Files
bee/audit/internal/app/component_status_db.go
T
Mikhail ChusavitinandClaude Sonnet 5 cfaa15ec7c platform/webui: fold confidential-computing into a GPU config + NVLink check
The standalone "confidential-computing" SAT target only ever checked CC
readiness, which most fleets never opt into (a NOT_READY verdict there
isn't a fault). Meanwhile DCGM diag never asserts GPU config compliance
(ECC/MIG/power-limit vs factory default) or NVLink topology (per NVIDIA's
own DGX BasePOD deployment guide, this needs a separate validation step)
— gaps confirmed against public DCGM docs and a real NV17-vs-expected-NV18
bonded pair found on a live bundle.

Repurposes the routine into "nvidia-config": reuses the existing
ListNvidiaGPUSettings() (already backing the GPU-settings page) to flag
ECC disabled, a MIG mode change stuck pending a reset/reboot, and a power
limit capped >5% below default; parses "nvidia-smi topo -m" bonded pairs
against "nvlink -s/-e" to flag any inactive lane or nonzero replay/
recovery/CRC counter on an otherwise-active bond. CC readiness is folded
in as one informational field (does not gate overall_status) rather than
a dedicated test. Reports under the same pcie:gpu:nvidia severity key as
every other nvidia-* SAT target instead of an isolated key, so a
config/NVLink FAILED result isn't invisible next to stress-test results.

Also fixes ApplySATResultToDB silently dropping any target with no
matching switch case (exactly what the old confidential-computing target
did) with a new coverage test enumerating every real SAT target.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-09 11:36:52 +03:00

330 lines
11 KiB
Go

package app
import (
"encoding/json"
"os"
"path/filepath"
"strings"
"sync"
"time"
)
// ComponentStatusDB is a persistent, append-only store of hardware component health records.
// Records are keyed by component identity strings (e.g. "pcie:0000:c8:00.0", "storage:nvme0n1").
// Once a component is marked Warning or Critical, subsequent OK entries do not downgrade it —
// the component stays at the highest observed severity until explicitly reset.
type ComponentStatusDB struct {
path string
mu sync.Mutex
records map[string]*ComponentStatusRecord
}
// ComponentStatusRecord holds the current and historical health of one hardware component.
type ComponentStatusRecord struct {
ComponentKey string `json:"component_key"`
Status string `json:"status"` // "OK", "Warning", "Critical", "Unknown"
LastCheckedAt time.Time `json:"last_checked_at"`
LastChangedAt time.Time `json:"last_changed_at"`
ErrorSummary string `json:"error_summary,omitempty"`
History []ComponentStatusEntry `json:"history"`
}
// ComponentStatusEntry is one observation written to a component's history.
type ComponentStatusEntry struct {
At time.Time `json:"at"`
Status string `json:"status"`
Source string `json:"source"` // e.g. "sat:nvidia", "sat:memory", "watchdog:kmsg"
Detail string `json:"detail,omitempty"`
}
// OpenComponentStatusDB opens (or creates) the JSON status DB at path.
func OpenComponentStatusDB(path string) (*ComponentStatusDB, error) {
db := &ComponentStatusDB{
path: path,
records: make(map[string]*ComponentStatusRecord),
}
if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
return nil, err
}
data, err := readFileLimited(path, 10<<20)
if err != nil && !os.IsNotExist(err) {
return nil, err
}
if len(data) > 0 {
var records []ComponentStatusRecord
if err := json.Unmarshal(data, &records); err == nil {
for i := range records {
db.records[records[i].ComponentKey] = &records[i]
}
}
}
return db, nil
}
// Record writes one observation for the given component key.
// source is a short label like "sat:nvidia" or "watchdog:kmsg".
// status is "OK", "Warning", "Critical", or "Unknown".
// OK never downgrades an existing Warning or Critical status.
func (db *ComponentStatusDB) Record(key, source, status, detail string) {
if db == nil || strings.TrimSpace(key) == "" {
return
}
db.mu.Lock()
defer db.mu.Unlock()
db.reloadLocked()
now := time.Now().UTC()
rec, exists := db.records[key]
if !exists {
rec = &ComponentStatusRecord{ComponentKey: key}
db.records[key] = rec
}
rec.LastCheckedAt = now
// History records status *transitions*, per the ingest contract
// (bible-local/docs/hardware-ingest-contract.md: "История переходов
// статусов" — status_history is a transition log, not a per-poll
// journal). Skip the append when this source's last recorded status for
// this key is unchanged, or a continuously-polled component (e.g. the
// PSU watchdog, every 60s indefinitely) grows an unbounded run of
// identical "still OK" entries and the file never stops growing.
// LastCheckedAt above already carries "we still saw this as of now" for
// a steady-state component, so nothing is lost by not repeating it here.
if last := lastEntryFromSource(rec.History, source); last == nil || last.Status != status {
rec.History = append(rec.History, ComponentStatusEntry{At: now, Status: status, Source: source, Detail: detail})
}
// Status merge: OK never downgrades Warning/Critical.
newSev := componentSeverity(status)
curSev := componentSeverity(rec.Status)
if newSev > curSev {
rec.Status = status
rec.LastChangedAt = now
rec.ErrorSummary = detail
} else if rec.Status == "" {
rec.Status = status
rec.LastChangedAt = now
}
_ = db.saveLocked()
}
// lastEntryFromSource returns the most recent history entry recorded by the
// given source, or nil if that source has never reported for this key.
func lastEntryFromSource(history []ComponentStatusEntry, source string) *ComponentStatusEntry {
for i := len(history) - 1; i >= 0; i-- {
if history[i].Source == source {
return &history[i]
}
}
return nil
}
// Get returns the current record for a component key.
func (db *ComponentStatusDB) Get(key string) (ComponentStatusRecord, bool) {
if db == nil {
return ComponentStatusRecord{}, false
}
db.mu.Lock()
defer db.mu.Unlock()
r, ok := db.records[key]
if !ok {
return ComponentStatusRecord{}, false
}
return *r, true
}
// All returns a snapshot of all records.
func (db *ComponentStatusDB) All() []ComponentStatusRecord {
if db == nil {
return nil
}
db.mu.Lock()
defer db.mu.Unlock()
out := make([]ComponentStatusRecord, 0, len(db.records))
for _, r := range db.records {
out = append(out, *r)
}
return out
}
// reloadLocked merges on-disk state into memory before this process applies
// its own update. component-status.json is shared by the long-lived bee-web
// process (writing PSU/kmsg watchdog records roughly every 60s) and each
// short-lived "bee bee-worker" subprocess spawned per SAT task (writing GPU/
// CPU/memory/storage records once on completion) — each holds its own
// in-memory copy backed by the same file. Without this reload, saveLocked
// below would dump the caller's stale in-memory map over the file and erase
// whatever the other process wrote in between: e.g. a GPU SAT task's
// pcie:gpu:nvidia record, written by a worker subprocess, silently vanishing
// the next time the main process's health poller ticks and saves its own
// GPU-blind snapshot. Only keys with a newer LastCheckedAt on disk are
// pulled in, so this process's own pending (not-yet-saved) update for key is
// never discarded by its own reload.
func (db *ComponentStatusDB) reloadLocked() {
data, err := readFileLimited(db.path, 10<<20)
if err != nil || len(data) == 0 {
return
}
var onDisk []ComponentStatusRecord
if err := json.Unmarshal(data, &onDisk); err != nil {
return
}
for i := range onDisk {
key := onDisk[i].ComponentKey
if existing, ok := db.records[key]; ok && !onDisk[i].LastCheckedAt.After(existing.LastCheckedAt) {
continue
}
rec := onDisk[i]
db.records[key] = &rec
}
}
func (db *ComponentStatusDB) saveLocked() error {
records := make([]ComponentStatusRecord, 0, len(db.records))
for _, r := range db.records {
records = append(records, *r)
}
data, err := json.MarshalIndent(records, "", " ")
if err != nil {
return err
}
return os.WriteFile(db.path, data, 0644)
}
// componentSeverity returns a numeric severity so higher values win.
func componentSeverity(status string) int {
switch strings.TrimSpace(status) {
case "Critical":
return 3
case "Warning":
return 2
case "OK":
return 1
default:
return 0
}
}
// ApplySATResultToDB reads a SAT summary.txt from the run directory next to archivePath
// and writes component status records to db for the given SAT target.
// archivePath may be either a bare .tar.gz path or "Archive written to /path/foo.tar.gz".
func ApplySATResultToDB(db *ComponentStatusDB, target, archivePath string) {
if db == nil || strings.TrimSpace(archivePath) == "" {
return
}
archivePath = extractArchivePath(archivePath)
if archivePath == "" {
return
}
runDir := strings.TrimSuffix(archivePath, ".tar.gz")
data, err := os.ReadFile(filepath.Join(runDir, "summary.txt"))
if err != nil {
return
}
kv := parseSATKV(string(data))
overall := strings.ToUpper(strings.TrimSpace(kv["overall_status"]))
if overall == "" {
return
}
source := "sat:" + target
dbStatus := satStatusToDBStatus(overall)
// Map SAT target to component keys. GPU targets are keyed by vendor, not
// by the raw target string: "nvidia" (Check tier) and "nvidia-stress" /
// "nvidia-targeted-stress" (Load/Burn tier) all exercise the same
// physical GPUs, so they must share one severity-tracked record — a
// severity-1 Check run after a severity-3 Load failure must not lose
// that failure just because it ran more recently. Recording each target
// under its own key (the previous behavior) also silently broke
// applyComponentStatusDB below, which expects "pcie:gpu:<vendor>" and
// otherwise fails to match any real BDF.
switch target {
case "nvidia", "nvidia-targeted-stress", "nvidia-compute", "nvidia-targeted-power", "nvidia-pulse",
"nvidia-interconnect", "nvidia-bandwidth", "nvidia-stress", "nvidia-config":
db.Record("pcie:gpu:nvidia", source, dbStatus, target+" SAT: "+overall)
case "amd", "amd-stress", "amd-mem", "amd-bandwidth":
db.Record("pcie:gpu:amd", source, dbStatus, target+" SAT: "+overall)
case "memory", "memory-stress", "sat-stress":
db.Record("memory:all", source, dbStatus, target+" SAT: "+overall)
case "cpu", "platform-stress":
db.Record("cpu:all", source, dbStatus, target+" SAT: "+overall)
case "storage":
// Try to record per-device if available in summary.
recordedAny := false
for key, val := range kv {
if !strings.HasSuffix(key, "_status") || key == "overall_status" {
continue
}
base := strings.TrimSuffix(key, "_status")
idx := strings.Index(base, "_")
if idx <= 0 {
continue
}
devName := base[:idx]
devStatus := satStatusToDBStatus(strings.ToUpper(strings.TrimSpace(val)))
db.Record("storage:"+devName, source, devStatus, "storage SAT: "+val)
recordedAny = true
}
if !recordedAny {
db.Record("storage:all", source, dbStatus, "storage SAT: "+overall)
}
}
}
func satStatusToDBStatus(overall string) string {
switch overall {
case "OK":
return "OK"
case "FAILED":
return "Warning"
case "PARTIAL", "UNSUPPORTED":
return "Unknown"
default:
return "Unknown"
}
}
// ExtractArchivePath extracts a bare path from a string that may be
// "Archive written to /path/to/run-dir" or already a bare path.
func ExtractArchivePath(s string) string {
return extractArchivePath(s)
}
// ReadSATOverallStatus reads the overall_status value from the summary.txt
// file located in the run directory alongside archivePath.
// Returns "" if the file cannot be read.
func ReadSATOverallStatus(archivePath string) string {
if strings.TrimSpace(archivePath) == "" {
return ""
}
runDir := strings.TrimSuffix(archivePath, ".tar.gz")
data, err := os.ReadFile(filepath.Join(runDir, "summary.txt"))
if err != nil {
return ""
}
kv := parseSATKV(string(data))
return strings.ToUpper(strings.TrimSpace(kv["overall_status"]))
}
func extractArchivePath(s string) string {
s = strings.TrimSpace(s)
if rest, ok := strings.CutPrefix(s, "Archive written to "); ok {
return strings.TrimSpace(rest)
}
return s
}
func parseSATKV(raw string) map[string]string {
kv := make(map[string]string)
for _, line := range strings.Split(raw, "\n") {
k, v, ok := strings.Cut(strings.TrimSpace(line), "=")
if ok {
kv[strings.TrimSpace(k)] = strings.TrimSpace(v)
}
}
return kv
}