From 0af43a764893028a0931c146afc439f62b42e2bd Mon Sep 17 00:00:00 2001
From: Martin Martinez Rivera <martinmr@dgraph.io>
Date: Fri, 26 Apr 2019 16:34:58 -0700
Subject: [PATCH] Use initial schema during bulk load.

Use the initial schema during bulk load, otherwise data might not be
loaded correctly for those predicates.

In particular, I observed that when loading the 21million dataset with
dgraph.type triples added to it, queries would not respond and would
eventually timeout. I figured this was because the index for that
predicate was not built during the bulkload. I reloaded the dataset with
my changes included and queries respond immediately.
---
 dgraph/cmd/bulk/schema.go | 10 ++++++++++
 schema/schema.go          | 18 +++++++++++++++++-
 2 files changed, 27 insertions(+), 1 deletion(-)

diff --git a/dgraph/cmd/bulk/schema.go b/dgraph/cmd/bulk/schema.go
index dcf28c1e9b0..369a1bfbf2f 100644
--- a/dgraph/cmd/bulk/schema.go
+++ b/dgraph/cmd/bulk/schema.go
@@ -25,6 +25,7 @@ import (
 	"github.com/dgraph-io/badger"
 	"github.com/dgraph-io/dgraph/posting"
 	"github.com/dgraph-io/dgraph/protos/pb"
+	"github.com/dgraph-io/dgraph/schema"
 	wk "github.com/dgraph-io/dgraph/worker"
 	"github.com/dgraph-io/dgraph/x"
 )
@@ -45,6 +46,15 @@ func newSchemaStore(initial []*pb.SchemaUpdate, opt options, state *state) *sche
 		},
 		state: state,
 	}
+
+	// Load all initial predicates. Some predicates that might not be used when
+	// the alpha is started (e.g ACL predicates) might be included but it's
+	// better to include them in case the input data contains triples with these
+	// predicates.
+	for _, update := range schema.CompleteInitialSchema() {
+		s.m[update.Predicate] = update
+	}
+
 	if opt.StoreXids {
 		s.m["xid"] = &pb.SchemaUpdate{
 			ValueType: pb.Posting_STRING,
diff --git a/schema/schema.go b/schema/schema.go
index 27213ce4c2e..20cedde9a9d 100644
--- a/schema/schema.go
+++ b/schema/schema.go
@@ -408,7 +408,23 @@ func LoadTypesFromDb() error {
 	return nil
 }
 
+// InitialSchema returns the schema updates to insert at the beginning of
+// Dgraph's execution. It looks at the worker options to determine which
+// attributes to insert.
 func InitialSchema() []*pb.SchemaUpdate {
+	return initialSchemaInternal(false)
+}
+
+// CompleteInitialSchema returns all the schema updates regardless of the worker
+// options. This is useful in situations where the worker options are not known
+// in advance and it's better to create all the reserved predicates and remove
+// them later than miss some of them. An example of such situation is during bulk
+// loading.
+func CompleteInitialSchema() []*pb.SchemaUpdate {
+	return initialSchemaInternal(true)
+}
+
+func initialSchemaInternal(all bool) []*pb.SchemaUpdate {
 	var initialSchema []*pb.SchemaUpdate
 
 	// propose the schema for _predicate_
@@ -428,7 +444,7 @@ func InitialSchema() []*pb.SchemaUpdate {
 		List:      true,
 	})
 
-	if x.WorkerConfig.AclEnabled {
+	if all || x.WorkerConfig.AclEnabled {
 		// propose the schema update for acl predicates
 		initialSchema = append(initialSchema, []*pb.SchemaUpdate{
 			{