Forum: Building VoltDB Applications

Post: Question about AdHoc queries

Question about AdHoc queries
Aug 15, 2015
Supposing that all tables are replicated, not partitioned.
And client application connects to VoltDB and uses sync. calls.

Should we expect same performance results for given 3 examples?
There is a 2 node cluster:
---client driver connects to 1 node and runs too many adhoc queries
---client driver connects to 2 nodes and runs too many adhoc queries
There is only 1 node cluster
---client driver runs too many adhoc queries

Aug 16, 2015
From memory (so could be wrong):

Each server will accept a fixed number of outstanding operations before turning on back pressure (I think 3000).

Each server has a planning queue that accepts ad-hoc work for planning. If the planning queue is longer than a fixed number (I think 300), then the caller sees an error. Note that not all planning is equal; a plan-cache hit zooms through this queue.

But the point is that the total number of outstanding transactions AND the total number of adhoc ops in the planning queue grows with the number of machines. However, you need to connect to both machines to see the benefit.