Sorting in mongo sharding environments degrades performance

user_1944986 0
Viewed 233

I have a query that performs something like this, 

last_shipment_id = OrderDelivery.where(platform: 'business').desc(:shipment_id).limit(1).pluck(:shipment_id)[0]

It works great with only 1 ~ 5ms when I have correctly index with { platform: 1, shipment_id: -1 } without sharding environment in staging machine

However, our production is set up with 4 sharding mongo db, the result ends up in 1000 ~ 3000ms.

Does anyone know how this could happen or how might solve this case?

I've read about this slides https://www.slideshare.net/mongodb/how-queries-work-with-sharding

Well, it's said on slide 13, but still not sure it has mentioned how to solve the case though.

mongoid mongodb ruby-on-rails
1 Answers

You don't mention what your shard key is, but this query has to be scattered to all three shards, and that means if any of the shards are slow, the overall result will be slow. In the gist you include explain for shard1 (which is fast) but omit it for shard2 and shard3 and overall numbers show that it's slow on one of those shards.

This means either the optimal index is not present on one of those shards, or there is a different index present that's being picked even though it's suboptimal. The solution in the former case is to build all correct indexes, the solution in the second case is to use hint with the query to force the use of the correct index.

user_4310120
Related