You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
At present, gitops-engine (and consumers of gitops-engine API) use /dev/shm to store temporary files that are passed to kubectl, such as manifests and cluster credentials. The temp directory logic can be found within pkg/utils/io/io.go.
One disadvantage of this is that since /dev/shm is backed by RAM, it has limited size. For many use cases this is not an issue, however, when concurrently deploying to a large number of Kubernetes clusters, or deploying large k8s manifests, /dev/shm can run out of disk space.
One option is that we could just move to using /tmp, rather than /dev/shm. But, it appears that /dev/shm was chosen intentionally to ensure that sensitive data (such as credentials) are kept off disk (absent disk caching of memory).
Thus, rather than moving to use /tmp, I propose instead allowing API consumers to optionally use a different path, which can then by enabled by API consumers (for example, Argo CD configmap value or env var)
I've also been tackling this problem by reducing the leaking of files into /dev/shm. You can see that work below.
Related issues:
In 'ManageResources' and 'runResourceCommand', temporary files written to '/dev/shm' may not be cleaned up, in rare cases
At present, gitops-engine (and consumers of gitops-engine API) use
/dev/shm
to store temporary files that are passed to kubectl, such as manifests and cluster credentials. The temp directory logic can be found withinpkg/utils/io/io.go
.One disadvantage of this is that since
/dev/shm
is backed by RAM, it has limited size. For many use cases this is not an issue, however, when concurrently deploying to a large number of Kubernetes clusters, or deploying large k8s manifests,/dev/shm
can run out of disk space.One option is that we could just move to using
/tmp
, rather than/dev/shm
. But, it appears that/dev/shm
was chosen intentionally to ensure that sensitive data (such as credentials) are kept off disk (absent disk caching of memory).Thus, rather than moving to use
/tmp
, I propose instead allowing API consumers to optionally use a different path, which can then by enabled by API consumers (for example, Argo CD configmap value or env var)I've also been tackling this problem by reducing the leaking of files into
/dev/shm
. You can see that work below.Related issues:
The text was updated successfully, but these errors were encountered: