[newrelic-logging] Adjust order of env list in daemonset to fix dependency reference #1441
Conversation
When using persistentVolume mode, the env variable FB_DB references NODE_NAME. This reference is currently not resolved, because NODE_NAME is defined after FB_DB. This commit fixes this, by moving the definition of NODE_NAME before the definition of FB_DB.
|
|
There was a problem hiding this comment.
Hi @nluedema ,
Thank you for identifying this issue. Indeed, I didn't realize that moving NODE_NAME would affect FB_DB when using persistent volumes due to ordering. The reason this is happening though is because that environment variable uses $(NODE_NAME) instead of ${NODE_NAME}, which means that it is Helm resolving the value of these variables rather than Fluent Bit when it starts.
To fix this, I would suggest to leave NODE_NAME untouched and instead fix FB_DB. We don't want environment variable definition to be dependent on order. To achieve this, could you please:
- Move
NODE_NAMEback to where it was - Modify this:
- name: FB_DB
value: "/db/$(NODE_NAME)-fb.db"
to this:
- name: FB_DB
value: "/db/${NODE_NAME}-fb.db"
This will ensure that environment variable definitions stay order-insensitive and that it is actually the Fluent Bit process that resolves NODE_NAME upon start rather than Helm.
This reverts commit 3b7a1b5.
This reverts commit 192cba5.
…on order The FB_DB environment variable depends on the environment variable NODE_NAME. This variable is currently resolved by Helm, which makes FB_DB depend on the environment variable definition order. To fix this, we make NODE_NAME being resolved by Fluent Bit instead of Helm.
|
Hi @jsubirat, Thanks for the feedback! I didn't know that Fluent Bit is able to resolve environment variables. Doing it that way is of course a lot more elegant. I updated the PR according to your suggestions. |
Is this a new chart
No
What this PR does / why we need it:
When using
persistentVolumemode, the env variableFB_DBreferencesNODE_NAME. This reference is currently not resolved, becauseNODE_NAMEis defined afterFB_DB. Therefore, only a single database file named'$(NODE_NAME)-fb.db'is created in the persistent volume, instead of one file for each node. #1408 traced the problem down to this change, which moved the definition ofNODE_NAMEafter the definition ofFB_DB.This PR fixes this, by moving the definition of
NODE_NAMEbefore the definition ofFB_DBagain.The behavior of kubernetes for dependent environment variables is documented here.
Which issue this PR fixes
Special notes for your reviewer:
I was thinking about a test that checks if
NODE_NAMEis defined beforeFB_DB, but I wasn't able to come up with a way to do this usinghelm-unittest.Checklist
[mychartname])