WIP: Add Task names to Pod names#1320
Conversation
|
Probably need to check the size of the resulting name to ensure it is below 253 characters as well. |
149c797 to
a43fb44
Compare
|
Good call. Name length is fixed in the last push. |
18eacca to
de9d989
Compare
|
Here's a before/after example of this feature: Before After One notable side effect of this change is that the pods are always listed in step-order. |
de9d989 to
3690cc9
Compare
|
What is this blocked on? I think this is a great addition. |
3690cc9 to
aacd99c
Compare
|
Hey @AudriusButkevicius, I honestly can't recall why I stopped working on this - it's definitely a more involved change than I at first thought it would be. I've rebased my branch, and I'll take another swing at it as soon as I find some time to do so. |
|
I am happy to help out if its clear what the remaining work is, but it seems PRs on this project aren't getting that much attention. |
8652897 to
330573b
Compare
|
@jessesuen @sarabala1979 I believe this is ready to review |
330573b to
673aa0d
Compare
|
Can you keep the same naming convention? can you change the |
673aa0d to
05e6562
Compare
|
@sarabala1979 done. |
|
@sarabala1979 could you take another look at this when you get a moment? |
|
LGTM. Resolve the conflicts |
This changes the way that Argo names pods. The new behavior will now create a display the parent-child relationships between nodes in the Workflow. For example, if Workflow dag-nested has a DAG named "a", which in turn has a sub-DAG named "b", the resultant Pod will be named "dag-nested-a-b-XXXXXXXXXX" where the last section is a unique hash. The argo CLI tool's "get" and "logs" commands have also been updated to reflect the new behavior.
05e6562 to
36116a8
Compare
|
@sarabala1979 Conflicts are resolved. Here's the message from the gate failure: I'm not sure how to fix this one without increasing the deadline on the linter... |
|
@ian-howell is this still wanted please? |
|
Added |
Seems like this PR is done and even approved, it just was never merged. @ian-howell do you think you could resolve the conflicts? |
|
@simster7 +1 |
|
Chiming in - this is a highly desired feature, and it has huge advantages for downstream metrics and monitoring, given podnames are a first class citizen in Grafana, GKE logs, Prometheus, Stackdriver, etc. I strongly support adding it and I hope to see it in a release |
|
+1, if (as seems) this feature exists in an MR and is ready to go it would be a very useful and addition! |
|
+1. We really need this in our project. Is there any other way in Argo to change the name of the pods being started by templates? |
|
I’ll take a look into this early next week |
|
Hi Simon, did you have a chance to have a look ever since? I was also looking for this feature. |
|
I need to add a +1 on this, since fine-grained pod monitoring with prometheus is impossible when all the pods in a workflow are named random jargon that does not differentiate what task they are performing. |
|
+1 on this, it will ease the monitoring process. |
|
Better to thumbs-up the issue, not this closed PR. |
|
Hello, what happened with this? It's a very useful feature to have, I am struggling to configure Grafana to observe some specific steps without having their names in the pod names |
|
this feature has causes a lot of upstream issues outside of Argo, particularly syncing of the kube downward API and etcd. I mentioned this in various open threads but pinging here so more engineers that worked on this feature are aware. This should be reverted until a real solution is created, not continuous patches. |
associated with. This makes it easier to keep track of which pod is
performing which task.
Still a work in progress - unit tests are failing, and there's probably a better way to deal with parentheses.
This would close #1319 if it is accepted.
Let me know what you think :)