Add disable evict vdisks option to config

This reverts commit 95fa103.

Author: pixcc

ydb/core/cms/cms.cpp

@@ -505,6 +505,12 @@ bool TCms::CheckEvictVDisks(const TAction &action, TErrorInfo &error) const {
         return false;
     }
 
+    if (State->Config.SentinelConfig.EvictVDisksStatus.Empty()) {
+        error.Code = TStatus::ERROR;
+        error.Reason = "Evict vdisks is disabled in Sentinel (self heal)";
+        return false;
+    }
+
     switch (action.GetType()) {
         case TAction::RESTART_SERVICES:
         case TAction::SHUTDOWN_HOST:

ydb/core/cms/cms_ut.cpp

@@ -1853,6 +1853,46 @@ Y_UNIT_TEST_SUITE(TCmsTest) {
         env.CheckDonePermission("user", permission2.GetPermissions(0).GetId());
     }
 
+    Y_UNIT_TEST(DisabledEvictVDisks)
+    {
+        auto opts = TTestEnvOpts(8).WithSentinel();
+        TCmsTestEnv env(opts);
+        env.SetLogPriority(NKikimrServices::CMS, NLog::PRI_DEBUG);
+
+        // Make transition faster for tests purposes
+        auto cmsConfig = env.GetCmsConfig();
+        cmsConfig.MutableSentinelConfig()->SetDefaultStateLimit(1);
+        env.SetCmsConfig(cmsConfig);
+
+        // Evict VDisks
+        auto request = env.CheckPermissionRequest(
+            MakePermissionRequest(TRequestOptions("user").WithEvictVDisks(),
+                MakeAction(TAction::RESTART_SERVICES, env.GetNodeId(0), 600000000, "storage")
+            ),
+            TStatus::DISALLOW_TEMP // ok, waiting for move VDisks
+        );
+     
+        // Check that FAULTY BSC request is sent
+        env.CheckBSCUpdateRequests({ env.GetNodeId(0) }, NKikimrBlobStorage::FAULTY);
+
+        // Disable VDisks eviction
+        cmsConfig.MutableSentinelConfig()->SetEvictVDisksStatus(NKikimrCms::TCmsConfig::TSentinelConfig::DISABLED);
+        env.SetCmsConfig(cmsConfig);
+
+        // Check that ACTIVE BSC request is sent
+        env.CheckBSCUpdateRequests({ env.GetNodeId(0) }, NKikimrBlobStorage::ACTIVE);
+
+        // Check that CMS returns ERROR when VDisks eviction is disabled
+        env.CheckRequest("user", request.GetRequestId(), false, TStatus::ERROR, 0);
+
+        // Enable VDisks eviction again
+        cmsConfig.MutableSentinelConfig()->SetEvictVDisksStatus(NKikimrCms::TCmsConfig::TSentinelConfig::FAULTY);
+        env.SetCmsConfig(cmsConfig);
+
+        // Check that FAULTY BSC request is sent again
+        env.CheckBSCUpdateRequests({ env.GetNodeId(0) }, NKikimrBlobStorage::FAULTY);
+    }
+
     Y_UNIT_TEST(EmergencyDuringRollingRestart)
     {
         TCmsTestEnv env(8);

ydb/core/cms/config.h

@@ -1,12 +1,14 @@
 #pragma once
 
 #include "pdisk_state.h"
+#include "pdisk_status.h"
 
 #include <ydb/core/protos/cms.pb.h>
 
 #include <util/datetime/base.h>
 #include <util/generic/hash.h>
 #include <util/generic/map.h>
+#include <util/generic/maybe.h>
 
 namespace NKikimr::NCms {
 
@@ -30,6 +32,8 @@ struct TCmsSentinelConfig {
     ui32 RoomRatio;
     ui32 RackRatio;
 
+    TMaybeFail<EPDiskStatus> EvictVDisksStatus;
+
     void Serialize(NKikimrCms::TCmsConfig::TSentinelConfig &config) const {
         config.SetEnable(Enable);
         config.SetDryRun(DryRun);
@@ -45,6 +49,7 @@ struct TCmsSentinelConfig {
         config.SetRackRatio(RackRatio);
 
         SaveStateLimits(config);
+        SaveEvictVDisksStatus(config);
     }
 
     void Deserialize(const NKikimrCms::TCmsConfig::TSentinelConfig &config) {
@@ -63,6 +68,8 @@ struct TCmsSentinelConfig {
 
         auto newStateLimits = LoadStateLimits(config);
         StateLimits.swap(newStateLimits);
+
+        EvictVDisksStatus = LoadEvictVDisksStatus(config);
     }
 
     void SaveStateLimits(NKikimrCms::TCmsConfig::TSentinelConfig &config) const {
@@ -129,6 +136,31 @@ struct TCmsSentinelConfig {
 
         return stateLimits;
     }
+
+    static TMaybeFail<EPDiskStatus> LoadEvictVDisksStatus(const NKikimrCms::TCmsConfig::TSentinelConfig &config) {
+        using EEvictVDisksStatus = NKikimrCms::TCmsConfig::TSentinelConfig;
+        switch (config.GetEvictVDisksStatus()) {
+            case EEvictVDisksStatus::UNKNOWN:
+            case EEvictVDisksStatus::FAULTY:
+                return EPDiskStatus::FAULTY;
+            case EEvictVDisksStatus::DISABLED:
+                return Nothing();
+        }
+        return EPDiskStatus::FAULTY;
+    }
+
+    void SaveEvictVDisksStatus(NKikimrCms::TCmsConfig::TSentinelConfig &config) const {
+        using EEvictVDisksStatus = NKikimrCms::TCmsConfig::TSentinelConfig;
+
+        if (EvictVDisksStatus.Empty()) {
+            config.SetEvictVDisksStatus(EEvictVDisksStatus::DISABLED);
+            return;
+        }
+
+        if (*EvictVDisksStatus == EPDiskStatus::FAULTY) {
+            config.SetEvictVDisksStatus(EEvictVDisksStatus::FAULTY);
+        }
+    }
 };
 
 struct TCmsLogConfig {

ydb/core/cms/sentinel.cpp

@@ -895,8 +895,8 @@ class TSentinel: public TActorBootstrapped<TSentinel> {
                 continue;
             }
 
-            if (it->second.HasFaultyMarker()) {
-                info.SetForcedStatus(EPDiskStatus::FAULTY);
+            if (it->second.HasFaultyMarker() && Config.EvictVDisksStatus.Defined()) {
+                info.SetForcedStatus(*Config.EvictVDisksStatus);
             } else {
                 info.ResetForcedStatus();
             }

ydb/core/protos/cms.proto

@@ -414,6 +414,12 @@ message TCmsConfig {
             optional uint32 Limit = 2;
         }
 
+        enum EEvictVDisksStatus {
+            UNKNOWN = 0;
+            DISABLED = 1;
+            FAULTY = 2;
+        }
+
         optional bool Enable = 1 [default = true];
         // Updater's config
         optional uint64 UpdateConfigInterval = 2 [default = 3600000000];
@@ -433,6 +439,7 @@ message TCmsConfig {
 
         optional bool DryRun = 13;
         repeated TStateLimit StateLimits = 14;
+        optional EEvictVDisksStatus EvictVDisksStatus = 15;
     }
 
     message TLogConfig {