Sysadmin Mega Thread

Just any regular disk controller should be fine. Ceph, much like ZFS, does not like it if a hardware RAID controller is sitting between it and the disks. This means that any controller you get must either provide direct disk access.

Gluster on the other hand does not care in the slightest since it is working at the filesystem level.

I meant like a server. I thought ceph had storage servers and controller server. Did I make that up? I think 45Drives sells them like that.

Oh, gotcha. Ceph has three services needed to run.

The “monitor” and “manager” hold the cluster state and provide an interface to it. These are usually coupled together but can be hosted on different boxes. The last service(s) are the “Object Storage Daemons” which manage the physical storage devices.

Definitely a bit of overhead involved.

1 Like

Anyone using cobbler?

Like for shoes…?

3 Likes

Haven’t tried it yet, but it’s an abstraction layer for PXE/tftp netboot stuff. Apparently made by one or several of the ansible devs. Might try it out…

1 Like

Ah yeah OG spacewalk used it.

Fucking annoying since telnet is a hard dependency IIRC.

2 Likes

For spacewalk or cobbler? I didn’t see anything about telnet, but I was only scanning the docs.

Cobbler, which spacewalk used.

1 Like

You’d think that a vanilla install of a RHEL distro would incur no change when you run sudo restorecon -vr / yet there are always a handful of things that are changed… should I run this periodically? It did fix an issue I was having being locked out of updating /etc/localtime.

https://brendangregg.com/blog/2021-07-05/computing-performance-on-the-horizon.html

I use windows for daily compute. Debian and pop is distros for my bad and Minecraft server

Anyone with have experience with Windows Clusters? I’m having an issue where 4k random writes to the CSVs are absolutely terrible, as in 1-2MB/s. It’s on SATA SSD RAID 5 arrays with Starwind VSAN. Writing to the arrays themselves is okay, about 80MB/s. The main victim here is Veeam restores which take forever to complete as a result.

So I just had an interesting experience where an improperly configured ALB caused a python WSGI server to hang and only process incoming requests once every tcp timeout duration.

Turns out that this server, for some reason, doesn’t handle tcp connections completely properly. Additionally this appears to be unique behavior of the pypi server application.

:man_shrugging:

This made for a fun debugging session with the rest of the team.

1 Like

wasnt this in the l1news today about having malware in it?

yep
image

No. We run our own private pypi server for some of our internal tools. Amazon’s load balancers are just short bus riding, pants on head, glue eaters that don’t always send valid http requests, apparently.

Another reason my hatred for AWS burns brighter than sol.

You can get pypi in docker format, and it’s awful. I am trying to get authorization to improve it and submit PRs on company time.

Apparently l1news was talking about a supply chain attack

1 Like

Does anyone know how to disable all the mitigations for the speculative execution vulns on Windows machines? I know that linux has make-linux-fast-again.com, but I haven’t heard of an equivalent for Windows.

1 Like

They do, I think it was called XP

3 Likes

Dammit, bobby!

3 Likes