While working on the inclusion of the large test suite (see freedomofpress/dangerzone-image#13), we found out that there are some test files that can cause the memory of the sandbox to increase, to a point where the machine becomes non-operational. This manifests both on the CI and in our local machines.
I'm not exactly sure if this was manifesting in Dangerzone 0.10.0, because the way we run conversions in this repo is different that in the Dangerzone repo. For example, the Dangerzone app makes sure to shut down all processes that belong to the same Podman process group. In any case, it seems we need to make the following changes to this repo:
- Add a memory limit in our gVisor sandbox, that will kill the container if it reaches past it. With our large test suite, we can experiment with a limit and be a bit more confident that we won't affect actual workloads, only problematic ones.
- Extend our test harness in this repo to kill all the Podman/gVisor processes, after a test fails, e.g., due to a test timeout.
While working on the inclusion of the large test suite (see freedomofpress/dangerzone-image#13), we found out that there are some test files that can cause the memory of the sandbox to increase, to a point where the machine becomes non-operational. This manifests both on the CI and in our local machines.
I'm not exactly sure if this was manifesting in Dangerzone 0.10.0, because the way we run conversions in this repo is different that in the Dangerzone repo. For example, the Dangerzone app makes sure to shut down all processes that belong to the same Podman process group. In any case, it seems we need to make the following changes to this repo: