What’s the connection scaling properties of CrateDB?

CrateDB
1

What’s the connection scaling properties of CrateDB?

Let’s assume a 3-node CrateDB cluster, how many concurrent connections can a single node handle?

Is there some benchmark about this?

2

Hi @Emre_Sevinc,

Let’s assume a 3-node CrateDB cluster , how many concurrent connections can a single node handle?

there is not a clear cut answer for this as it depends on multiple factors like number of CPUs, type of queries, number of shards, etc.

The following discussion on Elastic Discussion forums does go into more details: Concurrent search request to elasticsearch - #2 by radu_gheorghe - Elasticsearch - Discuss the Elastic Stack

3

As mentioned by @jayeff there typically it would not be a connection problem, rather the amount of queries that can be handled in general by the cluster.

1 Like
4

To add some more technical insights here. CrateDB uses an asynchronous event loop model to handle connections, so a single connection has little overhead (Likely the range of a couple hundred bytes, plus file descriptor)

PostgreSQL in comparison spawns a process per connection, which is much heavier. (But also has some resource isolation advantages)

2 Likes