NBS-4748: migrate from solomon to monitoring

ydb-platform · Jan 19, 2024 · e0e4fce · e0e4fce
1 parent fc1dd1f
commit e0e4fce
Show file tree

Hide file tree

Showing 19 changed files with 271 additions and 99 deletions.
diff --git a/cloud/blockstore/config/diagnostics.proto b/cloud/blockstore/config/diagnostics.proto
@@ -56,6 +56,25 @@ message TVolumePerfSettings
     optional uint32 CriticalFactor = 3;
 };
 
+////////////////////////////////////////////////////////////////////////////////
+// Monitoring URL components
+
+message TMonitoringUrlData
+{
+    // Monitoring cluster name (e.g. yandexcloud_prod_vla).
+    optional string MonitoringClusterName = 1;
+    // Monitoring host url.
+    optional string MonitoringUrl = 2;
+    // Project name in monitoring.
+    optional string MonitoringProject = 3;
+
+    // Dashboards ids for monitoring URL formation.
+    optional string MonitoringVolumeDashboard = 4;
+    optional string MonitoringPartitionDashboard = 5;
+    optional string MonitoringNBSAlertsDashboard = 6;
+    optional string MonitoringNBSTVDashboard = 7;
+};
+
 ////////////////////////////////////////////////////////////////////////////////
 
 message TDiagnosticsConfig
@@ -72,8 +91,9 @@ message TDiagnosticsConfig
     // Kikimr monitoring port.
     optional string KikimrMonitoringPort = 4;
 
+    // [obsolete]
     // Name of cluster in Solomon.
-    optional string SolomonClusterName = 5;
+    // optional string SolomonClusterName = 5;
 
     // Kikimr monitoring port.
     optional uint32 KikimrMonPort = 6;
@@ -94,17 +114,19 @@ message TDiagnosticsConfig
     // HDD disk performance threshold
     // optional TVolumePerfThreshold HddPerfThreshold = 15;
 
+    // [obsolete]
     // Solomon host url.
-    optional string SolomonUrl = 16;
+    // optional string SolomonUrl = 16;
 
     // Allow destructive LWTrace actions.
     optional bool UnsafeLWTrace = 17;
 
     // Path to LWTrace query file.
     optional string LWTraceDebugInitializationQuery = 18;
 
+    // [obsolete]
     // Project name in solomon.
-    optional string SolomonProject = 20;
+    // optional string SolomonProject = 20;
 
     // Sampling rate for request tracking
     optional uint32 SamplingRate = 21;
@@ -178,4 +200,7 @@ message TDiagnosticsConfig
 
     // Performance measurements coefficients for HDD NRD.
     optional TVolumePerfSettings HddNonreplPerfSettings = 46;
+
+    // Monitoring data necessary for link generation on monpages.
+    optional TMonitoringUrlData MonitoringUrlData = 47;
 }
diff --git a/cloud/blockstore/libs/diagnostics/config.cpp b/cloud/blockstore/libs/diagnostics/config.cpp
@@ -15,7 +15,6 @@ namespace {
     xxx(HostNameScheme,                  NProto::EHostNameScheme, NProto::EHostNameScheme::HOSTNAME_RAW )\
     xxx(BastionNameSuffix,               TString,         ""                                            )\
     xxx(ViewerHostName,                  TString,         ""                                            )\
-    xxx(SolomonClusterName,              TString,         ""                                            )\
     xxx(KikimrMonPort,                   ui32,            8765                                          )\
     xxx(NbsMonPort,                      ui32,            8766                                          )\
                                                                                                          \
@@ -26,8 +25,6 @@ namespace {
                                                                                                          \
     xxx(ProfileLogTimeThreshold,         TDuration,       TDuration::Seconds(15)                        )\
     xxx(UseAsyncLogger,                  bool,            false                                         )\
-    xxx(SolomonUrl,                      TString,         ""                                            )\
-    xxx(SolomonProject,                  TString,         "nbs"                                         )\
     xxx(UnsafeLWTrace,                   bool,            false                                         )\
     xxx(LWTraceDebugInitializationQuery, TString,         ""                                            )\
     xxx(SsdPerfSettings,                TVolumePerfSettings,  {}                                        )\
@@ -40,6 +37,7 @@ namespace {
     xxx(ExpectedIoParallelism,          ui32,                 32                                        )\
     xxx(CloudIdsWithStrictSLA,          TVector<TString>,     {}                                        )\
     xxx(LWTraceShuttleCount,            ui32,                 2000                                      )\
+    xxx(MonitoringUrlData,              TMonitoringUrlData,   {}                                        )\
                                                                                                          \
     xxx(CpuWaitFilename,            TString, "/sys/fs/cgroup/cpu/system.slice/nbs.service/cpuacct.wait" )\
                                                                                                          \
@@ -85,6 +83,14 @@ ConvertValue<TVolumePerfSettings, NProto::TVolumePerfSettings>(
     return TVolumePerfSettings(value);
 }
 
+template <>
+TMonitoringUrlData
+ConvertValue<TMonitoringUrlData, NProto::TMonitoringUrlData>(
+    const NProto::TMonitoringUrlData& value)
+{
+    return TMonitoringUrlData(value);
+}
+
 template <>
 TRequestThresholds
 ConvertValue<TRequestThresholds, TProtoRequestThresholds>(
@@ -115,6 +121,11 @@ bool IsEmpty(const NProto::TVolumePerfSettings& t)
     return t.ByteSizeLong() == 0;
 }
 
+bool IsEmpty(const NProto::TMonitoringUrlData& t)
+{
+    return t.ByteSizeLong() == 0;
+}
+
 template <typename T>
 bool IsEmpty(const google::protobuf::RepeatedPtrField<T>& value)
 {
@@ -253,6 +264,22 @@ void Out<NCloud::NBlockStore::TVolumePerfSettings>(
     SerializeToTextFormat(v, out);
 }
 
+template <>
+void Out<NCloud::NBlockStore::TMonitoringUrlData>(
+    IOutputStream& out,
+    const NCloud::NBlockStore::TMonitoringUrlData& value)
+{
+    NCloud::NBlockStore::NProto::TMonitoringUrlData v;
+    v.SetMonitoringClusterName(value.MonitoringClusterName);
+    v.SetMonitoringUrl(value.MonitoringUrl);
+    v.SetMonitoringProject(value.MonitoringProject);
+    v.SetMonitoringVolumeDashboard(value.MonitoringVolumeDashboard);
+    v.SetMonitoringPartitionDashboard(value.MonitoringPartitionDashboard);
+    v.SetMonitoringNBSAlertsDashboard(value.MonitoringNBSAlertsDashboard);
+    v.SetMonitoringNBSTVDashboard(value.MonitoringNBSTVDashboard);
+    SerializeToTextFormat(v, out);
+}
+
 template <>
 void Out<NCloud::TRequestThresholds>(
     IOutputStream& out,

diff --git a/cloud/blockstore/libs/diagnostics/config.h b/cloud/blockstore/libs/diagnostics/config.h
@@ -70,6 +70,34 @@ struct TVolumePerfSettings:
 
 ////////////////////////////////////////////////////////////////////////////////
 
+struct TMonitoringUrlData: public TAtomicRefCount<TMonitoringUrlData>
+{
+    TString MonitoringClusterName;
+    TString MonitoringUrl;
+    TString MonitoringProject;
+    TString MonitoringVolumeDashboard;
+    TString MonitoringPartitionDashboard;
+    TString MonitoringNBSAlertsDashboard;
+    TString MonitoringNBSTVDashboard;
+
+    TMonitoringUrlData()
+        : MonitoringProject("nbs")
+    {}
+    TMonitoringUrlData(const TMonitoringUrlData& rhs) = default;
+
+    explicit TMonitoringUrlData(const NProto::TMonitoringUrlData& data)
+        : MonitoringClusterName(data.GetMonitoringClusterName())
+        , MonitoringUrl(data.GetMonitoringUrl())
+        , MonitoringProject(data.GetMonitoringProject())
+        , MonitoringVolumeDashboard(data.GetMonitoringVolumeDashboard())
+        , MonitoringPartitionDashboard(data.GetMonitoringPartitionDashboard())
+        , MonitoringNBSAlertsDashboard(data.GetMonitoringNBSAlertsDashboard())
+        , MonitoringNBSTVDashboard(data.GetMonitoringNBSTVDashboard())
+    {}
+};
+
+////////////////////////////////////////////////////////////////////////////////
+
 class TDiagnosticsConfig
 {
 private:
@@ -81,9 +109,6 @@ class TDiagnosticsConfig
     NProto::EHostNameScheme GetHostNameScheme() const;
     TString GetBastionNameSuffix() const;
     TString GetViewerHostName() const;
-    TString GetSolomonClusterName() const;
-    TString GetSolomonUrl() const;
-    TString GetSolomonProject() const;
     ui32 GetKikimrMonPort() const;
     ui32 GetNbsMonPort() const;
     ui32 GetSamplingRate() const;
@@ -106,6 +131,7 @@ class TDiagnosticsConfig
     TVolumePerfSettings GetLocalSSDPerfSettings() const;
     ui32 GetExpectedIoParallelism() const;
     TVector<TString> GetCloudIdsWithStrictSLA() const;
+    TMonitoringUrlData GetMonitoringUrlData() const;
 
     TString GetCpuWaitFilename() const;
 

diff --git a/cloud/blockstore/libs/diagnostics/hostname.cpp b/cloud/blockstore/libs/diagnostics/hostname.cpp
@@ -72,74 +72,65 @@ TString GetExternalHostUrl(
     return out;
 }
 
-TString GetSolomonVolumeUrl(
+TString GetMonitoringVolumeUrl(
     const TDiagnosticsConfig& config,
-    const TString& diskId,
-    const TString& dashboard)
+    const TString& diskId)
 {
+    TMonitoringUrlData data = config.GetMonitoringUrlData();
     return TStringBuilder()
-        << config.GetSolomonUrl()
-        << "/?project=" << config.GetSolomonProject()
-        << "&service=service_volume"
-        << "&cluster="<< config.GetSolomonClusterName()
-        << "&volume=" << diskId
-        << "&dashboard=" << dashboard;
+           << data.MonitoringUrl << "/projects/" << data.MonitoringProject
+           << "/dashboards/" << data.MonitoringVolumeDashboard
+           << "?from=now-1d&to=now&refresh=60000&p.cluster="
+           << data.MonitoringClusterName << "&p.volume=" << diskId;
 }
 
-TString GetSolomonPartitionUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard)
+TString GetMonitoringPartitionUrl(const TDiagnosticsConfig& config)
 {
+    TMonitoringUrlData data = config.GetMonitoringUrlData();
     return TStringBuilder()
-        << config.GetSolomonUrl()
-        << "/?project=" << config.GetSolomonProject()
-        << "&service=tablets"
-        << "&cluster=" << config.GetSolomonClusterName()
-        << "&host=" << GetShortHostName()
-        << "&dashboard=" << dashboard;
+           << data.MonitoringUrl << "/projects/" << data.MonitoringProject
+           << "/dashboards/" << data.MonitoringPartitionDashboard
+           << "?from=now-1d&to=now&"
+              "refresh=60000&p.service=tablets&p.cluster="
+           << data.MonitoringClusterName << "&p.host=" << GetShortHostName();
 }
 
-TString GetSolomonServerUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard)
+TString GetMonitoringNBSAlertsUrl(const TDiagnosticsConfig& config)
 {
+    TMonitoringUrlData data = config.GetMonitoringUrlData();
     return TStringBuilder()
-        << config.GetSolomonUrl()
-        << "/?project" << config.GetSolomonProject()
-        << "&service=server"
-        << "&cluster=" << config.GetSolomonClusterName()
-        << "&host=" << GetShortHostName()
-        << "&type=-"
-        << "&dashboard="<< dashboard;
+           << data.MonitoringUrl << "/projects/" << data.MonitoringProject
+           << "/dashboards/" << data.MonitoringNBSAlertsDashboard
+           << "?from=now-1d&to=now&refresh=60000&p.cluster="
+           << data.MonitoringClusterName << "&p.host=" << GetShortHostName();
 }
 
-TString GetSolomonClientUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard)
+TString GetMonitoringNBSOverviewToTVUrl(const TDiagnosticsConfig& config)
 {
+    TMonitoringUrlData data = config.GetMonitoringUrlData();
     return TStringBuilder()
-        << config.GetSolomonUrl()
-        << "/?project=" << config.GetSolomonProject()
-        << "&service=client"
-        << "&cluster="<< config.GetSolomonClusterName()
-        << "&host=" << GetShortHostName()
-        << "&type=-"
-        << "&dashboard=" << dashboard;
+           << data.MonitoringUrl << "/projects/" << data.MonitoringProject
+           << "/dashboards/" << data.MonitoringNBSTVDashboard
+           << "?from=now-1d&to=now&refresh=60000&p.cluster="
+           << data.MonitoringClusterName << "&p.host=cluster";
 }
 
 TString GetSolomonBsProxyUrl(
     const TDiagnosticsConfig& config,
     ui32 groupId,
-    const TString& dashboard)
+    const TString& storagePool)
 {
+    TMonitoringUrlData data = config.GetMonitoringUrlData();
     return TStringBuilder()
-        << config.GetSolomonUrl()
-        << "/?project" << config.GetSolomonProject()
-        << "&service=dsproxy_percentile"
-        << "&cluster=" << config.GetSolomonClusterName()
-        << "&host=" << GetShortHostName()
-        << "&blobstorageproxy=" << groupId
-        << "&dashboard=" << dashboard;
+           << data.MonitoringUrl
+           << "/projects/kikimr/explorer/"
+              "queries?q.0.s=histogram_percentile(99, {project=\"kikimr"
+           << "\", cluster=\"" << data.MonitoringClusterName
+           << "\", storagePool=\"" << storagePool << "\", group=\"" << groupId
+           << "\", host=\"*\", service=\"vdisks\", "
+              "subsystem=\"latency_histo\", "
+              "handleclass=\"GetFast\"})&q.0.name=A&from=now-1d&to=now&refresh="
+              "60000";
 }
 
 }   // namespace NCloud::NBlockStore
diff --git a/cloud/blockstore/libs/diagnostics/hostname.h b/cloud/blockstore/libs/diagnostics/hostname.h
@@ -24,26 +24,19 @@ TString GetExternalHostUrl(
     EHostService serviceType,
     const TDiagnosticsConfig& config);
 
-TString GetSolomonServerUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard);
+TString GetMonitoringNBSAlertsUrl(const TDiagnosticsConfig& config);
 
-TString GetSolomonClientUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard);
+TString GetMonitoringNBSOverviewToTVUrl(const TDiagnosticsConfig& config);
 
-TString GetSolomonVolumeUrl(
+TString GetMonitoringVolumeUrl(
     const TDiagnosticsConfig& config,
-    const TString& diskId,
-    const TString& dashboard);
+    const TString& diskId);
 
-TString GetSolomonPartitionUrl(
-    const TDiagnosticsConfig& config,
-    const TString& dashboard);
+TString GetMonitoringPartitionUrl(const TDiagnosticsConfig& config);
 
 TString GetSolomonBsProxyUrl(
     const TDiagnosticsConfig& config,
     ui32 groupId,
-    const TString& dashboard);
+    const TString& storagePool);
 
 }   // namespace NCloud::NBlockStore
diff --git a/cloud/blockstore/libs/diagnostics/server_stats.cpp b/cloud/blockstore/libs/diagnostics/server_stats.cpp
@@ -601,15 +601,15 @@ void TServerStats::OutputHtml(IOutputStream& out, const IMonHttpRequest& request
         if (DiagnosticsConfig) {
             TAG(TH3) {
                 out << "<a href='"
-                    << GetSolomonServerUrl(*DiagnosticsConfig, "nbs-server-monitoring")
-                    << "'>Server dashboards</a>";
+                    << GetMonitoringNBSAlertsUrl(*DiagnosticsConfig)
+                    << "'>NBS Alerts dashboard</a>";
             };
 
             TAG(TH3) {
                 out << "<a href='"
-                    << GetSolomonClientUrl(*DiagnosticsConfig, "nbs-compute-client-monitoring")
-                    << "'>Client dashboards</a>";
-            };
+                    << GetMonitoringNBSOverviewToTVUrl(*DiagnosticsConfig)
+                    << "'>NBS overview To TV</a>";
+            }
         }
 
         TAG(TH3) { out << "Config"; }

diff --git a/cloud/blockstore/libs/storage/core/monitoring_utils.cpp b/cloud/blockstore/libs/storage/core/monitoring_utils.cpp
@@ -968,15 +968,14 @@ void DumpCompactionMap(
     }
 }
 
-void DumpSolomonVolumeLink(
+void DumpMonitoringVolumeLink(
     IOutputStream& out,
     const TDiagnosticsConfig& config,
     const TString& diskId)
 {
     HTML(out) {
         TAG(TH3) {
-            out << "<a href='"
-                << GetSolomonVolumeUrl(config, diskId, "nbs-volume-overview")
+            out << "<a href='" << GetMonitoringVolumeUrl(config, diskId)
                 << "'>Volume dashboards</a>";
         }
     }
@@ -988,8 +987,7 @@ void DumpSolomonPartitionLink(
 {
     HTML(out) {
         TAG(TH3) {
-            out << "<a href='"
-                << GetSolomonPartitionUrl(config, "nbs-tablets-transactions")
+            out << "<a href='" << GetMonitoringPartitionUrl(config)
                 << "'>Partition dashboards</a>";
         }
     }

diff --git a/cloud/blockstore/libs/storage/core/monitoring_utils.h b/cloud/blockstore/libs/storage/core/monitoring_utils.h
@@ -159,7 +159,7 @@ void DumpCompactionMap(
     const TVector<TCompactionCounter>& items,
     const ui32 rangeSize);
 
-void DumpSolomonVolumeLink(
+void DumpMonitoringVolumeLink(
     IOutputStream& out,
     const TDiagnosticsConfig& config,
     const TString& diskId);