2026–W05 #
Back on BGP for HA -- I'm starting to think that the requirements to run several BGP daemons is making my plan worse. I'm better off scripting what BIRD does, or manipulating a static table of my own to re-announce into BGP? The allure of exabgp/gobgp was there, alright.
Update: paused the BGP part, but I did work on a new website for Colocataires that I'm enjoying. Straight up and HTML and CSS with no frameworks and I'm pulling off some tricks to keep the HTML semantics but still look like an 80x25 glass teletype.
This was also a period of huge churn in the homelab, made worse because every "move" requires a long running Ceph operation to run and then calm down before the next. I have purged SATA from the Ceph cluster, storage is in three tiers: 7.2K LFF SAS, 10K SFF SAS and NVMe. VMs are on NVMe, RBD on 10K SAS and CephFS on bulk storage. I've also learned that CephFS metadata is kind-of problematic with millions of small files / inodes, and so it's not very well suited to backing up whole Linux machines (especially de-duplicated at the inode level with hard links).
I also replaced the switches: the Aruba and UniFi are replaced with one Brocade 7250 with 8x10Gs and 48x1G ports and PoE. It's loud though. Glad I don't work next to it.
I dislike that storage stuff seems to eat the time poured into it so I would like to leave this alone for a little bit an concentrate on other personal projects, along with Colocatairies/OttawaSystems stuff in my down time.