💾 Archived View for gem.kirigiri.me › gemlog › 2024-09-21.gmi captured on 2024-09-29 at 00:22:35. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
Wow, it's been a while huh? Work kind of consumed me and I don't write blog posts often, so sorry about that. Some things have happened in the world of Me(TM), so this is going to be a long one.
This is the most recent Happening, and possibly the most drastic one for me specifically.
Fedora had been having issues for a while. I initially said that the kernel no longer automatically gets copied to the boot partition and that DNF's system upgrade plugin didn't generate an initramfs, but it got worse over time. The behaviour Fedora exhibited, at the end of it all, was as follows:
I had put up with the sleep issues for years at that point and they'd come in all sorts of shapes and sizes, from GDM freezing to the most recent complete kernel panics. The degradation however seemingly continued to advance, and I really didn't want to reinstall it just for it to break again. I'd had enough.
So I picked up a cheap NVMe SSD (a Patriot P310 240GB) and installed Debian 12 Bookworm to test out. I'm glad I did as I learned a bunch.
I installed Debian 12 with Xfce for a change of pace. This generally went well but there were many snags I hit along the way.
Primarily it's that the Xfce variant lacks so many packages that I'd consider vital - for example, it lacks the GVFS packages to enable connection to SMB shares through Thunar by default. LibreOffice also was unable to access anything on an SMB share - I needed to install the INCREDIBLY poorly named 'libreoffice-gnome' package to get that working. Note that this also resolves SMB access problems with SoftMaker's FreeOffice.
Other than that I was learning the ways of Debian for the most part.
When the time came to install Debian for real, I made a final backup from my Fedora install and nuked it. During that time I got to experience the annoyance that is Debian's outright demand to use encrypted swap if you configure a swap partition alongside a LUKS encrypted root partition. I'm not joking - if you say no, it aborts the installation. Go fucking figure.
So okay, it generates a random key to use for the swap each boot and that should all be fine. Except it's not. On first boot (and all subsequent boots), cryptsetup gets confused as to the type of partition the swap actually is and keeps trying to determine it for a good ten seconds before assuming it's plain and just continuing on. This is a brute force method out of the box and adds a massive delay to the boot time of the system. What the hell is the point of using an SSD...
Initially I tried to stick with it by tweaking crypttab to specify that the partition type should just be 'plain', but it still added a good delay to the boot times, 4-5 seconds or so. Not great. Eventually I actually upgraded my boot SSD from a 500GB Samsung 980 to a 1TB Western Digital Black SN770 - so after I cloned it I removed the encrypted swap partition and all of its references, then created a 32GB swapfile. Booting post-unlock of the root partition is almost instantaneous now, and I don't have a weird swap partition either.
After that it has generally been plain sailing. I'm liking Debian's stability and while yes, it has its quirks, I can work around them relatively easily. I just wish there was a decent apt GUI frontend that isn't Synaptic or GNOME Software.
Debian devs - please use a goddamn swapfile. Using an encrypted swap partition feels like a bodge.
Another laptop, yes! This one's actually pretty recent, too. It's a 8th generation ThinkPad X1 Carbon. It comes in at just 1.09kg (2.4lbs for anyone using Imperial units) and is the lightest laptop I've ever used. I specifically wanted an ultrabook / thin-&-light after struggling carrying my 2015 MacBook Pro around at school. Its specs are as follows:
It's the newest laptop I have ever owned, aside from the Inspiron 14 5485 I had which was pretty terrible. And yeah, despite it being the 8th gen X1 Carbon it does sport a 10th generation U-series i7. And... well... DDR3? In 2020? On a FLAGSHIP ultrabook?! Lenovo, what the hell?
Yes yes, go ahead and hiss and screech, I put Windows 10 on this thing after trialling it with Linux. I honestly do need a Windows machine kicking around and I still like Windows 10 a lot, so I decided to put the LTSC release of 21H2 on it. It's very snappy but it's very behind on features, so it's caused some problems... like predating the Windows Store on desktop.
I had to search around for a script that got the store installed, so I could install the App Installer to be able to read MSIx packages so I could install Affinity Photo 2. Maybe Canva will get Serif to release a version for Linux at some point... please? Would be so good.
It's been fine. No issues aside from Windows being Windows. It's been ages since I used Windows 10, I definitely prefer it over Windows 11.
So... in August a bit of a catastrophic event happened. I'll let you in on a bit of a disturbing secret - all the data drives are connected to Europa, my primary server, over USB. Yep. And one of those cables (a cheap one that came with the Orico enclosures) went bad. And it just happened to be the one for the enclosure that contained the NVMe SSD with the Conduit RocksDB database on it.
Because the drive is LUKS encrypted, it did not get remounted. A subsequent reboot meant that anything in memory was lost - the database was unrecoverable, despite me trying to get it back with rocksdb-tools. The manifest file and the most recent changes simply were not saved to the disk. There were no backups - only one initial one made in April. All messages were lost.
The cable was replaced with a spare Ugreen cable I had on hand - there have been zero problems since. I'm now eyeballing replacing the cable connecting the Mastodon drive to the system.
I now also make nightly backups.
Sorry, Toko and Bard.
Earlier this month (Sept 2024) I evaluated my options for remote monitoring again having shut down my Zabbix server in March, during my exodus from Hetzner. I was looking at OpenNMS Horizon but it proved more annoying than expected to deal with, so I quickly abandoned it.
I settled back with Zabbix, this time version 7 LTS, on an Arm64 VM in Oracle's public cloud. Alerting is now set up along with monitoring of Systemd service units using v2 of the agent on registered machines. I also like graphs.
I actually chose MySQL instead of Postgres this time - during my first foray into Zabbix in February I used pSQL. In order to make it work, I had to make it use MD5 internally for authentication instead of SCRAM-SHA256 which I regarded as a massive bodge and didn't like. I wasn't aware if the issue had been fixed or not with version 7 of Zabbix (I still do not know) so I went with MySQL instead to simplify things.
Maybe that's not a very popular decision, and if you don't like it... dear god, get a grip. I discovered that there are MySQL and pSQL fanbeings today and I've never been more disappointed. Seriously, it's a relational database. Fucking hell.
Anyway, it's working well but alerting needs serious tuning. And I need to learn how to write expressions.
Today I did what I've been putting off since I received the NUC in December 2022, and that's replace its boot SSD. It had over 50TB of writes, which is a significant amount and I was becoming concerned.
It was a 250GB Samsung 970 Evo Plus so I was semi-confident it'd last until I got around to replacing it, and it did. I have replaced it with a cheaper and lower-end Western Digital Blue SN580 as I don't believe it's necessary to have such an expensive drive as long as a cheaper one is just as reliable. The SN580 was rated well and a good solid choice despite being a generally unimpressive upgrade over the SN570 it replaced in WD's lineup.
That should keep it chugging along for a good while now.
That's about it - everything else that's happened is generally not something I'd prefer to put here.
Thanks for reading. I'll be around.