CLI

The Kubernetes integration includes a small operations CLI for inspecting and controlling active jobs and inference model catalog entries.

Commands are run on the host where the Integration Module is installed:

$ ucloud <command> <subcommand> [options]

The job commands are available under ucloud jobs .... Inference model catalog commands are available under ucloud inference models ....

Notes

Commands must be run as root.
Query flags use regular expressions unless otherwise noted.
The CLI shows active jobs tracked by the Integration Module (non-final jobs).
Most commands support --json for machine-readable output.

`jobs` command

`ucloud jobs ls`

Lists active jobs.

Optional filters:

--state <regex>: Match UCloud job state.
--job-id <regex>: Match UCloud job ID.
--user <regex>: Match submitting UCloud username.
--project <regex>: Match UCloud project ID.
--application <regex>: Match application name.
--category <regex>: Match machine category.
--queue <regex>: Match scheduler queue name.
--json: Print JSON output.

Aliases: list.

`ucloud jobs get <jobId>`

Shows detailed information for one job.

Options:

--json: Print JSON output.

Aliases: retrieve, stat, view.

`ucloud jobs queue [queue]`

Shows queue insight for all queues (or a selected queue), including:

queued job count
running job count
queue/product availability statuses
queued jobs per queue (when enabled)

Options:

--queue <regex>: Filter queue name.
--jobs=<true|false>: Include/exclude queued job listings (default: true).
--json: Print JSON output.

If a positional queue argument is given, it is treated as an exact queue-name match.

Alias: queues.

`ucloud jobs queue debug [queue]`

Shows internal scheduler state intended for debugging and operations.

The output includes, per scheduler queue:

queue tick/time
all known node entries (including remaining/capacity/limits and unschedulable flag)
individual queue entries (priority and factor values)
individual replica entries (job, rank, node, dimensions)

Options:

--queue <regex>: Filter queue name.
--json: Print JSON output.

Alias: ucloud jobs queue state [queue].

`ucloud jobs kill <jobId...>`

Stops one or more jobs.

Behavior by backend:

Virtual machine jobs are suspended.
Container jobs are terminated.

You can pass multiple IDs in one command.

Aliases:

stop
rm, remove, delete, del

`ucloud jobs suspend <jobId...>`

Suspends one or more virtual machine jobs.

If a listed job is not a virtual machine job, the command reports an error for that job.

`ucloud jobs unsuspend <jobId...>`

Unsuspends one or more virtual machine jobs.

If a listed job is not a virtual machine job, the command reports an error for that job.

Alias: resume.

`inference models` command

The inference model commands manage the model catalog stored by the Integration Module. They do not query the backing inference platform directly.

`ucloud inference models ls`

Lists inference model catalog entries, including name, title, capabilities, pricing multipliers, endpoint information, and availability.

Options:

--json: Print JSON output.

Alias: list.

`ucloud inference models update <name>`

Updates an existing inference model catalog entry.

Options:

--name <name>: Rename the model. Renaming is only allowed while the model is not public.
--title <title>: Set the display title.
--capabilities <list>: Set a comma-separated capability list. Allowed values are TextGeneration, TextToImage, and SpeechToText.
--price-cached <n>: Set cached input token price multiplier in fixed-point thousandths.
--price-input <n>: Set input token price multiplier in fixed-point thousandths.
--price-output <n>: Set output token price multiplier in fixed-point thousandths.
--public <bool>: Set whether the model is publicly available.
--available-to <list>: Set a comma-separated list of project IDs allowed to use the model when it is not public.
--base-path <url>: Set the OpenAI-compatible endpoint base path for the backing model.
--backend-model-name <name>: Set the model name sent to the backing inference provider.

Price multipliers must be non-negative integers. A value of 1000 means 1x; 500 means 0.5x.

`ucloud inference models rm <name>`

Removes an inference model catalog entry. Removal only affects the Integration Module catalog and does not depend on whether the backing inference platform still lists the model.

Aliases: remove, delete, del.

Examples

$ ucloud jobs ls
$ ucloud jobs ls --state IN_QUEUE --category gpu-a10
$ ucloud jobs get 1234
$ ucloud jobs queue
$ ucloud jobs queue cpu-standard/general
$ ucloud jobs queue --queue 'gpu.*/.*' --jobs=false
$ ucloud jobs queue debug
$ ucloud jobs queue state gpu-a10/a10
$ ucloud jobs stop 1234 5678
$ ucloud jobs suspend 1234
$ ucloud jobs resume 1234
$ ucloud inference models ls
$ ucloud inference models update qwen --title "Qwen" --capabilities TextGeneration --price-input 1000 --price-output 1000
$ ucloud inference models update qwen --name qwen-public --public true
$ ucloud inference models rm qwen-public