Files
llmgw/internal/cli/agent.go
T
ilkerandClaude Opus 5 6a13e3dddf Phase 2 (1/2): node discovery, engine adapters, process supervisor
Everything here works without GPUs and is covered by tests. The control plane
connection is not built yet, so llmgw agent itself is still a stub and the phase
exit criterion is not met.

- discovery: IB from /sys/class/infiniband, parsing the formats sysfs actually
  uses and telling native InfiniBand from RoCE; GPUs via nvidia-smi, the
  cgo-free path, with health flags for uncorrected ECC and memory already held
  by another process. A node without IB is not an error, it is a single-node
  island.
- engines: vLLM and SGLang command and environment construction, kept free of
  process handling so it stays testable. The NCCL environment is the part the
  plan marks critical: NCCL_IB_HCA lists only the chosen fabric's adapters,
  because leaving it open lets NCCL pick an adapter on a fabric where the peer
  is unreachable, and the job then hangs instead of failing.
- validation before launch: no mixed GPU types in a tensor-parallel group, the
  node/GPU/TP arithmetic has to add up, and fp8 is refused below compute
  capability 8.9 rather than silently falling back to a slower path.
- supervisor: process groups, because engines fork workers that hold GPU memory
  and outlive a signal sent only to the leader; a bounded log tail, since engine
  output is the only diagnosis when loading fails; Starting separate from
  Loading, because a 70B model legitimately takes minutes; and orphan adoption
  through pid files fingerprinted with the command line, since pids get reused
  and adopting an unrelated process is worse than leaking the original.
- readiness is two questions: /health says the process is up, /v1/models says
  the weights loaded. Routing on /health alone sends traffic into a loading model.
- llmgw agent inventory reports what the scheduler would see on a node.

One plan correction: GPU class sizes come from the driver in GiB, which does not
always match the marketing name. An L40S is sold as 48GB and reports 44.99 GiB,
so its class is l40s-45g, not l40s-48g. Consistency matters more than matching
the box, because that number decides whether a model fits. Plan §2 and §5.1
updated.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-16 02:01:59 +02:00

126 lines
3.7 KiB
Go

package cli
import (
"encoding/json"
"fmt"
"os"
"text/tabwriter"
"github.com/spf13/cobra"
"gitea.mynodes.xyz/ilker/llmgw/internal/agent/discovery"
)
func newAgentCmd() *cobra.Command {
var configPath string
cmd := &cobra.Command{
Use: "agent",
Short: "Node agent: discovery and engine supervision (phase 2)",
Long: "Runs on every GPU node. It discovers the hardware, starts and supervises\n" +
"inference engines, and reports to the control plane.\n\n" +
"The control plane connection is not built yet, so for now the useful\n" +
"subcommand is `agent inventory`.",
RunE: func(_ *cobra.Command, _ []string) error {
return errNotImplemented
},
}
cmd.Flags().StringVar(&configPath, "config", "agent.yaml", "configuration file")
cmd.AddCommand(newInventoryCmd())
return cmd
}
func newInventoryCmd() *cobra.Command {
var (
asJSON bool
smiBin string
ibRoot string
skipGPU bool
)
cmd := &cobra.Command{
Use: "inventory",
Short: "Show this node's GPUs and InfiniBand adapters",
Long: "Reports what the scheduler would see on this node: GPU classes, health,\n" +
"and the fabric each adapter is attached to.\n\n" +
"Run it before adding a node to the cluster — it is the fastest way to find\n" +
"a card another process is already using, or a port with no subnet manager.",
RunE: func(cmd *cobra.Command, _ []string) error {
inv := discovery.Inventory{Source: "nvidia-smi"}
if !skipGPU {
gpus, err := discovery.DiscoverGPUsSMI(cmd.Context(), smiBin)
if err != nil {
// Not fatal: a node with no GPUs still has IB worth showing,
// and the error itself is the useful output.
fmt.Fprintf(os.Stderr, "GPU discovery failed: %v\n", err)
inv.Source = "unavailable"
}
inv.GPUs = gpus
}
hcas, err := discovery.DiscoverIB(ibRoot)
if err != nil {
return err
}
inv.HCAs = hcas
if asJSON {
enc := json.NewEncoder(os.Stdout)
enc.SetIndent("", " ")
return enc.Encode(inv)
}
printInventory(inv)
return nil
},
}
f := cmd.Flags()
f.BoolVar(&asJSON, "json", false, "emit JSON")
f.StringVar(&smiBin, "nvidia-smi", "", "path to nvidia-smi")
f.StringVar(&ibRoot, "ib-root", discovery.SysfsRoot, "sysfs InfiniBand directory")
f.BoolVar(&skipGPU, "no-gpu", false, "skip GPU discovery")
return cmd
}
func printInventory(inv discovery.Inventory) {
w := tabwriter.NewWriter(os.Stdout, 0, 0, 2, ' ', 0)
fmt.Fprintf(w, "GPUs (%d)\n", len(inv.GPUs))
if len(inv.GPUs) == 0 {
fmt.Fprintln(w, " none")
} else {
fmt.Fprintln(w, " IDX\tCLASS\tNAME\tMEM\tCC\tHEALTH")
for _, g := range inv.GPUs {
health := "ok"
if !g.Health.OK() {
health = g.Health.Problems[0]
}
fmt.Fprintf(w, " %d\t%s\t%s\t%d MiB\t%s\t%s\n",
g.Index, g.Class(), g.Name, g.MemoryTotalMB, g.ComputeCapability, health)
}
fmt.Fprintf(w, " classes: %v\n", inv.Classes())
}
fmt.Fprintf(w, "\nInfiniBand adapters (%d)\n", len(inv.HCAs))
if len(inv.HCAs) == 0 {
// Worth stating plainly: such a node is its own single-node island and
// can never host a multi-node replica.
fmt.Fprintln(w, " none — this node cannot host multi-node replicas")
} else {
fmt.Fprintln(w, " HCA\tPORT\tSTATE\tRATE\tLINK\tGID PREFIX\tNUMA")
for _, h := range inv.HCAs {
for _, p := range h.Ports {
fmt.Fprintf(w, " %s\t%d\t%s\t%d Gb/s\t%s\t%s\t%d\n",
h.Name, p.Num, p.State, p.RateGbps, p.LinkLayer, p.GIDPrefix, h.NUMANode)
}
}
fmt.Fprintln(w, "\n Fabric identity is not derived from the GID prefix by default:")
fmt.Fprintln(w, " OpenSM's default is fe80:: on every fabric, so a shared prefix proves")
fmt.Fprintln(w, " nothing. Label fabrics explicitly in agent.yaml (plan §5.2).")
}
_ = w.Flush()
}