Sentinel add ignore reason

add ignore reason

5 files changed, 50 insertions, 12 deletions

diff --git a/ydb/core/cms/sentinel.cpp b/ydb/core/cms/sentinel.cpp
index e7ccde90a9a..a818da004f4 100644
--- a/ydb/core/cms/sentinel.cpp
+++ b/ydb/core/cms/sentinel.cpp
@@ -213,7 +213,7 @@ TGuardian::TGuardian(TSentinelState::TPtr state, ui32 dataCenterRatio, ui32 room
 }
 
 TClusterMap::TPDiskIDSet TGuardian::GetAllowedPDisks(const TClusterMap& all, TString& issues,
-        TPDiskIDSet& disallowed) const {
+        TPDiskIgnoredMap& disallowed) const {
     TPDiskIDSet result;
     TStringBuilder issuesBuilder;
 
@@ -232,7 +232,9 @@ TClusterMap::TPDiskIDSet TGuardian::GetAllowedPDisks(const TClusterMap& all, TSt
             result.insert(kv.second.begin(), kv.second.end());
         } else {
             LOG_IGNORED(DataCenter);
-            disallowed.insert(kv.second.begin(), kv.second.end());
+            for (auto& pdisk : kv.second) {
+                disallowed.emplace(pdisk, NKikimrCms::TPDiskInfo::RATIO_BY_DATACENTER);
+            }
         }
     }
 
@@ -241,7 +243,9 @@ TClusterMap::TPDiskIDSet TGuardian::GetAllowedPDisks(const TClusterMap& all, TSt
 
         if (kv.first && !CheckRatio(kv, all.ByRoom, RoomRatio)) {
             LOG_IGNORED(Room);
-            disallowed.insert(kv.second.begin(), kv.second.end());
+            for (auto& pdisk : kv.second) {
+                disallowed.emplace(pdisk, NKikimrCms::TPDiskInfo::RATIO_BY_ROOM);
+            }
             EraseNodesIf(result, [&room = kv.second](const TPDiskID& id) {
                 return room.contains(id);
             });
@@ -257,7 +261,9 @@ TClusterMap::TPDiskIDSet TGuardian::GetAllowedPDisks(const TClusterMap& all, TSt
         }
         if (kv.first && !CheckRatio(kv, all.ByRack, RackRatio)) {
             LOG_IGNORED(Rack);
-            disallowed.insert(kv.second.begin(), kv.second.end());
+            for (auto& pdisk : kv.second) {
+                disallowed.emplace(pdisk, NKikimrCms::TPDiskInfo::RATIO_BY_RACK);
+            }
             EraseNodesIf(result, [&rack = kv.second](const TPDiskID& id) {
                 return rack.contains(id);
             });
@@ -967,6 +973,7 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
             if (!SentinelState->Nodes.contains(id.NodeId)) {
                 LOG_E("Missing node info"
                     << ": pdiskId# " << id);
+                info.IgnoreReason = NKikimrCms::TPDiskInfo::MISSING_NODE;
                 continue;
             }
 
@@ -983,7 +990,7 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
         }
 
         TString issues;
-        THashSet<TPDiskID, TPDiskIDHash> disallowed;
+        TClusterMap::TPDiskIgnoredMap disallowed;
         TClusterMap::TPDiskIDSet allowed = changed.GetAllowedPDisks(all, issues, disallowed);
         std::move(alwaysAllowed.begin(), alwaysAllowed.end(), std::inserter(allowed, allowed.begin()));
 
@@ -991,6 +998,8 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
             Y_VERIFY(SentinelState->PDisks.contains(id));
             TPDiskInfo::TPtr info = SentinelState->PDisks.at(id);
 
+            info->IgnoreReason = NKikimrCms::TPDiskInfo::NOT_IGNORED;
+
             if (!info->IsChangingAllowed()) {
                 info->AllowChanging();
                 continue;
@@ -1019,9 +1028,11 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
             }
         }
 
-        for (const auto& id : disallowed) {
+        for (const auto& [id, reason] : disallowed) {
             Y_VERIFY(SentinelState->PDisks.contains(id));
-            SentinelState->PDisks.at(id)->DisallowChanging();
+            auto& pdisk = SentinelState->PDisks.at(id);
+            pdisk->DisallowChanging();
+            pdisk->IgnoreReason = reason;
         }
 
         if (issues) {
@@ -1127,6 +1138,7 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
                         entry.MutableInfo()->SetPrevDesiredStatus(info->PrevStatusChangerState->Status);
                         entry.MutableInfo()->SetPrevStatusChangeAttempts(info->PrevStatusChangerState->Attempt);
                     }
+                    entry.MutableInfo()->SetIgnoreReason(info->IgnoreReason);
                 }
             }
         }
diff --git a/ydb/core/cms/sentinel_impl.h b/ydb/core/cms/sentinel_impl.h
index 00029ed6169..bdb380858b1 100644
--- a/ydb/core/cms/sentinel_impl.h
+++ b/ydb/core/cms/sentinel_impl.h
@@ -88,10 +88,13 @@ struct TPDiskInfo
 {
     using TPtr = TIntrusivePtr<TPDiskInfo>;
 
+    using EIgnoreReason = NKikimrCms::TPDiskInfo::EIgnoreReason;
+
     TActorId StatusChanger;
     TInstant LastStatusChange;
     TStatusChangerState::TPtr StatusChangerState;
     TStatusChangerState::TPtr PrevStatusChangerState;
+    EIgnoreReason IgnoreReason = NKikimrCms::TPDiskInfo::NOT_IGNORED;
 
     explicit TPDiskInfo(EPDiskStatus initialStatus, const ui32& defaultStateLimit, const TLimitsMap& stateLimits);
 
@@ -137,6 +140,7 @@ struct TSentinelState: public TSimpleRefCount<TSentinelState> {
 class TClusterMap {
 public:
     using TPDiskIDSet = THashSet<TPDiskID, TPDiskIDHash>;
+    using TPDiskIgnoredMap = THashMap<TPDiskID, TPDiskInfo::EIgnoreReason, TPDiskIDHash>;
     using TDistribution = THashMap<TString, TPDiskIDSet>;
     using TNodeIDSet = THashSet<ui32>;
 
@@ -163,7 +167,7 @@ class TGuardian : public TClusterMap {
 public:
     explicit TGuardian(TSentinelState::TPtr state, ui32 dataCenterRatio = 100, ui32 roomRatio = 100, ui32 rackRatio = 100);
 
-    TPDiskIDSet GetAllowedPDisks(const TClusterMap& all, TString& issues, TPDiskIDSet& disallowed) const;
+    TPDiskIDSet GetAllowedPDisks(const TClusterMap& all, TString& issues, TPDiskIgnoredMap& disallowed) const;
 
 private:
     const ui32 DataCenterRatio;
diff --git a/ydb/core/cms/sentinel_ut.cpp b/ydb/core/cms/sentinel_ut.cpp
index 0e1f33ae94d..1867ef3d032 100644
--- a/ydb/core/cms/sentinel_ut.cpp
+++ b/ydb/core/cms/sentinel_ut.cpp
@@ -174,6 +174,16 @@ Y_UNIT_TEST_SUITE(TSentinelBaseTests) {
         return {state, sentinelState};
     }
 
+    THashSet<TPDiskID, TPDiskIDHash> MapKeys(TClusterMap::TPDiskIgnoredMap& map) {
+        THashSet<TPDiskID, TPDiskIDHash> result;
+
+        for (auto& [k, _] : map) {
+            result.insert(k);
+        }
+
+        return result;
+    };
+
     void GuardianDataCenterRatio(ui16 numDataCenter, const TVector<ui16>& nodesPerDataCenterVariants, bool anyDC = false) {
         UNIT_ASSERT(!anyDC || numDataCenter == 1);
 
@@ -198,7 +208,7 @@ Y_UNIT_TEST_SUITE(TSentinelBaseTests) {
             }
 
             TString issues;
-            THashSet<TPDiskID, TPDiskIDHash> disallowed;
+            TClusterMap::TPDiskIgnoredMap disallowed;
 
             UNIT_ASSERT_VALUES_EQUAL(changed.GetAllowedPDisks(all, issues, disallowed), changedSet);
             UNIT_ASSERT(disallowed.empty());
@@ -218,7 +228,7 @@ Y_UNIT_TEST_SUITE(TSentinelBaseTests) {
             disallowed.clear();
             if (!anyDC) {
                 UNIT_ASSERT(changed.GetAllowedPDisks(all, issues, disallowed).empty());
-                UNIT_ASSERT_VALUES_EQUAL(disallowed, changedSet);
+                UNIT_ASSERT_VALUES_EQUAL(MapKeys(disallowed), changedSet);
                 UNIT_ASSERT_STRING_CONTAINS(issues, "due to DataCenterRatio");
             } else {
                 UNIT_ASSERT_VALUES_EQUAL(changed.GetAllowedPDisks(all, issues, disallowed), changedSet);
@@ -259,7 +269,7 @@ Y_UNIT_TEST_SUITE(TSentinelBaseTests) {
             }
 
             TString issues;
-            THashSet<TPDiskID, TPDiskIDHash> disallowed;
+            TClusterMap::TPDiskIgnoredMap disallowed;
 
             UNIT_ASSERT_VALUES_EQUAL(changed.GetAllowedPDisks(all, issues, disallowed), changedSet);
             UNIT_ASSERT(disallowed.empty());
@@ -287,7 +297,7 @@ Y_UNIT_TEST_SUITE(TSentinelBaseTests) {
                 UNIT_ASSERT(issues.empty());
             } else {
                 UNIT_ASSERT_VALUES_EQUAL(allowed, decltype(allowed){});
-                UNIT_ASSERT_VALUES_EQUAL(disallowed, changedSet);
+                UNIT_ASSERT_VALUES_EQUAL(MapKeys(disallowed), changedSet);
                 UNIT_ASSERT_STRING_CONTAINS(issues, "due to RackRatio");
             }
         }
diff --git a/ydb/core/cms/ui/sentinel_state.js b/ydb/core/cms/ui/sentinel_state.js
index d151441b059..9ae740493d4 100644
--- a/ydb/core/cms/ui/sentinel_state.js
+++ b/ydb/core/cms/ui/sentinel_state.js
@@ -44,6 +44,7 @@ const PDiskHeaders = [
     "PrevDesiredStatus",
     "PrevStatusChangeAttempts",
     "LastStatusChange",
+    "IgnoreReason",
 ];
 
 class CmsSentinelState {
@@ -145,6 +146,7 @@ class CmsSentinelState {
             "PrevDesiredStatus": this.id.bind(this),
             "PrevStatusChangeAttempts": this.id.bind(this),
             "LastStatusChange": this.id.bind(this),
+            "IgnoreReason": this.id.bind(this),
         };
     }
 
diff --git a/ydb/core/protos/cms.proto b/ydb/core/protos/cms.proto
index 85168eab00a..05ae41eac08 100644
--- a/ydb/core/protos/cms.proto
+++ b/ydb/core/protos/cms.proto
@@ -601,11 +601,20 @@ message TGetSentinelStateRequest {
         SUSPICIOUS = 2;
         ALL = 3;
     }
+
     optional EShow Show = 1;
     repeated TFilterRange Ranges = 2;
 }
 
 message TPDiskInfo {
+    enum EIgnoreReason {
+        NOT_IGNORED = 1;
+        MISSING_NODE = 2;
+        RATIO_BY_DATACENTER = 3;
+        RATIO_BY_ROOM = 4;
+        RATIO_BY_RACK = 5;
+    }
+
     optional uint32 State = 1; // EPDiskState
     optional uint32 PrevState = 2; // EPDiskState
     optional uint64 StateCounter = 3;
@@ -617,6 +626,7 @@ message TPDiskInfo {
     optional uint32 PrevDesiredStatus = 9;
     optional uint32 PrevStatusChangeAttempts = 10;
     optional string LastStatusChange = 11;
+    optional EIgnoreReason IgnoreReason = 12;
 }
 
 message TPDisk {