil {
				glog.Errorf("Error from alpha client subscribe: %v", err)
				time.Sleep(100 * time.Millisecond)

I am assuming this is done to avoid hogging the CPU, right?

---

worker/worker.go, line 128 at r1 (raw file):

	}

	posting.Config.Mu.Lock()

If possible, I would change the definition of posting.Config from

type Config struct {
...
Mu sync.Mutex
...
}

to

type Config struct {
...
sync.Mutex
...
}

so you can write this line as posting.Config.Lock().

[MichaelJCompton]

Looks good.

I think after release we should go through and reduce boilerplate in these admin fns.

Same as with backup, we need to confirm if the old endpoints are being removed and then put up other PRs that do that and convert the tests over.

Reviewable status: 4 of 9 files reviewed, 6 unresolved discussions (waiting on @manishrjain, @MichaelJCompton, and @pawanrawal)

---

graphql/admin/admin.go, line 98 at r1 (raw file):

	}

	type ExportPayload {

(same comment holds here for backup endpoint)

How sure are we about these responses? If Dgraph always responds with just code and message, has done so for a while and we don't think we'll change that ... then it's probably ok to just return type Response for all of these.

The way you have done it is more flexible and might avoid changes in the future, but this is one spot where we'd save a bunch of types.

---

graphql/admin/admin.go, line 298 at r1 (raw file):

				})
		}).
		WithMutationResolver("export", func(m schema.Mutation) resolve.MutationResolver {

same as with backup, check that these can be called while the server is starting, otherwise move to addConnectedAdminResolvers

---

graphql/admin/admin.go, line 494 at r1 (raw file):

}

func writeResponse(m schema.Mutation, code, message string) []byte {

see note in backup.

Just write code, message and m.Name here. Aliases and selections are all handled generically now.

---

graphql/admin/config.go, line 57 at r1 (raw file):

	result map[string]interface{}) (*gql.GraphQuery, error) {

	return nil, nil

this is fine for now ... if we weren't under release pressure, we'd change some of this setup so you have to write less boilerplate.

Let's do a pass after release and make the works of all these admin bits one step nicer.

[pawanrawal]

Agree, I'll check with @manishrjain again if we are removing the old endpoints.

Reviewable status: 2 of 11 files reviewed, 6 unresolved discussions (waiting on @manishrjain, @martinmr, and @MichaelJCompton)

---

graphql/admin/admin.go, line 98 at r1 (raw file):

Previously, MichaelJCompton (Michael Compton) wrote…

(same comment holds here for backup endpoint)

How sure are we about these responses? If Dgraph always responds with just code and message, has done so for a while and we don't think we'll change that ... then it's probably ok to just return type Response for all of these.

The way you have done it is more flexible and might avoid changes in the future, but this is one spot where we'd save a bunch of types.

Keeping it as it is, will do this change along with GRAPHQL-250 before the release.

---

graphql/admin/admin.go, line 298 at r1 (raw file):

Previously, MichaelJCompton (Michael Compton) wrote…

same as with backup, check that these can be called while the server is starting, otherwise move to addConnectedAdminResolvers

Hmm, looking at the code I am not sure why we have two functions for doing all this as initServer is called right after newAdminResolverFactory so there isn't anything substantial that happens between the two. Maybe earlier we were doing a schema Alter, hence we had these differentiation?

Dgraph itself should return an error if its not ready to perform these operations, so we should be safe here.

---

graphql/admin/admin.go, line 494 at r1 (raw file):

Previously, MichaelJCompton (Michael Compton) wrote…

see note in backup.

Just write code, message and m.Name here. Aliases and selections are all handled generically now.

I'll add a card to clean this up after the health endpoint PR is merged next week. Keeping it as it is for now.

https://dgraph.atlassian.net/browse/GRAPHQL-250

---

graphql/admin/config.go, line 57 at r1 (raw file):

Previously, MichaelJCompton (Michael Compton) wrote…

this is fine for now ... if we weren't under release pressure, we'd change some of this setup so you have to write less boilerplate.

Let's do a pass after release and make the works of all these admin bits one step nicer.

Agree, there is too much boilerplate to conform to the interface that is defined here.
Created https://dgraph.atlassian.net/browse/GRAPHQL-249

---

worker/groups.go, line 1084 at r1 (raw file):

Previously, martinmr (Martin Martinez Rivera) wrote…

I am assuming this is done to avoid hogging the CPU, right?

That is right

---

worker/worker.go, line 128 at r1 (raw file):

Previously, martinmr (Martin Martinez Rivera) wrote…

If possible, I would change the definition of posting.Config from

type Config struct {
...
Mu sync.Mutex
...
}

to

type Config struct {
...
sync.Mutex
...
}

so you can write this line as posting.Config.Lock().

done