Aerospike Big Memory Usage

1

Im using Aerospike to store events from 2 Kafka Topics, on production, each of topic would be with 100-300k TPS; However, in dev environment, I only sampling around 40mio events each, so in total i would insert 80mio events to Aerospike

In Aerospike, each events will go to separated Sets, imitating original topic; I know that for 1 record, the size will be roughly 64B for Primary Index and 850B for Data;

Then, i set my config with several points

service {
        proto-fd-max 40000
        cluster-name cakery
}

namespace bar {
        replication-factor 1

        # Enable nsup default expiration for namespace
        # By default is 0, and no records with ttl > 0 can be written
        nsup-period 900

        storage-engine device {
                file /opt/aerospike/data/bar.dat
                filesize 20G
        }
}

I also commented the namespace ‘test’, so practically i only use ‘bar’

Im using Aerospike 8 Community Edition

My VM is around 16Core and 16GB RAM and using SSD, i believe this is not big deal because it is not in production nor i do performance test;

Upon insertion, when the records is around 3-4mio each, i notice on ‘top’ command, that the memory used is 13GB;

My confusion and my questions are:

  • By stating storage-engine device, Aerospike should be automatically store index on memory and data on disk right?
  • Therefore, the memory occupancy should around 2 sets * (64B * 4mio records) which is under 13GB;
  • Stopping Aerospike server is not releasing the memory, only after i DELETED the .dat file, the memory goes back to buff/cache

Why does this mechanism happens? How do i ended with much bigger resources that theoritically needed? And What configurations should i have if i want Index on Memory and Data on disk?

2

Cross posted on Stackoverflo. Posted a response there.