If these are unit agents, then there are definitely some useful logs you could get us. Particularly if you are seeing the same unit agents running out of memory.
All juju agents have built in access to get the current goroutines and heap profiles.
Memory leaks in Juju tend to be one of two things: goroutine leaks, lost references to blocks of memory. The best way to find them is to look at goroutines and heap profiles over time.
To gather these, ssh into the machine and run the following, using unit-telegraf-0 as an example:
If these are unit agents, then there are definitely some useful logs you could get us. Particularly if you are seeing the same unit agents running out of memory.
All juju agents have built in access to get the current goroutines and heap profiles.
Memory leaks in Juju tend to be one of two things: goroutine leaks, lost references to blocks of memory. The best way to find them is to look at goroutines and heap profiles over time.
To gather these, ssh into the machine and run the following, using unit-telegraf-0 as an example:
$ juju_goroutines unit-telegraf-0 > goroutines.log
$ juju_heap_profile unit-telegraf-0 > heap-profile.log
Ideally include the date/time in the filename so we know which is earlier. Ideally at least an hour apart as this should make any leak more obvious.