I have a parent-child document mapping and parent only have one contact_id field. And I need to make sure that this parent document exists when I insert new child document. It may or may not be already existing.
So I use Bulk API to insert a parent if it not exist and insert a child in one request.
My Question is which method is faster: update
with doc_as_upsert
and detect_noop
OR index
new record with the same data that probably already exist:
{ update: { _index: 'index_name', _type: 'contact', _id: 25, _routing: 14}}
{ doc: { contact_id: 25 }, doc_as_upsert: true, detect_noop: true }
{ index: { _index: 'index_name', _type: 'event', _routing: 14, _parent: 25}}
{ ... event document body ...}
OR
{ index: { _index: 'index_name', _type: 'contact', _id: 25, _routing: 14}}
{ contact_id: 25 }
{ index: { _index: 'index_name', _type: 'event', _routing: 14, _parent: 25}}
{ ... event document body ...}
It seems like it's performs the same:
user system total real
update_10k_x1 6.460000 1.720000 8.180000 ( 79.737009)
index_10k_x1 6.300000 1.680000 7.980000 ( 80.067855)
update_10k_x2 12.660000 3.350000 16.010000 (159.787347)
index_10k_x2 12.690000 3.380000 16.070000 (160.276717)
update_10k_x3 18.870000 5.000000 23.870000 (242.023184)
index_10k_x3 18.940000 5.030000 23.970000 (240.063431)
Here is the benchmark code:
require 'benchmark'
require 'elasticsearch-ruby'
$client = Elasticsearch::Client.new
def update_10k(n)
index_name = "#{__method__}_x#{n}"
n.times do
(1..10000).each do |id|
body = []
body << { update: {_index: index_name, _type: :contact, _id: id }}
body << { doc: { contact_id: id }, doc_as_upsert: true, detect_noop: true }
$client.bulk body: body
end
end
end
def index_10k(n)
index_name = "#{__method__}_x#{n}"
n.times do
(1..10000).each do |id|
body = []
body << { index: {_index: index_name, _type: :contact, _id: id }}
body << { contact_id: id }
$client.bulk body: body
end
end
end
Benchmark.bm do |x|
(1..3).each do |n|
x.report("update_10k_x#{n}") { update_10k(n) }
x.report("index_10k_x#{n}") { index_10k(n) }
end
end