Cache directory limit, leading to OOM #977
Description
Echoes #976
The new cache directory for storing OCI artifacts is now out of the replicated volume (at least we hope it is). This means that if the hosting node is out of memory, we can end up with OOM that will further propagate to the adjacent systems, leading to partial DoS of the overall system.
This would be very inconvenient 😬
For this reason, a limit to the size of the ephemeral volume created by the deploy/ directory would be a good thing !
Moreover, testing what happens to the API operations if this happens during one --especially ensuring consistency-- is now to check, manually first then with automated functional tests.