Indexer is inconsistent when a field goes from a small value to an extremely large value.

Steps to reproduce

1. Create a index with the following definition:

   CREATE INDEX `test` ON `default`(`name`)

2. Run the following script:

   #!/usr/bin/python

   from couchbase.bucket import Bucket

   from couchbase.n1ql import N1QLQuery

   cb = Bucket('couchbase://10.111.151.101/default')

   def get_name_length():

   query = N1QLQuery("SELECT name FROM `default` WHERE name IS NOT MISSING")

   for row in cb.n1ql_query(query):

   return len(row['name'])

   for x in [10, 20971420]:

   cb.upsert('large', {'name': ("x" * x) })

   length_check = True

   while length_check:

   length = get_name_length()

   if length == x:

   length_check = False

   elif length == None and x == 20971420 :

   print "Index is empty at x: {} - length which is expected {}.\n Test has passed".format(x, length)

   length_check = False

   elif length < x:

   print "Retring x: {} - length: {}".format(x, length)

   The script will keep on looping as the index is never updated.

Problem

The problem here is that the projector does not inform the indexer about the document with the large field:

• Projector logs:

  2017-01-18T22:44:04.845+00:00 [Error] WRKR[43<-default<-127.0.0.1:8091 #MAINT_STREAM_TOPIC_e5:11:29:9:92:3b:f1:cc] ##4 TransformRoute: collatejson.outputLen

  2017-01-18T22:44:09.064+00:00 [Error] CollateJSONEncode: index field for docid: large (err: collatejson.outputLen)

Expected behaviour

The index not to contain any value at all, especially not the old value when the field is above the maximum index length size