High Availability Cluster in Proxmox 5 + LXC & Watchdogs

S

starnetwork

Renowned Member
Dec 8, 2009
430
10
83
Hi,
just set HA Cluster with Proxmox 5.x with Ceph shared Storage
now, I set the fencing via IPMI Watchdog
1. I set /etc/modprobe.d/ipmi_watchdog.conf with
options ipmi_watchdog action=power_cycle panic_wdt_timeout=10

2. I added nmi_watchdog=0
for: /etc/default/grub
and did reboot.

now, my questions:
1. how can I know that fencing / Watchdogs is set work work well?

2. I did test, powered Off one mahine with LXC containers and it' take about 2 minutes to power on the LXC container from new node, it's reasonable?
should it's take about 2 minutes?

Regards,
 
  • Like
Reactions: chocosabrinax2017
starnetwork said:
1. how can I know that fencing / Watchdogs is set work work well?
Click to expand...

Test, for example by disconnecting the network cable.

starnetwork said:
2. I did test, powered Off one mahine with LXC containers and it' take about 2 minutes to power on the LXC container from new node, it's reasonable?
should it's take about 2 minutes?
Click to expand...

yes.
 
Hi dietmar,
thanks for that answers!
so:
1. it's work!

2. why it's take too long, in Video guides, I see 5-6 seconds for VM and LXC container should be even faster, no?
who can I Reduce the downtime time?

Regards,
 
starnetwork said:
who can I Reduce the downtime time?
Click to expand...

The watchdog based self fencing needs a 2 minutes timeout for safety reasons. In future, we will implement active fencing using hardware fence devices, which can reduce timeout to about 1 minute (minimum to detect node failure).
 
Hi dietmar,
Apologize in advance for insistence, it's just important issue.

just curious how it's was work in previews versions and allow 5-6 seconds ping downtime only?
possible to disable the watchdog?
maybe any option to set the X seconds timeout before activate the HA?
also 1 minute is too long for downtime...
Especially when we saw that it worked in the past with 5-6 downtime pings

Regards,
 
dietmar said:
The watchdog based self fencing needs a 2 minutes timeout for safety reasons. In future, we will implement active fencing using hardware fence devices, which can reduce timeout to about 1 minute (minimum to detect node failure).
Click to expand...

Is it correct that Proxmox VE 5.1 currently "only" has watchdog based fencing?
In PVE3 hardware fencing (IPMI over LAN, Dell iDRAC, etc) could be used to stonith (shoot the other node in the head, poweroff the system). Currently there seems to be no option to configure this any more.

The documentation at https://pve.proxmox.com/pve-docs/chapter-ha-manager.html#ha_manager_fencing talks about "Possible Fencing Methods" including external power switches. I assume that this also means something like Dell's iDRAC to remotely power off another node. Are there plans to implement that as a feature?
 
You must log in or register to reply here.
Top Bottom