Service Monitors and Compactor Service #68
Conversation
ts-mini
force-pushed
the
servicemonitors
branch
from
af272c5
to
e3b2884
Compare
ts-mini
force-pushed
the
servicemonitors
branch
3 times, most recently
from
36f0d5e
to
8409990
Compare
Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
ts-mini
force-pushed
the
servicemonitors
branch
from
8409990
to
474c5a8
Compare
Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
| @@ -85,6 +85,8 @@ spec: | |||
| {{- toYaml .Values.alertmanager.tolerations | nindent 8 }} | |||
| terminationGracePeriodSeconds: {{ .Values.alertmanager.terminationGracePeriodSeconds }} | |||
| volumes: | |||
| - emptyDir: {} | |||
There was a problem hiding this comment.
I would recommend avoiding specifying non-cortex path mounts here. I get the intention, but it prevents others from customizing /tmp in this setup. There are overridable settings for volumes and mounts, which allow for this.
There was a problem hiding this comment.
@ts-mini - Sorry, should have flagged this earlier, but I think this last item needs correction prior to merge
There was a problem hiding this comment.
sounds good I'll get to this asap
There was a problem hiding this comment.
if I remember correctly, it's cause the code inside of cortex writes to tmp no matter what, and it's not customizable, so if you use the default psp the file system is ro, making it fail when it validates configs. I think I should be able to make it customizable, and adjust to tmp in our setup
There was a problem hiding this comment.
Yes, that might be fair that cortex uses /tmp in some cases (I'll have to go look where again), but in the PR you've fixed that solution to be an emptyDir. If others are already mapping /tmp to other volume types, this will potentially break their setup.
Also correcting any path mounts should be in its own PR, given the title of this one is adding service monitors. Both valid issues, but it makes it much easier to review, provide feedback and merge when PRs are focused on a single scope.
There was a problem hiding this comment.
Honestly I look back at this change and idk what I was thinking. 1) sorry for conflating prs, 2) I totally glossed over the 'extraVolume' and 'extraVolumeMount' options. I am just going to use those going forward - and while I think we could solve this ro-filesystem issue within the chart - the real "problem" imho is this section. I might just attempt to make this change on the cortex side so one could specify the 'tmpDIr' path as a config param.
There was a problem hiding this comment.
Thank you for being open to the feedback and adjusting the changes :) LGTM now for merging
Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
* updating alertmanager and servicemonitors Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com> * adding clusterDomain back Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com> * adding - back Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com> * adding scope to clusterDomain Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com> * revert the tmp filesystem changes in this pr Signed-off-by: Tyler Horvath <tyler.horvath@gmail.com>
Seeing as this is included by default in the Loki chart - I figure this might not be too opinionated of an addition.
I updated the docs as well
edit: I discovered a alertmanager clustering bug (in the replicas code) that I also fixed