feat(obs): 访问日志 SSOT 与 edge_health,去掉协议兼容层

Agent 仅上报 host_metrics/edge_health/access_logs;业务流量与 UV 由
Server 侧访问日志聚合。新增 of_node_edge_health 与 of_access_log_hourly,
删除 request_reports/openresty 吞吐路径;API 不再暴露 traffic_reports
与 openresty_rx|tx。心跳/离线默认阈值与回填迁移一并入库。
This commit is contained in:
ryan
2026-07-18 11:53:11 +08:00
parent 9a0974cce8
commit f0e234df1f
60 changed files with 1799 additions and 2164 deletions
+32 -127
View File
@@ -2,147 +2,52 @@ package nginx
import "github.com/Rain-kl/Wavelet/internal/apps/agent/protocol"
const (
openRestyObservabilityWindowTTL = "7200"
openRestyObservabilityWindowSize = "60"
)
// Local OpenResty observability endpoint (target model):
// GET /openflare/observability returns instantaneous health/connections only.
// Business traffic is collected exclusively from access.log.
const openRestyObservabilityInitLua = `local dict = ngx.shared.openflare_observability
if not dict then
return
end
return
const openRestyObservabilityInitLua = `return
`
const openRestyObservabilityLogLua = `local dict = ngx.shared.openflare_observability
if not dict then
return
end
local request_uri = tostring(ngx.var.uri or "")
if request_uri == "/openflare/observability" or request_uri == "/openflare/stub_status" then
return
end
local ttl = ` + openRestyObservabilityWindowTTL + `
local now = ngx.time()
local window_size = ` + openRestyObservabilityWindowSize + `
local window_start = now - (now % window_size)
local function ensure_counter(key)
dict:add(key, 0, ttl)
end
local function incr(key, delta)
ensure_counter(key)
local value, err = dict:incr(key, delta)
if not value and err == "not found" then
dict:set(key, delta, ttl)
end
end
local function remember_value(list_key, marker_key, value)
if value == "" then
return
end
if not dict:add(marker_key, 1, ttl) then
return
end
local existing = dict:get(list_key)
if not existing or existing == "" then
dict:set(list_key, value, ttl)
return
end
dict:set(list_key, existing .. "\n" .. value, ttl)
end
local window_prefix = tostring(window_start)
incr("request_count:" .. window_prefix, 1)
local status = tostring(ngx.status or 0)
if status ~= "0" then
incr("status:" .. window_prefix .. ":" .. status, 1)
remember_value(
"status_keys:" .. window_prefix,
"status_marker:" .. window_prefix .. ":" .. status,
status
)
if tonumber(status) and tonumber(status) >= 500 then
incr("error_count:" .. window_prefix, 1)
end
end
local host = tostring(ngx.var.host or "")
if host ~= "" then
incr("domain:" .. window_prefix .. ":" .. host, 1)
remember_value(
"domain_keys:" .. window_prefix,
"domain_marker:" .. window_prefix .. ":" .. host,
host
)
end
local remote_addr = tostring(ngx.var.binary_remote_addr or ngx.var.remote_addr or "")
if remote_addr ~= "" and dict:add("visitor:" .. window_prefix .. ":" .. remote_addr, 1, ttl) then
incr("unique_visitor_count:" .. window_prefix, 1)
end
local request_length = tonumber(ngx.var.request_length) or 0
if request_length > 0 then
incr("openresty_rx_bytes:" .. window_prefix, request_length)
end
local bytes_sent = tonumber(ngx.var.bytes_sent) or tonumber(ngx.var.body_bytes_sent) or 0
if bytes_sent > 0 then
incr("openresty_tx_bytes:" .. window_prefix, bytes_sent)
end
// log.lua no longer accumulates business counters (access.log is the authority).
const openRestyObservabilityLogLua = `return
`
// read.lua exposes stub_status-style connection gauges as JSON.
const openRestyObservabilityReadLua = `local cjson = require "cjson.safe"
local dict = ngx.shared.openflare_observability
if not dict then
ngx.status = ngx.HTTP_SERVICE_UNAVAILABLE
ngx.say(cjson.encode({ message = "shared dict unavailable" }))
return
end
local now = ngx.time()
local window_size = ` + openRestyObservabilityWindowSize + `
local window_start = now - (now % window_size)
local current_window = tostring(window_start)
local function read_counter(key)
return tonumber(dict:get(key) or 0) or 0
end
local function read_map(window_id, prefix, list_key)
local result = {}
local raw = dict:get(list_key .. ":" .. window_id)
if not raw or raw == "" then
return result
local function read_stub_status()
local res = ngx.location.capture("/openflare/stub_status")
if not res or res.status ~= 200 or not res.body then
return nil
end
for value in string.gmatch(raw, "[^\n]+") do
result[value] = read_counter(prefix .. ":" .. window_id .. ":" .. value)
end
return result
local body = res.body
local active = tonumber(string.match(body, "Active connections:%s*(%d+)")) or 0
local reading = tonumber(string.match(body, "Reading:%s*(%d+)")) or 0
local writing = tonumber(string.match(body, "Writing:%s*(%d+)")) or 0
local waiting = tonumber(string.match(body, "Waiting:%s*(%d+)")) or 0
return {
active = active,
reading = reading,
writing = writing,
waiting = waiting
}
end
local connections = read_stub_status()
local payload = {
window_started_at_unix = window_start,
window_ended_at_unix = now,
request_count = read_counter("request_count:" .. current_window),
error_count = read_counter("error_count:" .. current_window),
unique_visitor_count = read_counter("unique_visitor_count:" .. current_window),
status_codes = read_map(current_window, "status", "status_keys"),
top_domains = read_map(current_window, "domain", "domain_keys"),
source_countries = {},
openresty_rx_bytes = read_counter("openresty_rx_bytes:" .. current_window),
openresty_tx_bytes = read_counter("openresty_tx_bytes:" .. current_window)
ok = connections ~= nil,
captured_at_unix = ngx.time(),
connections = connections or {
active = 0,
reading = 0,
writing = 0,
waiting = 0
}
}
ngx.header.content_type = "application/json"
ngx.status = ngx.HTTP_OK
ngx.say(cjson.encode(payload))
`
@@ -0,0 +1,35 @@
package nginx
import (
"strings"
"testing"
)
func TestManagedObservabilityLuaIsHealthOnly(t *testing.T) {
t.Parallel()
files := ManagedObservabilityLuaFiles()
var logLua, readLua string
for _, file := range files {
switch file.Path {
case "log.lua":
logLua = file.Content
case "read.lua":
readLua = file.Content
}
}
if logLua == "" || readLua == "" {
t.Fatal("expected log.lua and read.lua")
}
// Business counters must not be written in log phase.
if strings.Contains(logLua, "openresty_rx_bytes") ||
strings.Contains(logLua, "request_count") {
t.Fatal("log.lua must not accumulate business counters")
}
if !strings.Contains(readLua, "connections") || !strings.Contains(readLua, "ok") {
t.Fatal("read.lua must expose ok + connections health snapshot")
}
if strings.Contains(readLua, "top_domains") || strings.Contains(readLua, "request_count") {
t.Fatal("read.lua must not expose business traffic aggregates")
}
}