harvester: [BUG] Windows VMs crashing

Describe the bug Windows VM crashing. When i use image with preinstalled windows server 2019 or trying to install VM from iso image the VM restarted by itself after some period (seconds, 2-3 minutes). I`m using Windows VM template as described in documentation. When i use resources like 2CPU and 4GB RAM the VM is stable, when i use 4CPU and 8GB RAM the VM is crashing. Tested on 3 different environments with 3 nodes and also with single node cluster.

To Reproduce Steps to reproduce the behavior: each time trying to run or install windows

Environment

Harvester ISO version: 1.0.2
Underlying Infrastructure: tried with different platforms and environments. 3x Supermicro, Intel XEON D2183IT, 256GB RAM

** support bundle ** i have uploaded the supportbundle by ftp

About this issue

Original URL
State: open
Created 2 years ago
Comments: 16 (12 by maintainers)

Most upvoted comments

After configuring 512MiB reserved memory my Windows 2019 Server VM has been running for 10 days without any crash. CPU: 4 Memory: 8GiB

janeczku on Aug 2, 2022

Hey, had the same issue and setting the reserved memory to 512MiB fixed the issue for me. 256 still caused issues on our side.

RegisHubelia on Sep 1, 2022

@futuretea @guangbochen

It seems unexpected that Harvester is automatically applying memory limits to the VMI: Normally, the memory limit is a user controlled parameter in Kubevirt.

If Harvester wants to be in control of the memory limit, then it must properly calculate the overhead required to avoid OOM killer.

Currently, Harvester uses a hardcoded overhead of 100MiB for the qemu process. Clearly, this overhead is insufficient and this causes the OOM events. https://github.com/harvester/harvester/commit/39c76d70e7ce178a26142bac43d09fc0002e1290

Advanced Options -> Reserved Memory does not appear to be the issue here.

VMI launched by Harvester:

    memory:
      guest: 8092Mi   <- Configured by Harvester (using hardcoded 100MiB overhead)
    resources:
      limits:
        cpu: "4"
        memory: 8Gi <- Limit configured by Harvester
      requests:
        cpu: 250m
        memory: 5461Mi

janeczku on Jul 21, 2022