Hi all,I had just been playing around a bit with a drop-in unit for the condor service and noticed, that the drop-in seems not to be inherited by the jobs' slice.
I had added a drop-in to the condor.service unit block some IPs [1]. nsenter'ing the condor master looks good. Laso as a test, I had set up a test unit with a drop-in to block traffic like [2], which works and the ping process in the unit could not access the blocked IPs.
However, this drop-in does not seem to get propagated to the actual job slice. I.e., we set the base cgroup for jobs like [3] and the job hierarchy is spawned underneath it (separately from condor.service). But when I run a test job or condor_ssh into a slot, the IPs are not blocked and I am able to reach the IPs denied for the actual condor service.
So, I guess that the drop-in is only evaluated by systemd for starting the condor service unit, but as the job slice is realized by condor itself, the drop-in additions are not actually inherited, or?
I am not sure, if I would actually be able to use systemd unit semantics in some way to influence the job slice?
Cheers, Thomas [1] > cat condor.service.d/02-condor-network-filters.conf [Service] IPAccounting=yes #IPAddressDeny=DESYIPs IPAddressDeny=8.8.8.8 8.8.4.4 [2] > cat ip-deny-ping.service [Service] ExecStart=/usr/bin/ping 8.8.4.4 IPAccounting=yes > cat ip-deny-ping.service.d/01-network-filter.conf [Service] # IPAddressDeny=8.8.8.8 8.8.4.4 2001:4860:4860::8888 2001:4860:4860::8844 IPAddressDeny=8.8.8.8 2001:4860:4860::8888 2001:4860:4860::8844 [3] condor_config_val BASE_CGROUP system.slice/condor.slice
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature