While setting up an instance of PostgreSQL on my home Kubernetes cluster, the issue of how to create a proper persistent volume came up. I wanted to use network-based storage for the pod primarily because I use FreeNAS and backing up the volume to S3 using a Cloud Sync Task would be trivially easy. Right away I ruled out using NFS for PostgreSQL storage over reliability and performance concerns, so that left iSCSI. But, in order to get the best performance out of an iSCSI volume I had to use a zvol (device) instead of a file, which meant that Cloud Sync Tasks couldn't be used.

Read more...

I've been using Ansible to provision a Kubernetes cluster and I've found one major problem with the k8s module: it has, until recently, had a complete inability to handle the sort of multi-document YAML files that are typical of how lots of Kubernetes software is distributed (the other being Helm charts).

Read more...

Recently, I started looking at using Amazon S3 as a shared storage service instead of Dropbox, which I have been using for quite some time. I found S3 appealing because:

Read more...

In a previous post I explored a method to pass through an AMD GPU to an ESXi Windows guest that involves running batch scripts to enable and disable the GPU automatically.

Read more...

If you've managed to pass through an AMD GPU to a Windows guest in ESXi you may notice that the VM will fail to start after performing a reset. This is because AMD GPUs have issues with the kind of PCIe bus reset that hypervisors perform, which in turn causes issues with the AMD GPU driver.

Read more...

Windows 10 has a fairly aggressive automatic update policy and the knobs and levers required to change it are well-hidden. While I think this is a good idea for average consumers, it frequently causes problems with my VM that includes GPU passthrough – automatic update restarts seem to bypass the scripts that disable the GPU upon shutdown and problems ensue.

Read more...