add health check overload shard hint (ydb-platform#14416)

Author: adameat

ydb/core/health_check/health_check.cpp

@@ -714,7 +714,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
 
     THashMap<TString, THintOverloadedShard> OverloadedShardHints;
     static constexpr size_t MAX_OVERLOADED_SHARDS_HINTS = 10;
-    static constexpr double OVERLOADED_SHARDS_CPU_CORES = 0.01; /* HACK */
+    static constexpr double OVERLOADED_SHARDS_CPU_CORES = 0.75;
 
     struct TTabletRequestsState {
         struct TTabletState {
@@ -992,7 +992,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
     [[nodiscard]] TRequestResponse<TEvSysView::TEvGetPartitionStatsResult> RequestPartitionStats(TTabletId schemeShardId, TSubDomainKey subDomainKey) {
         THolder<TEvSysView::TEvGetPartitionStats> request = MakeHolder<TEvSysView::TEvGetPartitionStats>();
         NKikimrSysView::TEvGetPartitionStats& record = request->Record;
-        record.MutableFilter()->MutableNotLess()->SetCPUCores(OVERLOADED_SHARDS_CPU_CORES); //
+        record.MutableFilter()->MutableNotLess()->SetCPUCores(OVERLOADED_SHARDS_CPU_CORES);
         record.SetDomainKeyOwnerId(subDomainKey.GetSchemeShard());
         record.SetDomainKeyPathId(subDomainKey.GetPathId());
         return RequestTabletPipe<TEvSysView::TEvGetPartitionStatsResult>(schemeShardId, request.Release(), TTabletRequestsState::RequestGetPartitionStats);
@@ -1455,7 +1455,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
 
     bool CheckOverloadedShardHint(const TEvSchemeShard::TEvDescribeSchemeResult& response, THintOverloadedShard& hint) {
         const auto& policy(response.GetRecord().GetPathDescription().GetTable().GetPartitionConfig().GetPartitioningPolicy());
-        if (/*HACK ! */policy.GetSplitByLoadSettings().GetEnabled()) {
+        if (!policy.GetSplitByLoadSettings().GetEnabled()) {
             hint.Message = "Split by load is disabled on the table"; // do not change without changing the logic in the UI
             return true;
         }

ydb/core/sys_view/partition_stats/partition_stats.cpp

@@ -309,6 +309,27 @@ class TPartitionStatsCollector : public TActorBootstrapped<TPartitionStatsCollec
         }
 
         bool includePathColumn = !record.HasIncludePathColumn() || record.GetIncludePathColumn();
+        bool hasFilter = record.HasFilter();
+        auto matchesFilter = [&](const NKikimrSysView::TPartitionStats& stats) {
+            const auto& filter = record.GetFilter();
+            if (filter.HasNotLess()) {
+                if (filter.GetNotLess().HasCPUCores() && stats.GetCPUCores() < filter.GetNotLess().GetCPUCores()) {
+                    return false;
+                }
+            }
+            return true;
+        };
+        auto addStats = [&](const auto& pathId, const auto& tableStats, ui64 partIdx) {
+            auto stats = result->Record.AddStats();
+            auto* key = stats->MutableKey();
+            key->SetOwnerId(pathId.OwnerId);
+            key->SetPathId(pathId.LocalPathId);
+            key->SetPartIdx(partIdx);
+            if (includePathColumn) {
+                stats->SetPath(tableStats.Path);
+            }
+            return stats;
+        };
 
         auto matchesFilter = [&](const NKikimrSysView::TPartitionStats& stats) {
                 if (record.HasFilter()) {
@@ -335,35 +356,24 @@ class TPartitionStatsCollector : public TActorBootstrapped<TPartitionStatsCollec
 
             for (ui64 partIdx = startPartIdx; partIdx < end; ++partIdx) {
                 NKikimrSysView::TPartitionStatsResult* stats = nullptr;
+                if (!hasFilter) {
+                    stats = addStats(pathId, tableStats, partIdx);
+                }
                 auto shardIdx = tableStats.ShardIndices[partIdx];
                 auto part = tableStats.Partitions.find(shardIdx);
                 if (part != tableStats.Partitions.end()) {
                     for (const auto& followerStat : part->second.FollowerStats) {
-                        if (!matchesFilter(followerStat.second)) {
+                        if (hasFilter && !matchesFilter(followerStat.second)) {
                             continue;
                         }
-                        if (stats == nullptr) {
-                            stats = result->Record.AddStats();
+                        if (!stats) {
+                            stats = addStats(pathId, tableStats, partIdx);
                         }
                         *stats->AddStats() = followerStat.second;
                     }
                 }
 
-                if (!stats) {
-                    continue;
-                }
-
-                auto* key = stats->MutableKey();
-
-                key->SetOwnerId(pathId.OwnerId);
-                key->SetPathId(pathId.LocalPathId);
-                key->SetPartIdx(partIdx);
-
-                if (includePathColumn) {
-                    stats->SetPath(tableStats.Path);
-                }
-
-                if (++count == BatchSize) {
+                if (stats && ++count == BatchSize) {
                     auto* next = result->Record.MutableNext();
                     next->SetOwnerId(pathId.OwnerId);
                     next->SetPathId(pathId.LocalPathId);