Alpha: Gracefully shutdown ludicrous mode

[jarifibrahim]

Fixes - DGRAPH-1353

When live loader in running on an Alpha in ludicrous mode, alpha will
crash with a segmentation fault error because badger was closed before
ludicrous mode goroutines were stopped. This PR fixed that issue.

The crash can be reproduced by

1. dgraph zero --ludicrous_mode
2. dgraph alpha --ludicrous_mode --enable_sentry=false
3. dgraph live -f $GOPATH/src/github.com/dgraph-io/benchmarks/data/21million.rdf.gz -s $GOPATH/src/github.com/dgraph-io/benchmarks/data/21million.schema -z localhost:5080 -a localhost:9080 --ludicrous_mode

Hit <CTRL+C> while alpha is receiving data from live loader.

---

This change is 

[ashish-goswami]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @jarifibrahim, @manishrjain, and @vvbalaji-dgraph)

---

worker/executor.go, line 68 at r1 (raw file):

				return
			default:
			}

We should avoid using it in critical path, might hamper performance significantly.

[jarifibrahim]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @ashish-goswami, @manishrjain, and @vvbalaji-dgraph)

---

worker/executor.go, line 68 at r1 (raw file):

Previously, ashish-goswami (Ashish Goswami) wrote…

We should avoid using it in critical path, might hamper performance significantly.

Without this return statement, Alpha took around 1 minute to close in my tests. It could take more than 1 minute if palyload.edges is huge.

[ashish-goswami]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @jarifibrahim, @manishrjain, and @vvbalaji-dgraph)

---

worker/executor.go, line 68 at r1 (raw file):

Previously, jarifibrahim (Ibrahim Jarif) wrote…

Without this return statement, Alpha took around 1 minute to close in my tests. It could take more than 1 minute if palyload.edges is huge.

How big is payload.edges in your case? If its too big we can do it every 10K or 100K edge.

[jarifibrahim]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @ashish-goswami, @manishrjain, and @vvbalaji-dgraph)

---

worker/executor.go, line 68 at r1 (raw file):

Previously, ashish-goswami (Ashish Goswami) wrote…

How big is payload.edges in your case? If its too big we can do it every 10K or 100K edge.

I see a maximum of 1000 edges but I think that wasn't the bottleneck. I've moved the select statement to check at the channel range loop. It takes 7 seconds now for Alpha to close after hitting ctrl+c

[jarifibrahim]

Dismissed @ashish-goswami from a discussion.
Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion (waiting on @manishrjain and @vvbalaji-dgraph)

---

worker/executor.go, line 68 at r1 (raw file):

Previously, jarifibrahim (Ibrahim Jarif) wrote…

I see a maximum of 1000 edges but I think that wasn't the bottleneck. I've moved the select statement to check at the channel range loop. It takes 7 seconds now for Alpha to close after hitting ctrl+c

I had a chat we @ashish-goswami and we decided that this change is not needed right now.

[ashish-goswami]

Reviewable status: 0 of 1 files reviewed, all discussions resolved (waiting on @manishrjain and @vvbalaji-dgraph)

[ashish-goswami]

Reviewed 1 of 2 files at r1, 1 of 1 files at r2.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @manishrjain and @vvbalaji-dgraph)

[martinmr]

Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @manishrjain and @vvbalaji-dgraph)

[manishrjain]

Let's separately clean up how closing happens.

Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @manishrjain and @vvbalaji-dgraph)