Change a model's replica count without editing a file on the control plane

The count is scaling.min, and until now the only way to change it was to edit
server.yaml on the host and restart — or, for a model added at runtime, to
remove it and add it back, which means having its whole definition to hand.

`llmgw ctl scale <model> --min N` instead. Only the scaling block moves:
everything else is read back from what was stored and written out untouched, so
changing a number cannot quietly become a redefinition.

The catalog is rebuilt and validated before anything is written, as in Add. A
max below a min is refused with the stored definition untouched — persisting a
rejected change would come back on the next restart as a catalog that does not
build.

The answer reports the scaling in force rather than the request. The catalog
fills in a max that was left out, and a caller echoing its own request would
print a number the cluster is not using.

A model declared in configuration is refused, for the reason Remove already
refuses one: the YAML is read on every start, so the new count would hold until
the next restart and then silently revert. A cluster disagreeing with its own
config file is worse than an error saying where to edit.

The help says what --max does, because it is not what it looks like: a model
with a minimum runs exactly that many and nothing scales it up under load, and
an on-demand model loads one replica however high its max.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Ilker ManapandClaude Opus 5 committed 2026-09-27 11:25:05 +02:00
1 parent 10e323132c
commit 707d774c76
6 files changed
+388

No files matched your search

+57
View File
@@ -4,12 +4,14 @@ import (
"encoding/json"
"fmt"
"os/user"
"time"
"github.com/knadh/koanf/parsers/yaml"
"github.com/knadh/koanf/providers/file"
"github.com/knadh/koanf/v2"
"github.com/spf13/cobra"
"gitea.mynodes.xyz/ilker/llmgw/internal/control/adminapi"
"gitea.mynodes.xyz/ilker/llmgw/internal/control/catalog"
)
@@ -128,3 +130,58 @@ func actor() string {
}
return "unknown"
}
func newScaleCmd(client func() *ctlClient) *cobra.Command {
var min, max int
var idle time.Duration
c := &cobra.Command{
Use: "scale <model> --min N [--max M]",
Short: "Change how many replicas a model runs",
Long: "The replica count is --min. --max is a ceiling, not a target: a model\n" +
"with a minimum runs exactly that many, and nothing scales it up under\n" +
"load. Leaving --max out makes it follow --min.\n\n" +
"--min 0 with a --max above it is the on-demand model: nothing runs until\n" +
"somebody asks, and it stops again after the idle timeout. That loads one\n" +
"replica however high --max is. --min 0 --max 0 is a model kept in the\n" +
"catalog and deliberately not running.\n\n" +
"Only models added at runtime. One declared in server.yaml is read again\n" +
"on every start, so a number set here would hold until the next restart\n" +
"and then revert — change it in the file instead.\n\n" +
"It takes effect on the reconciler's next pass, and survives a restart.",
Args: cobra.ExactArgs(1),
RunE: func(cmd *cobra.Command, args []string) error {
if !cmd.Flags().Changed("min") {
return fmt.Errorf("--min is required: it is the replica count")
}
body, err := json.Marshal(adminapi.ScaleRequest{Min: min, Max: max, IdleTimeout: idle})
if err != nil {
return err
}
var res adminapi.ScaleResponse
if err := client().post(cmd.Context(),
"/api/v1/models/"+args[0]+"/scale", body, &res); err != nil {
return err
}
// What is in force, not what was asked for: the catalog fills in a
// max that was left out, and printing the request back would show a
// number the cluster is not using.
out := cmd.OutOrStdout()
switch {
case res.OnDemand:
fmt.Fprintf(out, "%s is on-demand: loaded when asked for, up to %d, stopped when idle.\n",
res.Model, res.Max)
case res.Min == 0:
fmt.Fprintf(out, "%s will not be placed; it stays in the catalog.\n", res.Model)
default:
fmt.Fprintf(out, "%s: %d replicas (max %d). It takes effect on the next pass; watch it with:\n"+
" llmgw ctl models\n", res.Model, res.Min, res.Max)
}
return nil
},
}
c.Flags().IntVar(&min, "min", 0, "replicas to keep running; this is the count")
c.Flags().IntVar(&max, "max", 0, "ceiling for an on-demand model; follows --min when left out")
c.Flags().DurationVar(&idle, "idle-timeout", 0,
"how long an on-demand model stays up unasked-for; zero takes the cluster default")
return c
}