Alerting: Support Unified Alerting with Grafana HA (#37920)

* Alerting: Support Unified Alerting in Grafana's HA mode.
This commit is contained in:
gotjosh
2021-09-16 15:33:51 +01:00
committed by GitHub
parent 92209f1011
commit 7db97097c9
25 changed files with 6377 additions and 79 deletions

View File

@@ -18,15 +18,14 @@ import (
"strings"
"time"
"github.com/gobwas/glob"
"github.com/prometheus/common/model"
"gopkg.in/ini.v1"
"github.com/grafana/grafana-aws-sdk/pkg/awsds"
"github.com/grafana/grafana/pkg/components/gtime"
"github.com/grafana/grafana/pkg/infra/log"
"github.com/grafana/grafana/pkg/util"
"github.com/gobwas/glob"
"github.com/prometheus/common/model"
"gopkg.in/ini.v1"
)
type Scheme string
@@ -420,7 +419,14 @@ type Cfg struct {
GeomapEnableCustomBaseLayers bool
// Unified Alerting
AdminConfigPollInterval time.Duration
AdminConfigPollInterval time.Duration
AlertmanagerConfigPollInterval time.Duration
HAListenAddr string
HAAdvertiseAddr string
HAPeers []string
HAPeerTimeout time.Duration
HAGossipInterval time.Duration
HAPushPullInterval time.Duration
}
// IsLiveConfigEnabled returns true if live should be able to save configs to SQL tables
@@ -916,8 +922,7 @@ func (cfg *Cfg) Load(args CommandLineArgs) error {
if err := readAlertingSettings(iniFile); err != nil {
return err
}
if err := cfg.readUnifiedAlertingSettings(iniFile); err != nil {
if err := cfg.ReadUnifiedAlertingSettings(iniFile); err != nil {
return err
}
@@ -1374,13 +1379,6 @@ func (cfg *Cfg) readRenderingSettings(iniFile *ini.File) error {
return nil
}
func (cfg *Cfg) readUnifiedAlertingSettings(iniFile *ini.File) error {
ua := iniFile.Section("unified_alerting")
s := ua.Key("admin_config_poll_interval_seconds").MustInt(60)
cfg.AdminConfigPollInterval = time.Second * time.Duration(s)
return nil
}
func readAlertingSettings(iniFile *ini.File) error {
alerting := iniFile.Section("alerting")
AlertingEnabled = alerting.Key("enabled").MustBool(true)

View File

@@ -0,0 +1,57 @@
package setting
import (
"strings"
"time"
"github.com/grafana/grafana/pkg/components/gtime"
"github.com/prometheus/alertmanager/cluster"
"gopkg.in/ini.v1"
)
const (
AlertmanagerDefaultClusterAddr = "0.0.0.0:9094"
AlertmanagerDefaultPeerTimeout = 15 * time.Second
AlertmanagerDefaultGossipInterval = cluster.DefaultGossipInterval
AlertmanagerDefaultPushPullInterval = cluster.DefaultPushPullInterval
SchedulerDefaultAdminConfigPollInterval = 60 * time.Second
AlertmanagerDefaultConfigPollInterval = 60 * time.Second
)
func (cfg *Cfg) ReadUnifiedAlertingSettings(iniFile *ini.File) error {
ua := iniFile.Section("unified_alerting")
var err error
cfg.AdminConfigPollInterval, err = gtime.ParseDuration(valueAsString(ua, "admin_config_poll_interval", (SchedulerDefaultAdminConfigPollInterval).String()))
if err != nil {
return err
}
cfg.AlertmanagerConfigPollInterval, err = gtime.ParseDuration(valueAsString(ua, "alertmanager_config_poll_interval", (AlertmanagerDefaultConfigPollInterval).String()))
if err != nil {
return err
}
cfg.HAPeerTimeout, err = gtime.ParseDuration(valueAsString(ua, "ha_peer_timeout", (AlertmanagerDefaultPeerTimeout).String()))
if err != nil {
return err
}
cfg.HAGossipInterval, err = gtime.ParseDuration(valueAsString(ua, "ha_gossip_interval", (AlertmanagerDefaultGossipInterval).String()))
if err != nil {
return err
}
cfg.HAPushPullInterval, err = gtime.ParseDuration(valueAsString(ua, "ha_push_pull_interval", (AlertmanagerDefaultPushPullInterval).String()))
if err != nil {
return err
}
cfg.HAListenAddr = ua.Key("ha_listen_address").MustString(AlertmanagerDefaultClusterAddr)
cfg.HAAdvertiseAddr = ua.Key("ha_advertise_address").MustString("")
peers := ua.Key("ha_peers").MustString("")
cfg.HAPeers = make([]string, 0)
if peers != "" {
for _, peer := range strings.Split(peers, ",") {
peer = strings.TrimSpace(peer)
cfg.HAPeers = append(cfg.HAPeers, peer)
}
}
return nil
}

View File

@@ -0,0 +1,39 @@
package setting
import (
"testing"
"time"
"github.com/stretchr/testify/require"
)
func TestCfg_ReadUnifiedAlertingSettings(t *testing.T) {
cfg := NewCfg()
err := cfg.Load(CommandLineArgs{HomePath: "../../", Config: "../../conf/defaults.ini"})
require.NoError(t, err)
// It sets the correct defaults.
{
require.Equal(t, 60*time.Second, cfg.AdminConfigPollInterval)
require.Equal(t, 60*time.Second, cfg.AlertmanagerConfigPollInterval)
require.Equal(t, 15*time.Second, cfg.HAPeerTimeout)
require.Equal(t, "0.0.0.0:9094", cfg.HAListenAddr)
require.Equal(t, "", cfg.HAAdvertiseAddr)
require.Len(t, cfg.HAPeers, 0)
require.Equal(t, 200*time.Millisecond, cfg.HAGossipInterval)
require.Equal(t, 60*time.Second, cfg.HAPushPullInterval)
}
// With peers set, it correctly parses them.
{
require.Len(t, cfg.HAPeers, 0)
s, err := cfg.Raw.NewSection("unified_alerting")
require.NoError(t, err)
_, err = s.NewKey("ha_peers", "hostname1:9090,hostname2:9090,hostname3:9090")
require.NoError(t, err)
require.NoError(t, cfg.ReadUnifiedAlertingSettings(cfg.Raw))
require.Len(t, cfg.HAPeers, 3)
require.ElementsMatch(t, []string{"hostname1:9090", "hostname2:9090", "hostname3:9090"}, cfg.HAPeers)
}
}