Zabbix on a VM vs Docker

[ian]

Hi

Is it better to run zabbix on virtual machine or in a docker container?

My options are

a) Running on a dedicated Ubuntu VM which is on a proxmox server
or
b) Running in docker where dockerd is running on a Debian 10 VM also on a proxmox server

It's a small network that I am looking to monitor, with a mixture of windows & unix OS's, already running in a mixture of bare metal, proxmox VMs and docker containers.

I appreciate there is additional network overheads with containerisation, but was wondering if someone had written and article on zabbix, installed on bare metal, verses VM, verses Docker, etc?

[Cloud8x4]

Hi,

I started with Zabbix server installed on Ubuntu Server 14 VM. Since Ubuntu 18 Server I'm using dockers instead.

For a small network, I don't believe there is a noticeable difference in performance.

I think it's easier to setup and maintain with Docker because you don't need to install and configure various services like MySQL and nginx in the host.
As you can read in the manual, you can install with 2 commands, git checkout and docker-domposer; only need to change the configuration files to your needs and they are all in the same directory; no need to copy them to the /etc/.. folders.
No experiences with bare metal.

[sjohn]

Hi

I never tried Zabbix in docker though have have run other docker apps on UnRaid. We have over 300 customer and network devices to monitor and 2 office locations half a block
apart. About a year ago I migrated the Cacti 0.88b database on bare metal to 1.2 on the UnRaid as a VM. I've had this Zabbix VM up for more than 6 months with Cacti as VM's on an
UnRaid hypervisor and just recently started using the Zabbix VM. I am impressed as it is a lot easier to do stuff with Zabbix compared to Cacti. Once I got started it only took me
about 90 minutes to figure out how to monitor a non standard MIB's on the Ubiquiti AF5XHD's. Import export of devices, copy items and graphs to other devices, it's slick. I was
able to extract mysql data with tree informaiton from Cacti format it as xml using the tree data as groups and import it with IP SNMP community and version into Zabbix.

The second office pointed out that the new Cacti VM on UnRaid has gaps in some graphs while the original does not. So I got a spare i7 4 cores 8 threads 16GB of memory and setup a
bare metal Zabbix server and when spot checking the VM I found some gaps in some of the graphs, ping is solid but Hw and Network SNMP uptime readings have gaps. These gaps do not
show up on the i7 bare metal machine. I increased the VM Zabbix cpus to 8 and memory to 16G, switched the VM to another drive, and changed the networking to be configured with
ifup ifdown same as the new bare metal but I still have gaps. I want to move the BM Zabbix to the same VLAN as the UnRaid hypervisor to test if it is some other network issue.
That being said, a VM will never perform as well as bare metal even if just running one VM on the hypervisor. I've run HyperV, VirtualBox, ESXI, KVM, and bhyve though most of my
experience is with KVM, HyperV, and bhyve with very little actual ESXI as I didn't like the fact I couldn't get a decent graphic resolution on the VM's and couldn't easily get
pass through to work. Network monitoring is important (customers will complain if services are not up) so I advise to use bare metal.