Added pod affinity if the default ReadWriteOnce is used. #13
Conversation
|
I am in progress of testing this, putting it here already for transparency. |
keskival
force-pushed
the
mastodon/22095/pod-affinity-for-readwriteonce
branch
4 times, most recently
from
56b9f32
to
e51d535
Compare
|
We literally just removed this in order to reduce unexpected affinities by just selecting a different volume access mode. I guess the question becomes towards what end of the spectrum we want tonstart optimising the chart. Do we want to keep it generic and unopinionated or move towards an ideal, predefined setup? See this conversation: mastodon/mastodon#20733 (comment) |
keskival
force-pushed
the
mastodon/22095/pod-affinity-for-readwriteonce
branch
from
e51d535
to
9a54181
Compare
|
Ok fair enough, but the default configuration isn't functional, and the jobs have analogous affinities set up. In any case, this has been tested as far as template generation goes and produces templates in the expected form. If I am fine with adding a visible warning to the This is now confirmed working in my own cluster as well. Previously the defaults caused it to become locked up in a difficult to understand manner, which is what this PR fixes. Added S3 conditioned disabling as well. |
keskival
force-pushed
the
mastodon/22095/pod-affinity-for-readwriteonce
branch
3 times, most recently
from
b20acd7
to
2f3fe27
Compare
If `accessMode` `ReadWriteOnce` is used, then the volume can only be accessed in a write mode from a single node. This means that all pods which require write access to the volume need to be co-located to the same node. Failing that typically leads to a difficult to untangle locked volume mount situation where containers are stuck in `ContainerCreating`. This change adds the necessary pod affinities if the related access modes are set in the `values.yaml`.
keskival
force-pushed
the
mastodon/22095/pod-affinity-for-readwriteonce
branch
from
2f3fe27
to
682a5a4
Compare
If
accessModeReadWriteOnceis used, then the volume can only be accessed in a write mode from a single node. This means that all pods which require write access to the volume need to be co-located to the same node. Failing that typically leads to a difficult to untangle locked volume mount situation where containers are stuck inContainerCreating.This change adds the necessary pod affinities if the related access modes are set in the
values.yaml.