Rclone moving files at 150mb out and websites down

disaint

Verified User
Joined
Apr 13, 2020
Messages
47
Hi there!

I am having an Issue with Rclone on my CentOS 8 VPS (KVM).
I started working with Onedrive to send all my backups. But Onedrive was very slow (2mb).
It takes about 7h to move 70gb.
So I tried BlackBlaze. Too fast to be true! (150mb). 70Gb moved in 1h. Crazy!

The issue with Rclone using BlackBlaze in my VPS is that It seems to be using to much resources somewhere in the moving process that makes some websites to be down (timeout response) for about 5 min.

I tried looking into logs with --DEBUG and -vv flags and everything seems fine.
Then I looked to NETDATA and saw something strange. Maybe is not real.

But Rclone seems to be using only one cpu for network process (sortirq) and not all the rest of CPUs (6 cores).
So I asked Datacenter and they said it was a software issue because it was only using one cpu to transfer data out of my server.
I do not know why a 150MB transfer out makes websites down (30% CPU usage).

- Is really a Rclone issue?
- Centos 8 issue?
- DA issue?
- Where to start looking?
- I tried changing with flags some settings about files chunked sizes, but not working at all.

Now some NETDATA screenshots:

system_overview.PNG
cpu_load.PNGsoftirqs_cpus.PNG
eth0.PNG


Thanks in advance!
 

jca

Verified User
Joined
Oct 31, 2006
Messages
281
Location
Allen, TX
I would start checking with --bwlimit flag: https://rclone.org/docs/#bwlimit-bandwidth-spec

Transfers should not be very CPU intensive and it's way more I/O intensive. Many people run it on Raspberry Pi devices with no issues at full gbit speeds.

Limiting the bandwidth being used should let the disks keep up and if it's network related, leave enough available bandwidth for the rest of the packets to arrive. Based on what you find you can figure out what is causing this behavior.
 

disaint

Verified User
Joined
Apr 13, 2020
Messages
47
Thanks @jca!
I will try that flag tonight.

I do not know why softirq is 100% load over one of the CPUs when transfering out.
Maybe some KVM issue? config? setting?
 

jca

Verified User
Joined
Oct 31, 2006
Messages
281
Location
Allen, TX
Doing a quick google search:

"Another important metric worth monitoring, is softirq. A constantly high percentage of softirq may indicate network driver issues."

I would check the driver just to be sure. Seems softirq is related to network.
 

disaint

Verified User
Joined
Apr 13, 2020
Messages
47
Yes, I know that but my VPS is a KVM virtualization with dedicated resources.
I do not event know where to check network drivers.
Datacenter told me it was a software issue. But I not quite shure.

NETDATA gives me a lot of infot a tech details to start working.
 
Top