Over the weekend thanos-fe2001 ran out of disk space on /. Turns out /var/spool/rsyslog/ was full of queue files such as:
-rw------- 1 root root 1049237 May 16 11:14 centrallog1001.eqiad.wmnet:6514.00093952 -rw------- 1 root root 606208 May 16 12:11 centrallog2002.codfw.wmnet:6514.00000977
The same thing was happening on thanos-fe2002, though the filesystem wasn't full yet. It looked like the transport to centrallog hosts was stuck somehow, I'm assuming post-reboot for Bullseye upgrades. I've restarted rsyslog on thanos-fe2002 and it looks like it is flushing its queue now.