Some bad_alloc on 6.0/6.1

ScyllaDB
, ,
1

Installation details
#ScyllaDB version: 6.1.5-0.20250119.c84780618297
#Cluster size: 150 To on 10 nodes
os (RHEL/CentOS/Ubuntu/AWS AMI): Ubuntu 22.04

Hi, since I have upgrade from scylla 5.4 to 6.0, i have ~20/min batch insert queries failing with “bad_alloc” reason (while between 5k and 15k queries beside are ok).
Upgrading to 6.1 did not helped, neither reducing the size of batches.
I did not encounter this problem with v5.4, and I have no idea of which node is generating this kind of error. I have tried to add a few gb of memory whithout any luck.

What can I do ?

2

How much memory do you have per node?
We usually recommend 8GB per shard - so if you have 150TB on 10 nodes, thus 15TB per node, the minimum memory on each node should be ~160GB. The recommended number of vCPUs in that case is ~20 per node.

3

Thanks for your reply
Initial setup was 125 Go with 4 vcpu, that went fine with 5.4 ; i tried to raise it up to 175G on 3 nodes but without any significant win. Payload is now ~10To per node, still 10 nodes.