修复二开 agent 流量增量丢失: 基线改上报确认后推进并落盘持久化

- xray 新增快照/确认接口 GetTrafficSnapshot/CommitTrafficSnapshot/
  RestoreTrafficBaseline, 与面板 GetTraffic 的 StatsLastValues 完全隔离
- agent 改走快照: 仅主控 HTTP 200 后才推进基线, 失败轮增量由累计
  计数做差在下轮自动补齐; 已确认基线原子落盘, agent 重启恢复续差
- 核心重启计数归零的从零规则保留; Onlines 仍每轮新鲜采集直报
- 新增 11 个测试覆盖断网恢复/连续失败补齐/重启恢复/计数归零/
  面板语义隔离, xray 与 agent 两包全绿
This commit is contained in:
Muse
2026-10-20 15:58:38 +00:00
parent 4828f946b1
commit 0f0335e943
5 changed files with 818 additions and 22 deletions
+121
View File
@@ -61,6 +61,19 @@ type XrayAPI struct {
grpcClient *grpc.ClientConn
isConnected bool
StatsLastValues map[string]int64
// committedStats is the agent reporting path's last ACKNOWLEDGED
// baseline: the raw core counters as of the most recent snapshot the
// master confirmed receiving (HTTP 200). It is deliberately separate
// from StatsLastValues — GetTraffic (the panel's local accounting)
// advances its baseline at read time, while the agent path may only
// advance on CommitTrafficSnapshot, so a failed upload never loses
// the deltas it carried. committedReady is false until a baseline has
// been established (first snapshot committed, or one restored from
// disk); while false, snapshots report no deltas, mirroring
// GetTraffic's first-poll rule.
committedStats map[string]int64
committedReady bool
}
// Init connects to the Xray API server.
@@ -650,6 +663,114 @@ func (x *XrayAPI) GetTraffic() ([]*Traffic, []*ClientTraffic, error) {
return mapToSlice(tagTrafficMap), mapToSlice(emailTrafficMap), nil
}
// TrafficSnapshot is one read of the core counters taken for the agent
// reporting path: the deltas accrued since the last COMMITTED baseline,
// plus Values, the raw cumulative counters the deltas were computed from.
//
// Unlike GetTraffic — which advances its baseline as it reads, the right
// contract for the panel's local accounting job — a snapshot changes no
// state. The caller reports the deltas and only then calls
// CommitTrafficSnapshot; if the report fails, the next snapshot recomputes
// against the same committed baseline and the unreported traffic is
// included again, because the core counters are cumulative and are never
// reset by polling.
type TrafficSnapshot struct {
Traffics []*Traffic
ClientTraffics []*ClientTraffic
// Values holds every counter returned by the core at snapshot time,
// keyed by stat name. CommitTrafficSnapshot installs exactly this map
// as the new committed baseline.
Values map[string]int64
}
// BaselineValues returns a copy of the baseline CommitTrafficSnapshot
// would install for this snapshot, so callers can persist it (the agent
// writes it to disk before/while committing).
func (s *TrafficSnapshot) BaselineValues() map[string]int64 {
out := make(map[string]int64, len(s.Values))
for k, v := range s.Values {
out[k] = v
}
return out
}
// GetTrafficSnapshot reads the core counters and computes deltas against
// the committed baseline WITHOUT advancing it. Per-stat rules match
// GetTraffic exactly: while no baseline has been established the snapshot
// reports no deltas (it only proposes one), a counter the baseline has
// never seen counts from zero, and a counter below its baseline (the core
// restarted) counts from zero as well.
func (x *XrayAPI) GetTrafficSnapshot() (*TrafficSnapshot, error) {
if x.grpcClient == nil {
return nil, common.NewError("xray api is not initialized")
}
ctx, cancel := context.WithTimeout(context.Background(), time.Second*10)
defer cancel()
stats, err := x.queryStats(ctx)
if err != nil {
logger.Debug("Failed to query Xray stats:", err)
return nil, err
}
snap := &TrafficSnapshot{Values: make(map[string]int64, len(stats))}
tagTrafficMap := make(map[string]*Traffic)
emailTrafficMap := make(map[string]*ClientTraffic)
for _, stat := range stats {
snap.Values[stat.Name] = stat.Value
if !x.committedReady {
continue
}
lastValue, ok := x.committedStats[stat.Name]
if !ok || stat.Value < lastValue {
lastValue = 0
}
value := stat.Value - lastValue
if matches := trafficRegex.FindStringSubmatch(stat.Name); len(matches) == 4 {
processTraffic(matches, value, tagTrafficMap)
} else if matches := clientTrafficRegex.FindStringSubmatch(stat.Name); len(matches) == 3 {
processClientTraffic(matches, value, emailTrafficMap)
}
}
snap.Traffics = mapToSlice(tagTrafficMap)
snap.ClientTraffics = mapToSlice(emailTrafficMap)
return snap, nil
}
// CommitTrafficSnapshot marks a snapshot as acknowledged by the master:
// its Values become the committed baseline that later snapshots diff
// against. Installing only the counters the core currently reports also
// retires baselines for deleted inbounds/clients, the same end state
// GetTraffic's pruning converges to.
func (x *XrayAPI) CommitTrafficSnapshot(s *TrafficSnapshot) {
if s == nil {
return
}
x.committedStats = s.BaselineValues()
x.committedReady = true
}
// RestoreTrafficBaseline installs a committed baseline recovered from
// persistent storage (the agent saves one after every acknowledged
// report), so a restarted agent keeps diffing against the last values the
// master confirmed instead of swallowing the unreported tail into a
// fresh baseline. An empty map leaves the baseline unestablished: the
// next snapshot is the baseline-only first pass, exactly like a first
// run with no state file.
func (x *XrayAPI) RestoreTrafficBaseline(values map[string]int64) {
if len(values) == 0 {
return
}
x.committedStats = make(map[string]int64, len(values))
for k, v := range values {
x.committedStats[k] = v
}
x.committedReady = true
}
// OnlineIP is one source address of a live connection, with the unix time (seconds)
// the core last dispatched a link from it.
type OnlineIP struct {