rengwar

I’m running into recurring CPU storms on my “downloader” VM. These events fully redline the VM’s CPU allocation and effectively lock up the entire VM, even though Docker CPU limits are configured. Unraid itself does not redline. Essentially, I start downloading, everything is fine; CPU usage is ~10% (6 CPUs) for a while. A download finishes, sometimes it continues normally, other times CPU will start increasing substantially and over ~15minutes CPU utilization is redlined. I’m trying to determine the root cause. I believe the issue is related to an IO-wait storm but, I am totally open to other recommendations. Here are important details of my setup: Proxmox (PVE) is the host. Unraid is running on a VM. Downloader runs on Docker inside a PVE VM (not an Unraid VM). Unraid owns 4x28TB HDDs (controller passthrough) and 1x4TB NVME. Proxmox owns 1TB NVME. I have ~100 active downloads, ~50 active uploads, with ~500 files available for upload. Shares are scoped to disks (i.e. content2 = disk2, content3 = disk3) but I mounted the user shares (i.e. /mnt/user/content2, not /mnt/disk2/content2). Download Flow: /cache/incoming => /mnt/user/content2/complete => /mnt/user/content2/published (separrate process) The issue started after two events: Increasing some advanced settings in my downloader (now reverted). Running a workflow that touched the same data using both SMB and NFS, though in a chained, non-overlapping way. NFS is now fully unmounted and was only active during a handful of downloads. A key detail leading me to believe it’s IO-wait related is because the issue completely bypassed Docker CPU limits. I set a limit of 4 (of 6) CPUs and it consumed all 6. This tells me the problem is occurring below the container level, inside the kernel’s I/O or filesystem layer – not inside the downloader itself. This also makes it unlikely to be hash-checking related, which was my initial suspicion. Has anyone seen Unraid cause downstream VMs to enter an I/O-wait storm like this? Does this pattern sound familiar to anyone? Sudden CPU runaway (over ~10-15 minutes) VM redlining despite Docker CPU limits Unraid not redlining Downloads running smoothly for a while, then freezing the VM completely Could a temporary SMB+NFS dual-mount setup have created metadata inconsistencies, locking states, or user-share (FUSE) confusion that persist even after NFS was unmounted? For a short period, I had both SMB and NFS mounted to the same Unraid share, but they were used in a strictly chained, non-overlapping workflow: (SMB) /cache/incoming => /mnt/user/content2/complete (NFS) /mnt/disk2/content2/complete => /mnt/disk2/content2/staging Any ideas?