metrics: introduce config alerts gauge

Since there is no full support of Tarantool 3 config instances in
luatest yet (only treegen support in master), I had borrowed some test
helpers from tarantool/crud [1].

1. https://github.com/tarantool/crud/blob/98b120ef7095fa34525ef9d335a1458a2edf0cca/test/tarantool3_helpers

Part of tarantool/grafana-dashboard#224

Author: DifferentialOrange

.luacheckrc

@@ -2,3 +2,4 @@ include_files = {"**/*.lua", "*.rockspec", "*.luacheckrc"}
 exclude_files = {"lua_modules/", ".luarocks/", ".rocks/", "tmp/", ".history/"}
 
 max_line_length = 120
+max_comment_line_length = 200

CHANGELOG.md

@@ -5,6 +5,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
+### Added
+- New Tarantool 3 metrics:
+  - tnt_config_alerts
 
 ## [1.1.0] - 2024-05-17
 ### Added

doc/monitoring/api_reference.rst

@@ -566,6 +566,7 @@ Metrics functions
     *   ``cartridge_failover``
     *   ``clock``
     *   ``event_loop``
+    *   ``config``
 
     See :ref:`metrics reference <metrics-reference>` for details.
     All metric collectors from the collection have ``metainfo.default = true``.

doc/monitoring/metrics_reference.rst

@@ -993,3 +993,21 @@ Read view statistics
 
         *   -   ``tnt_memtx_index_read_view``
             -   Memory (in bytes) held for read views.
+
+
+Tarantool configuration
+-----------------------
+
+These metrics are available starting from Tarantool 3.0.
+
+..  container:: table
+
+    ..  list-table::
+        :widths: 25 75
+        :header-rows: 0
+
+        *   -   ``tnt_config_alerts``
+            -   Count of current instance :ref:`configuration apply alerts <config_api_reference_info>`.
+                ``{level="warn"}`` label covers warnings and
+                ``{level="error"}`` covers errors.
+                

metrics/tarantool.lua

@@ -23,6 +23,7 @@ local default_metrics = {
     cartridge_failover  = require('metrics.cartridge.failover'),
     clock               = require('metrics.tarantool.clock'),
     event_loop          = require('metrics.tarantool.event_loop'),
+    config              = require('metrics.tarantool.config'),
 }
 
 local all_metrics_map = {}

metrics/tarantool/config.lua

@@ -0,0 +1,46 @@
+local utils = require('metrics.utils')
+
+local collectors_list = {}
+
+local function get_config_alerts(config_info)
+    -- https://github.com/tarantool/tarantool/blob/319357d5973d15d08b8eda6a230eada08b710802/src/box/lua/config/utils/aboard.lua#L17-L18
+    local config_alerts = {
+        warn = 0,
+        error = 0,
+    }
+
+    for _, alert in pairs(config_info.alerts) do
+        config_alerts[alert.type] = config_alerts[alert.type] + 1
+    end
+
+    return config_alerts
+end
+
+local function update()
+    if not utils.is_tarantool3() then
+        return
+    end
+
+    -- Can migrate to box.info().config later
+    -- https://github.com/tarantool/tarantool/commit/a1544d3bbc029c6fb2a148e580afe2b20e269b8d
+    local config = require('config')
+    local config_info = config:info()
+
+    local config_alerts = get_config_alerts(config_info)
+
+    for level, count in pairs(config_alerts) do
+        collectors_list.config_alerts = utils.set_gauge(
+            'config_alerts',
+            'Tarantool 3 configuration alerts',
+            count,
+            {level = level},
+            nil,
+            {default = true}
+        )
+    end
+end
+
+return {
+    update = update,
+    list = collectors_list,
+}

metrics/utils.lua

@@ -36,4 +36,20 @@ function utils.delete_collectors(list)
     table.clear(list)
 end
 
+local function get_tarantool_version()
+    local version_parts = rawget(_G, '_TARANTOOL'):split('-', 3)
+
+    local major_minor_patch_parts = version_parts[1]:split('.', 2)
+    local major = tonumber(major_minor_patch_parts[1])
+    local minor = tonumber(major_minor_patch_parts[2])
+    local patch = tonumber(major_minor_patch_parts[3])
+
+    return major, minor, patch
+end
+
+function utils.is_tarantool3()
+    local major = get_tarantool_version()
+    return major == 3
+end
+
 return utils

test/tarantool/config_metrics_test.lua

@@ -0,0 +1,179 @@
+local t = require('luatest')
+local g = t.group()
+
+local fio = require('fio')
+local yaml = require('yaml')
+
+local utils = require('test.utils')
+local treegen = require('test.tarantool3_helpers.treegen')
+local server_helper = require('test.tarantool3_helpers.server')
+
+g.before_all(function(cg)
+    cg.treegen = {}
+    treegen.init(cg.treegen)
+end)
+
+g.after_all(function(cg)
+    treegen.clean(cg.treegen)
+end)
+
+
+local default_config =  {
+    credentials = {
+        users = {
+            guest = {
+                roles = {'super'},
+            },
+            replicator = {
+                password = 'replicating',
+                roles = {'replication'},
+            },
+        },
+    },
+    iproto = {
+        advertise = {
+            peer = {
+                login = 'replicator',
+            },
+        },
+    },
+    groups = {
+        servers = {
+            replicasets = {
+                ['server-001'] = {
+                    leader = 'server-001-a',
+                    instances = {
+                        ['server-001-a'] = {
+                            iproto = {
+                                listen = {{uri = 'localhost:3301'}},
+                            },
+                        },
+                    },
+                },
+            },
+        },
+    },
+    replication = {
+        failover = 'manual',
+    },
+    metrics = {
+        include = {'all'},
+    },
+}
+
+local function write_config(cg, config)
+    return treegen.write_script(cg.server_dir, 'config.yaml', yaml.encode(config))
+end
+
+local function start_server(cg)
+    t.skip_if(not utils.is_tarantool_3_config_supported(),
+              'Skip since Tarantool 3 config is unsupported')
+
+    cg.server_dir = treegen.prepare_directory(cg.treegen, {}, {})
+    local config_file = write_config(cg, default_config)
+
+    cg.server = server_helper:new{
+        alias = 'server-001-a',
+        config_file = config_file,
+        chdir = cg.server_dir,
+    }
+    cg.server:start{wait_until_ready = true}
+end
+
+local function stop_server(cg)
+    if cg.server ~= nil then
+        cg.server:stop()
+        cg.server = nil
+    end
+
+    if cg.server_dir ~= nil then
+        fio.rmtree(cg.server_dir)
+        cg.server_dir = nil
+    end
+end
+
+local function reload_config(cg, config)
+    write_config(cg, config)
+    cg.server:exec(function()
+        pcall(function()
+            require('config'):reload()
+        end)
+    end)
+end
+
+local function assert_config_alerts_metrics(server, expected_values)
+    local observations = server:exec(function()
+        local metrics = require('metrics')
+        metrics.invoke_callbacks()
+        return metrics.collect()
+    end)
+
+    local warnings = utils.find_obs(
+        'tnt_config_alerts',
+        {level = 'warn', alias = 'server-001-a'},
+        observations
+    )
+    t.assert_equals(warnings.value, expected_values['warn'])
+
+    local errors = utils.find_obs(
+        'tnt_config_alerts',
+        {level = 'error', alias = 'server-001-a'},
+        observations
+    )
+    t.assert_equals(errors.value, expected_values['error'])
+end
+
+
+g.before_test('test_config_alerts_if_healthy', start_server)
+g.after_test('test_config_alerts_if_healthy', stop_server)
+
+g.test_config_alerts_if_healthy = function(cg)
+    assert_config_alerts_metrics(cg.server, {warn = 0, error = 0})
+end
+
+
+g.before_test('test_config_alerts_if_minor_trouble', start_server)
+g.after_test('test_config_alerts_if_minor_trouble', stop_server)
+
+g.test_config_alerts_if_minor_trouble = function(cg)
+    local config = table.deepcopy(default_config)
+    config['credentials']['users']['user_one'] = {roles = {'role_two'}}
+    reload_config(cg, config)
+
+    assert_config_alerts_metrics(cg.server, {warn = 1, error = 0})
+end
+
+
+g.before_test('test_config_alerts_if_critical_failure', start_server)
+g.after_test('test_config_alerts_if_critical_failure', stop_server)
+
+g.test_config_alerts_if_critical_failure = function(cg)
+    local config = table.deepcopy(default_config)
+    config['groups']['servers'] = {}
+    reload_config(cg, config)
+
+    assert_config_alerts_metrics(cg.server, {warn = 0, error = 1})
+end
+
+
+g.before_test('test_config_alerts_if_unsupported', function(cg)
+    t.skip_if(utils.is_tarantool_3_config_supported(),
+              'Skip since Tarantool 3 config is supported')
+    utils.create_server(cg)
+end)
+
+g.after_test('test_config_alerts_if_unsupported', function(cg)
+    utils.drop_server(cg)
+    cg.server = nil
+end)
+
+g.test_config_alerts_if_unsupported = function(cg)
+    local observations = cg.server:exec(function()
+        local metrics = require('metrics')
+        metrics.invoke_callbacks()
+        return metrics.collect()
+    end)
+
+    local alerts = utils.find_metric('tnt_config_alerts', observations)
+    t.assert_equals(alerts, nil)
+end