CI / test (push) Successful in 45s
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>
104 lines
2.8 KiB
Go
104 lines
2.8 KiB
Go
package discovery
|
|
|
|
import (
|
|
"context"
|
|
"encoding/csv"
|
|
"fmt"
|
|
"os/exec"
|
|
"strconv"
|
|
"strings"
|
|
)
|
|
|
|
// smiQueryFields are the columns requested from nvidia-smi, in order.
|
|
var smiQueryFields = []string{
|
|
"index",
|
|
"uuid",
|
|
"name",
|
|
"memory.total",
|
|
"memory.used",
|
|
"compute_cap",
|
|
"pci.bus_id",
|
|
"driver_version",
|
|
"temperature.gpu",
|
|
"ecc.errors.uncorrected.volatile.total",
|
|
}
|
|
|
|
// DiscoverGPUsSMI enumerates GPUs by shelling out to nvidia-smi.
|
|
//
|
|
// This is the cgo-free path (plan §3.2): the NVML bindings need cgo, which
|
|
// would force the whole binary — server included — to be built with a C
|
|
// toolchain. nvidia-smi gives everything except NVLink topology, which the
|
|
// NVML build adds.
|
|
func DiscoverGPUsSMI(ctx context.Context, binary string) ([]GPU, error) {
|
|
if binary == "" {
|
|
binary = "nvidia-smi"
|
|
}
|
|
out, err := exec.CommandContext(ctx, binary,
|
|
"--query-gpu="+strings.Join(smiQueryFields, ","),
|
|
"--format=csv,noheader,nounits",
|
|
).Output()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("running %s: %w", binary, err)
|
|
}
|
|
return parseSMI(string(out))
|
|
}
|
|
|
|
func parseSMI(out string) ([]GPU, error) {
|
|
r := csv.NewReader(strings.NewReader(out))
|
|
r.TrimLeadingSpace = true
|
|
r.FieldsPerRecord = -1 // tolerate driver versions with extra columns
|
|
|
|
rows, err := r.ReadAll()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("parsing nvidia-smi output: %w", err)
|
|
}
|
|
|
|
gpus := make([]GPU, 0, len(rows))
|
|
for _, row := range rows {
|
|
if len(row) < len(smiQueryFields) {
|
|
continue
|
|
}
|
|
idx, err := strconv.Atoi(strings.TrimSpace(row[0]))
|
|
if err != nil {
|
|
continue
|
|
}
|
|
|
|
g := GPU{
|
|
Index: idx,
|
|
UUID: strings.TrimSpace(row[1]),
|
|
Name: strings.TrimSpace(row[2]),
|
|
ComputeCapability: strings.TrimSpace(row[5]),
|
|
PCIBusID: strings.TrimSpace(row[6]),
|
|
DriverVersion: strings.TrimSpace(row[7]),
|
|
NUMANode: -1,
|
|
}
|
|
g.MemoryTotalMB, _ = parseCSVInt(row[3])
|
|
g.MemoryUsedMB, _ = parseCSVInt(row[4])
|
|
g.Health.TemperatureC, _ = parseCSVInt(row[8])
|
|
g.Health.ECCErrors, _ = parseCSVInt(row[9])
|
|
|
|
assessHealth(&g)
|
|
gpus = append(gpus, g)
|
|
}
|
|
return gpus, nil
|
|
}
|
|
|
|
// residentMemoryToleranceMB is how much allocated memory is written off as
|
|
// driver overhead rather than a foreign process.
|
|
const residentMemoryToleranceMB = 512
|
|
|
|
// assessHealth flags conditions that should keep a GPU out of placement.
|
|
func assessHealth(g *GPU) {
|
|
if g.Health.ECCErrors > 0 {
|
|
g.Health.Problems = append(g.Health.Problems,
|
|
fmt.Sprintf("%d uncorrected ECC errors", g.Health.ECCErrors))
|
|
}
|
|
// Memory already in use means something else is on the card. Starting an
|
|
// engine there would either fail to allocate or quietly evict the other
|
|
// process's throughput.
|
|
if g.MemoryUsedMB > residentMemoryToleranceMB {
|
|
g.Health.Problems = append(g.Health.Problems,
|
|
fmt.Sprintf("%d MiB already allocated by another process", g.MemoryUsedMB))
|
|
}
|
|
}
|