避免部署构建占满临时文件系统

deploy.sh

@@ -36,8 +36,15 @@ readonly REPO_DIR="/home/program/llmsproxy"
 readonly HEALTH_URL="http://127.0.0.1:8081/v1/models"
 readonly BIN_NAME="llmsproxy"
 
-# 构建目录在 /tmp（tmpfs 更快）
+# 构建目录。默认落在仓库下的构建工作区，而不是 /tmp：本机 /tmp 是 8.8G
-readonly TMP_DIR="/tmp/llmsproxy-deploy"
+# tmpfs，实测被 Go 构建缓存占满后会让 go test/build 在链接阶段直接失败。
+# 可用 LLMSPROXY_BUILD_DIR 覆盖；同一目录也被测试用作 GOTMPDIR。
+readonly BUILD_SUBDIR=".build-work"
+if [[ -n "${LLMSPROXY_BUILD_DIR:-}" ]]; then
+    readonly TMP_DIR="${LLMSPROXY_BUILD_DIR}"
+else
+    readonly TMP_DIR="${REPO_DIR}/${BUILD_SUBDIR}"
+fi
 # 同文件系统暂存文件（用于原子 rename）
 readonly STAGING_BIN="${TARGET_BIN}.staging"
 readonly STAGING_CONFIG="${TARGET_CONFIG}.staging"
@@ -142,7 +149,7 @@ build() {
     mkdir -p "$TMP_DIR/bin"
 
     cd "$REPO_DIR"
-    CGO_ENABLED=1 go build -tags luajit \
+    CGO_ENABLED=1 GOTMPDIR="$TMP_DIR" go build -tags luajit \
         -trimpath \
         -ldflags="-s -w" \
         -o "$TMP_DIR/bin/$BIN_NAME" \

internal/gateway/stats.go

@@ -228,6 +228,26 @@ func incStatus(a *Stat, name string, r Req) {
 	}
 }
 
+// addStat merges an already aggregated row into a total. It deliberately
+// includes failed requests without a model: routing can fail before a model is
+// resolved, but that request still belongs in the gateway total.
+func addStat(dst, src *Stat) {
+	if dst == nil || src == nil {
+		return
+	}
+	dst.Reqs += src.Reqs
+	dst.OK += src.OK
+	dst.Err += src.Err
+	dst.Tokens += src.Tokens
+	dst.Prompt += src.Prompt
+	dst.Compl += src.Compl
+	dst.LatSum += src.LatSum
+	dst.FirstByteSum += src.FirstByteSum
+	if src.LatMax > dst.LatMax {
+		dst.LatMax = src.LatMax
+	}
+}
+
 // LoadAudit primes the in-memory state from the audit file.
 //
 // The aggregates (totals, per-key/model/source rows, status counts and the
@@ -1219,17 +1239,13 @@ func (s *Stats) Snapshot(limit int, key string) map[string]interface{} {
 		byModel = s.byKeyModel[key]
 		bySrc = s.byKeySrc[key]
 	}
-	for _, a := range byModel {
+	// The total is the sum of every recorded request, not the model rows.
-		total.Reqs += a.Reqs
+	// Routing failures are valid requests but may have no resolved model; deriving
-		total.OK += a.OK
+	// total from byModel silently dropped those 502/503 records. Use byKey, which
-		total.Err += a.Err
+	// is populated for every request (including an empty key), and keep the same
-		total.Tokens += a.Tokens
+	// scope as the selected key filter.
-		total.Prompt += a.Prompt
+	for _, a := range byKey {
-		total.Compl += a.Compl
+		addStat(&total, a)
-		total.LatSum += a.LatSum
-		if a.LatMax > total.LatMax {
-			total.LatMax = a.LatMax
-		}
 	}
 	bs := make([]agrRow, 0, len(s.byStatus))
 	for code, st := range s.byStatus {

internal/gateway/stats_test.go

@@ -560,3 +560,17 @@ func TestAuditRotationThresholds(t *testing.T) {
 		t.Fatalf("total audit budget = %d, want 256 MiB", total)
 	}
 }
+
+func TestSnapshotTotalIncludesRequestsWithoutModel(t *testing.T) {
+	s := NewStats(100)
+	s.Record(Req{Key: "k", Status: 200, OK: true})
+	s.Record(Req{Key: "k", Model: "m", Status: 200, OK: true, Prompt: 3, Compl: 2})
+	snap := s.Snapshot(0, "")
+	total, ok := snap["total"].(Stat)
+	if !ok {
+		t.Fatalf("total missing: %#v", snap["total"])
+	}
+	if total.Reqs != 2 || total.OK != 2 || total.Tokens != 5 {
+		t.Fatalf("total must include model-less requests: %#v", total)
+	}
+}

internal/lua/plugins/channels.lua

@@ -7110,27 +7110,24 @@ function plugin.usage_token_ledger(req)
   if key ~= "" then
     url = url .. "&key=" .. (key:len() <= 6 and key or ("***" .. key:sub(-6)))
   end
-  local resp, err = __llmsproxy_http({ url = url, method = "GET", timeout_ms = 4000 })
+  local headers = {}
+  if key ~= "" then
+    headers["Authorization"] = "Bearer " .. key
+  end
+  local resp, err = __llmsproxy_http({ url = url, method = "GET", timeout_ms = 4000, headers = headers })
   if err or not resp or resp.status ~= 200 then
-    return { status = 200, json = { ok = true, snapshot = { available = false,
+    return { status = 503, json = { ok = false, error = "usage snapshot unavailable: " .. tostring(err or (resp and resp.status or "no response")) } }
-      error = err or (resp and tostring(resp.status) or "no response") } } }
   end
   local ok, rec = pcall(json.decode, resp.body or "")
   if not ok or type(rec) ~= "table" then
-    return { status = 200, json = { ok = true, snapshot = { available = false, error = "stats 无法解析" } } }
+    return { status = 502, json = { ok = false, error = "usage snapshot 无法解析" } }
-  end
-  -- 汇总 token 总量（兼容 stats 的不同形状）
-  local prompt, completion, requests = 0, 0, 0
-  local rows = rec.requests or rec.rows or rec.data or {}
-  if type(rows) == "table" then
-    for _, r in ipairs(rows) do
-      if type(r) == "table" then
-        prompt = prompt + (tonumber(r.prompt_tokens) or 0)
-        completion = completion + (tonumber(r.completion_tokens) or 0)
-        requests = requests + 1
-      end
-    end
   end
+  -- 汇总 token 总量。/api/stats 的聚合在 `total` 下，不在顶层数组中。
+  -- 之前错误地遍历 rec.requests/rows/data，导致该端点永远返回 0。
+  local total = type(rec.total) == "table" and rec.total or {}
+  local prompt = tonumber(total.prompt_tokens) or 0
+  local completion = tonumber(total.completion_tokens) or 0
+  local requests = tonumber(total.reqs) or 0
   return { status = 200, json = { ok = true, snapshot = { available = true,
     tokens = { prompt = prompt, completion = completion, total = prompt + completion },
     requests = requests, raw = rec } } }