Pipelines
Run history, triggers and run detail (task DAG, logs) of the Train, Register, Deploy and CT pipelines.
Required permission: View
Each pipeline kind (Train, Register, Deploy, CT — Continuous Training) runs one configured container image; this screen runs it and shows its history. One run is one Kubernetes Job in the tenant namespace. The image per kind is set by the server setting pipeline_images (GEO_MLOPS_PIPELINE_IMAGES); the default is the placeholder busybox:1.36. Training runs from the Training screen and serving image builds do not appear here. Every member can view; triggering, rerunning and stopping need operate (OPERATOR, ADMIN).
| # | Area | Description |
|---|---|---|
| ① | Pipeline tabs | Train (Train) · Register (Register) · Deploy (Deploy) · CT (M3+) (CT (M3+)). Each tab shows the run history of that kind. |
| ② | New trigger (새 트리거) | Creates one Job from the current tab's image (operate). The screen passes no parameters. |
| ③ | Run history | Run (Job name), status, type, start, end. Up to the 200 most recent Jobs, newest first, refreshed every 15 seconds. The cluster deletes finished Jobs a day later, and they disappear from the list too. Click a row to open the run detail. |
Run detail
The top of the run detail shows the Job name, the status badge, and Rerun (재실행) · Stop (중단) (operate). One run is one Job, so Progress (진행률) reads "0/1 tasks completed (0/1 태스크 완료)" or "1/1 tasks completed (100%)" (1/1 also when it ended in failure), and the Tasks (DAG) (태스크 (DAG)) diagram holds a single Job node. Click the node to see the Job's container log; download it with Download (다운로드). A run in progress refreshes every 5 seconds.
| Button | Result |
|---|---|
| Rerun (재실행) | Creates a new Job with the same image and settings. The original run stays. |
| Stop (중단) | Deletes the Job and its pods without a confirmation dialog. The deleted run disappears from the list. |
Main tasks
- Choose the tab of the pipeline to run.
- Press New trigger (새 트리거). A "Pipeline triggered (파이프라인 트리거됨)" notice appears and a new row is added to the history.
- Click the new row to see progress and logs.
Statuses and badges
The status is one of four values the server derives from the Kubernetes Job status, shown in English. Jobs are not retried, so a single pod failure makes the run Failed.
| Value | Color | Meaning |
|---|---|---|
| Pending | Yellow | No pod running yet (waiting for the pod to start) |
| Running | Blue | Running |
| Succeeded | Green | Ended normally |
| Failed | Red | Failed — see the task log for the cause |