Score:0

High Disk IO with Network and CPU spikes

cn flag

I have a Linode server that has been under load for a few hours. The configuration is 2 CPU Cores, 4GB memory, 80GB SSD storage mounted as 1 partition, 500MB swap. There's a second hard drive of 30GB which is also SSD. Software is Ubuntu 18.04, Apache 2.4.52, PHP-FPM (Most sites using 8.0, some on 7.2 and 8.1). I'm also using fail2ban and iptables for protection.

Here's the graph that Linode's dashboard shows me

Linode Dashboard

Earlier I was getting spikes in network which corresponded to spikes in CPU usage. It was enough to make the websites on the server not respond (500 error) during the spikes. Now, I have constantly higher than usual disk IO.

Linode has a built in DDoS protection and I asked them about it and they said it didn't trigger any actions from DDoS protection layer. Does this look like an attack?

I've tried several commands but none of them has given me any clues so far.

iostat

iostat

iotop -a :

iotop

vmstat 1 10

vmstat

top : Again there's no process constantly on the list

top

jp flag
150 `tps` on `sda` and 200 `IOPS` on the graph for >6 hours and no processes in `iotop`?
Whip avatar
cn flag
Not no process but processes come and go with only a few kb/s read and writes. Is it possible to get an average over time by each process?
jp flag
you can run `iotop -a` to accumulate stats. Read `man iotop`.
Whip avatar
cn flag
The problem with that is when the process exits, its removed from the list. So not useful still
Whip avatar
cn flag
I don't know if there should be so many www-data processes. I've added the screenshot to the question
jp flag
You can try an alternative `iotop` implementation, it remembers dead processes. https://github.com/Tomas-M/iotop
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.