Add monitoring counters for GroupLayoutSanitizer, KIKIMR-18170

Add monitoring counters for GroupLayoutSanitizer, KIKIMR-18170

4 files changed, 20 insertions, 3 deletions

diff --git a/ydb/core/blobstorage/base/blobstorage_events.h b/ydb/core/blobstorage/base/blobstorage_events.h
index 905d343163..78dd1e8571 100644
--- a/ydb/core/blobstorage/base/blobstorage_events.h
+++ b/ydb/core/blobstorage/base/blobstorage_events.h
@@ -258,6 +258,7 @@ namespace NKikimr {
         NKikimrBlobStorage::TEvControllerConfigRequest, TEvBlobStorage::EvControllerConfigRequest>
     {
         bool SelfHeal = false;
+        bool GroupLayoutSanitizer = false;
 
         TEvControllerConfigRequest() = default;
     };
diff --git a/ydb/core/mind/bscontroller/config_cmd.cpp b/ydb/core/mind/bscontroller/config_cmd.cpp
index 7c2437a610..ccffeb2edf 100644
--- a/ydb/core/mind/bscontroller/config_cmd.cpp
+++ b/ydb/core/mind/bscontroller/config_cmd.cpp
@@ -11,6 +11,7 @@ namespace NKikimr::NBsController {
             const ui64 Cookie;
             const NKikimrBlobStorage::TConfigRequest Cmd;
             const bool SelfHeal;
+            const bool GroupLayoutSanitizer;
             THolder<TEvBlobStorage::TEvControllerConfigResponse> Ev;
             NKikimrBlobStorage::TConfigResponse *Response;
             std::optional<TConfigState> State;
@@ -19,12 +20,13 @@ namespace NKikimr::NBsController {
 
         public:
             TTxConfigCmd(const NKikimrBlobStorage::TConfigRequest &cmd, const TActorId &notifyId, ui64 cookie,
-                    bool selfHeal, TBlobStorageController *controller)
+                    bool selfHeal, bool groupLayoutSanitizer, TBlobStorageController *controller)
                 : TTransactionBase(controller)
                 , NotifyId(notifyId)
                 , Cookie(cookie)
                 , Cmd(cmd)
                 , SelfHeal(selfHeal)
+                , GroupLayoutSanitizer(groupLayoutSanitizer)
                 , Ev(new TEvBlobStorage::TEvControllerConfigResponse())
                 , Response(Ev->Record.MutableResponse())
             {}
@@ -270,6 +272,12 @@ namespace NKikimr::NBsController {
                         : NBlobStorageController::COUNTER_SELFHEAL_REASSIGN_BSC_ERR;
                     Self->TabletCounters->Cumulative()[counter].Increment(1);
                 }
+                if (GroupLayoutSanitizer) {
+                    const auto counter = Success
+                        ? NBlobStorageController::COUNTER_GROUP_LAYOUT_SANITIZER_BSC_OK
+                        : NBlobStorageController::COUNTER_GROUP_LAYOUT_SANITIZER_BSC_ERR;
+                    Self->TabletCounters->Cumulative()[counter].Increment(1);
+                }
 
                 if (!Success) {
                     // rollback transaction
@@ -365,11 +373,14 @@ namespace NKikimr::NBsController {
             if (ev->Get()->SelfHeal) {
                 TabletCounters->Cumulative()[NBlobStorageController::COUNTER_SELFHEAL_REASSIGN_BSC_REQUESTS].Increment(1);
             }
+            if (ev->Get()->GroupLayoutSanitizer) {
+                TabletCounters->Cumulative()[NBlobStorageController::COUNTER_GROUP_LAYOUT_SANITIZER_BSC_REQUESTS].Increment(1);
+            }
 
             NKikimrBlobStorage::TEvControllerConfigRequest& record(ev->Get()->Record);
             const NKikimrBlobStorage::TConfigRequest& request = record.GetRequest();
             STLOG(PRI_DEBUG, BS_CONTROLLER, BSCTXCC01, "Execute TEvControllerConfigRequest", (Request, request));
-            Execute(new TTxConfigCmd(request, ev->Sender, ev->Cookie, ev->Get()->SelfHeal, this));
+            Execute(new TTxConfigCmd(request, ev->Sender, ev->Cookie, ev->Get()->SelfHeal, ev->Get()->GroupLayoutSanitizer, this));
         }
 
 } // NKikimr::NBsController
diff --git a/ydb/core/mind/bscontroller/self_heal.cpp b/ydb/core/mind/bscontroller/self_heal.cpp
index c3254838b1..dfebc047b5 100644
--- a/ydb/core/mind/bscontroller/self_heal.cpp
+++ b/ydb/core/mind/bscontroller/self_heal.cpp
@@ -147,13 +147,13 @@ namespace NKikimr::NBsController {
             }
 
             auto ev = MakeHolder<TEvBlobStorage::TEvControllerConfigRequest>();
-            ev->SelfHeal = true;
             auto& record = ev->Record;
             auto *request = record.MutableRequest();
             request->SetIgnoreGroupReserve(true);
             request->SetSettleOnlyOnOperationalDisks(true);
             request->SetIsSelfHealReasonDecommit(IsSelfHealReasonDecommit);
             if (VDiskToReplace) {
+                ev->SelfHeal = true;
                 auto *cmd = request->AddCommand()->MutableReassignGroupDisk();
                 cmd->SetGroupId(VDiskToReplace->GroupID);
                 cmd->SetGroupGeneration(VDiskToReplace->GroupGeneration);
@@ -161,6 +161,7 @@ namespace NKikimr::NBsController {
                 cmd->SetFailDomainIdx(VDiskToReplace->FailDomain);
                 cmd->SetVDiskIdx(VDiskToReplace->VDisk);
             } else {
+                ev->GroupLayoutSanitizer = true;
                 auto *cmd = request->AddCommand()->MutableSanitizeGroup();
                 cmd->SetGroupId(GroupId);
             }
diff --git a/ydb/core/protos/counters_bs_controller.proto b/ydb/core/protos/counters_bs_controller.proto
index 77e2dfb8c3..0a74ca8ab7 100644
--- a/ydb/core/protos/counters_bs_controller.proto
+++ b/ydb/core/protos/counters_bs_controller.proto
@@ -80,6 +80,10 @@ enum ECumulativeCounters {
     COUNTER_CONFIGCMD_ALLOCATE_VIRTUAL_GROUP_USEC = 42 [(CounterOpts) = {Name: "AllocateVirtualGroup"}];
     COUNTER_CONFIGCMD_DECOMMIT_GROUPS_USEC = 43 [(CounterOpts) = {Name: "DecommitGroups"}];
     COUNTER_CONFIGCMD_CANCEL_VIRTUAL_GROUP_USEC = 44 [(CounterOpts) = {Name: "CancelVirtualGroup"}];
+
+    COUNTER_GROUP_LAYOUT_SANITIZER_BSC_REQUESTS = 45 [(CounterOpts) = {Name: "GroupLayoutSanitizerBscRequests"}];
+    COUNTER_GROUP_LAYOUT_SANITIZER_BSC_OK = 46 [(CounterOpts) = {Name: "GroupLayoutSanitizerBscOk"}];
+    COUNTER_GROUP_LAYOUT_SANITIZER_BSC_ERR = 47 [(CounterOpts) = {Name: "GroupLayoutSanitizerBscErr"}];
 }
 
 enum EPercentileCounters {