You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While doing the load testing, I found out the grpc call make the whole system slow down and cant run in parallel.
After digging into it, the grpc can use multiple cpu cores but not disk. Disk become bottleneck if some grpc call is heavily read/write disk.
Also the more delegations the slower it is. This is matching how disk performs as well.
And while doing this load testing, the babylon node can't receive any block either response peer. I guess all the operations involve reading disk are impacted.
Nov 15 00:48:25 babylon-rpc-0 babylond[15275]: 12:48AM ERR failed to write response err="write tcp 45.32.122.69:26657->18.162.253.87:59633: write: broken pipe" module=rpc-server
Nov 15 00:48:28 babylon-rpc-0 babylond[15275]: 12:48AM ERR failed to write response err="write tcp 45.32.122.69:26657->139.180.191.142:29214: write: connection reset by peer" module=rpc-server
Nov 15 00:48:28 babylon-rpc-0 babylond[15275]: 12:48AM ERR failed to write response err="write tcp 45.32.122.69:26657->139.180.191.142:29232: write: broken pipe" module=rpc-server
The text was updated successfully, but these errors were encountered:
liam-icheng-lai
changed the title
GRPC Performance issue related to disk utilization
GRPC Call Performance issue and lead to node not functioning
Nov 21, 2024
liam-icheng-lai
changed the title
GRPC Call Performance issue and lead to node not functioning
GRPC Call Performance issue and lead to node not function
Nov 21, 2024
Summary of Bug
While doing the load testing, I found out the grpc call make the whole system slow down and cant run in parallel.
After digging into it, the grpc can use multiple cpu cores but not disk. Disk become bottleneck if some grpc call is heavily read/write disk.
Also the more delegations the slower it is. This is matching how disk performs as well.
And while doing this load testing, the babylon node can't receive any block either response peer. I guess all the operations involve reading disk are impacted.
Version
v0.16.0
Steps to Reproduce
grpc call:
Experiment
CPU and Disk Resource Consumption
2 requests
CPU Utilization
Disk Utilization
8 requests
CPU Utilization
Disk Utilization
The text was updated successfully, but these errors were encountered: