feat(xid): simplify xid component

[cardyok]

No description provided.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 33.62319% with 229 lines in your changes missing coverage. Please review.

Project coverage is 21.15%. Comparing base (02d1814) to head (eddea4c).

Files with missing lines	Patch %	Lines
...mponents/accelerator/nvidia/error/xid/component.go	8.33%	132 Missing ⚠️
components/accelerator/nvidia/error/xid/helper.go	78.57%	19 Missing and 2 partials ⚠️
internal/session/serve.go	0.00%	20 Missing ⚠️
components/db/events.go	81.25%	4 Missing and 2 partials ⚠️
internal/server/server.go	0.00%	6 Missing ⚠️
components/metrics/metrics.go	0.00%	3 Missing ⚠️
...omponents/accelerator/nvidia/bad-envs/component.go	0.00%	1 Missing ⚠️
...onents/accelerator/nvidia/clock-speed/component.go	0.00%	1 Missing ⚠️
components/accelerator/nvidia/ecc/component.go	0.00%	1 Missing ⚠️
...nts/accelerator/nvidia/error-xid-sxid/component.go	0.00%	1 Missing ⚠️
... and 37 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #321      +/-   ##
==========================================
- Coverage   21.21%   21.15%   -0.06%     
==========================================
  Files         300      301       +1     
  Lines       27089    27302     +213     
==========================================
+ Hits         5747     5776      +29     
- Misses      20697    20895     +198     
+ Partials      645      631      -14     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gyuho]

Also need to use this events DB in:

gpud/components/accelerator/nvidia/query/nvml/xid.go

Lines 179 to 190 in 02d1814

	// no need to check duplicate entries, assuming nvml event poller does not return old events
	ctx, cancel := context.WithTimeout(inst.rootCtx, 10*time.Second)
	werr := nvidia_xid_sxid_state.InsertEvent(ctx, inst.dbRW, nvidia_xid_sxid_state.Event{
	UnixSeconds: event.Time.Unix(),
	DataSource: "nvml",
	EventType: "xid",
	EventID: int64(event.Xid),
	DeviceID: event.DeviceUUID,
	EventDetails: string(eb),
	})
	cancel()
	if werr != nil {

(this PR only covers dmesg ones, we also need the ones from NVML)

---

TODO: need to deprecate old xid matcher:

gpud/internal/server/server.go

Lines 332 to 390 in 02d1814

	dmesgProcessMatched := func(ts time.Time, line []byte, matchedFilter *query_log_common.Filter) {
	if ts.IsZero() {
	return
	}
	if len(line) == 0 {
	return
	}
	if matchedFilter == nil {
	return
	}
	cctx, ccancel := context.WithTimeout(ctx, 10*time.Second)
	defer ccancel()
	for _, ref := range matchedFilter.OwnerReferences {
	switch ref {
	case nvidia_component_error_xid_id.Name:
	ev, err := nvidia_query_xid.ParseDmesgLogLine(metav1.Time{Time: ts}, string(line))
	if err != nil {
	log.Logger.Errorw("failed to parse xid dmesg line", "line", string(line), "error", err)
	continue
	}
	if ev.Detail == nil {
	log.Logger.Errorw("failed to parse xid dmesg line", "line", string(line), "error", "no detail")
	continue
	}
	eventToInsert := nvidia_xid_sxid_state.Event{
	UnixSeconds: ts.Unix(),
	DataSource: "dmesg",
	EventType: "xid",
	EventID: int64(ev.Detail.Xid),
	DeviceID: ev.DeviceUUID,
	EventDetails: ev.LogItem.Line,
	}
	if xidSxidEventDeduper.Get(eventToInsert) {
	log.Logger.Debugw("xid event already exists in cache -- deduped", "event", eventToInsert)
	continue
	}
	if err := xidSxidEventDeduper.Add(eventToInsert); err != nil {
	log.Logger.Errorw("failed to add xid event to deduper", "error", err)
	continue
	}
	// not found in cache, fallback to db lookup
	found, err := nvidia_xid_sxid_state.FindEvent(cctx, dbRO, eventToInsert)
	if err != nil {
	log.Logger.Errorw("failed to find xid event in database", "error", err)
	continue
	}
	if found {
	log.Logger.Debugw("xid event already exists in database", "event", eventToInsert)
	continue
	}
	if werr := nvidia_xid_sxid_state.InsertEvent(cctx, dbRW, eventToInsert); werr != nil {
	log.Logger.Errorw("failed to insert xid event into database", "error", werr)
	continue
	}

[gyuho]

Discussed offline, we will do NVML in a separate PR

[gyuho]

nice catch

[xiang90]

let us improve the test coverage to make the CI green.