fix(vitals): control plane only create table rotater timer once (#9083)

* fix(vitals): control plane only create table rotater timer once

After enabling vitals, whenever each data plane connects to the control plane,
it will trigger the control plane to generate a table rotater timer,
and the table rotater timer will cyclically create itself, causing timer leaks.

This fix moves the creation of the table rotater timer to the start function,
where it checks the self:get_running() status to determine if the table rotater
timer has already started, ensuring that the table rotater timer is only
created once.

Fix [FTI-5905](https://konghq.atlassian.net/browse/FTI-5905)

Signed-off-by: tzssangglass <tzssangglass@gmail.com>

changelog/unreleased/kong-ee/create-table-rotater-timer-once.yml

@@ -0,0 +1,4 @@
+message: |
+  **Vitals**: Fixed a bug that each data plane connecting to the control plane would trigger the control plane to create a redundant table rotater timer.
+type: bugfix
+scope: Core

kong/vitals/postgres/strategy.lua

@@ -219,7 +219,7 @@ function _M.new(db, opts)
   local table_rotater = table_rotater_module.new(
     {
       connector = db.connector,
-      rotation_interval = opts.ttl_seconds or 3600,
+      ttl_seconds = opts.ttl_seconds or 3600,
       list_cache = ngx.shared.kong_vitals_lists,
     }
   )
@@ -266,12 +266,6 @@ function _M:init(node_id, hostname)
     return nil, "node_id is required"
   end
 
-  local ok, err = self.table_rotater:init()
-
-  if not ok then
-    return nil, "failed to init table rotator: " .. err
-  end
-
   self.node_id = node_id
   self.hostname = hostname
 
@@ -285,13 +279,19 @@ end
 
 
 function _M:start()
-  -- delete timer
-  -- make sure the timer is started only once, clustering strategy will call
+  -- table rotater timer and delete timer
+  -- make sure these two timers are started only once, clustering strategy will call
   -- this function multiple times
   if self:get_running() then
     return true
   end
 
+  local ok, err = self.table_rotater:init()
+
+  if not ok then
+    return nil, "failed to init table rotator: " .. err
+  end
+
   local when = self.delete_interval
   log(INFO, _log_prefix, "starting initial postgres delete timer in ", when, " seconds")
 

kong/vitals/postgres/table_rotater.lua

@@ -12,6 +12,7 @@ local log        = ngx.log
 local WARN       = ngx.WARN
 local DEBUG      = ngx.DEBUG
 local ERR        = ngx.ERR
+local NOTICE     = ngx.NOTICE
 local time       = ngx.time
 
 
@@ -93,6 +94,7 @@ end
 
 
 function _M:init()
+  log(NOTICE, _log_prefix, "init vitals table rotater")
   -- make sure we have a current vitals_stats_seconds table
   local query = fmt(CREATE_VITALS_STATS_SECONDS, self:current_table_name())
   local _, err = self.connector:query(query)

spec-ee/02-integration/03-vitals/13-table_rotater_spec.lua

@@ -0,0 +1,102 @@
+-- This software is copyright Kong Inc. and its licensors.
+-- Use of the software is subject to the agreement between your organization
+-- and Kong Inc. If there is no such agreement, use is governed by and
+-- subject to the terms of the Kong Master Software License Agreement found
+-- at https://konghq.com/enterprisesoftwarelicense/.
+-- [ END OF LICENSE 0867164ffc95e54f04670b5169c09574bdbd9bba ]
+
+local helpers = require "spec.helpers"
+local clear_license_env = require("spec-ee.helpers").clear_license_env
+local get_portal_and_vitals_key = require("spec-ee.helpers").get_portal_and_vitals_key
+local ngx_time = ngx.time
+local tonumber = tonumber
+local max = math.max
+local unpack = unpack
+
+
+local function select_vitals_seconds_tables(db)
+  local query = [[
+  select table_name from information_schema.tables
+   where table_name like 'vitals_stats_seconds_%'
+  ]]
+
+  local res = assert(db:query(query))
+  return res
+end
+
+
+for _, strategy in helpers.each_strategy() do
+  describe("vitals table rotater #" .. strategy, function()
+    local reset_license_data, db
+
+    lazy_setup(function()
+      reset_license_data = clear_license_env()
+      local _
+      _, db = helpers.get_db_utils(strategy, {
+        "clustering_data_planes",
+        "licenses",
+      })
+      db = db.connector
+
+      assert(helpers.start_kong({
+        role = "control_plane",
+        cluster_cert = "spec/fixtures/kong_clustering.crt",
+        cluster_cert_key = "spec/fixtures/kong_clustering.key",
+        lua_ssl_trusted_certificate = "spec/fixtures/kong_clustering.crt",
+        database = strategy,
+        db_update_frequency = 0.1,
+        cluster_listen = "127.0.0.1:9005",
+        nginx_conf = "spec/fixtures/custom_nginx.template",
+        log_level = "notice",
+        license_path = "spec-ee/fixtures/mock_license.json",
+        portal_and_vitals_key = get_portal_and_vitals_key(),
+        vitals = true,
+        vitals_ttl_seconds = 2,
+      }))
+    end)
+
+    lazy_teardown(function()
+      helpers.stop_kong("servroot")
+      helpers.stop_kong("servroot2")
+      reset_license_data()
+    end)
+
+    it("only init in control plane init worker phase", function()
+      assert.logfile("servroot/logs/error.log").has.line("[vitals-table-rotater] init vitals table rotater, context: init_worker_by_lua", true, 1)
+
+      helpers.clean_logfile("servroot/logs/error.log")
+
+      assert(helpers.start_kong({
+        role = "data_plane",
+        database = "off",
+        prefix = "servroot2",
+        cluster_cert = "spec/fixtures/kong_clustering.crt",
+        cluster_cert_key = "spec/fixtures/kong_clustering.key",
+        lua_ssl_trusted_certificate = "spec/fixtures/kong_clustering.crt",
+        cluster_control_plane = "127.0.0.1:9005",
+        proxy_listen = "0.0.0.0:19002",
+        nginx_conf = "spec/fixtures/custom_nginx.template",
+        log_level = "info",
+        vitals_flush_interval = 1,
+        license_path = "spec-ee/fixtures/mock_license.json",
+        portal_and_vitals_key = get_portal_and_vitals_key(),
+        vitals = true,
+      }))
+
+      -- check if vitals table rotater init trigger by dataplane connect
+      assert.logfile("servroot/logs/error.log").has.no.line("init vitals table rotater, context: ngx.timer", true, 5)
+
+      local res = select_vitals_seconds_tables(db)
+      local seconds_times = {}
+      local seconds_table_prefix = "vitals_stats_seconds_"
+      for i = 1, #res do
+        seconds_times[i] = tonumber(res[i].table_name:sub(#seconds_table_prefix + 1))
+      end
+
+      local nearly_table_time = max(unpack(seconds_times))
+      local now = ngx_time()
+      -- verify that table rotater timer work(table rotater timer should create new table every 2s during ngx.sleep(5))
+      assert.is_true(now - nearly_table_time < 3)
+    end)
+  end)
+end