issue-2404: don't start migration if it is already finished (#2890)

* issue-2404: add check is migration finished and tests * issue-2404: correct issues * issue-2404: correct issues * issue-2404: correct issues * issue-2404: remove lines * issue-2404: fmt * issue-2404: correct issue
ydb-platform · Jan 30, 2025 · 4510ec7 · 4510ec7
1 parent 86fcf1f
commit 4510ec7
Show file tree

Hide file tree

Showing 5 changed files with 579 additions and 15 deletions.
diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_monitoring.cpp b/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_monitoring.cpp
@@ -896,7 +896,7 @@ void TDiskRegistryActor::RenderDiskHtmlInfo(
                         TABLEH() { out << "SeqNo"; }
                     }
 
-                    for (const auto& [uuid, seqNo]: info.FinishedMigrations) {
+                    for (const auto& [uuid, seqNo, _]: info.FinishedMigrations) {
                         TABLER() {
                             TABLED() { DumpDeviceLink(out, TabletID(), uuid); }
                             TABLED() { out << seqNo; }

diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_database.h b/cloud/blockstore/libs/storage/disk_registry/disk_registry_database.h
@@ -30,6 +30,10 @@ struct TFinishedMigration
 {
     TString DeviceId;
     ui64 SeqNo = 0;
+    bool IsCanceled =   // by default, this value is set to true, because
+                        // we may not start migration if this field is set to
+                        // false
+        true;
 };
 
 ////////////////////////////////////////////////////////////////////////////////

diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_state.cpp b/cloud/blockstore/libs/storage/disk_registry/disk_registry_state.cpp
@@ -2975,7 +2975,7 @@ auto TDiskRegistryState::DeallocateSimpleDisk(
         }
     }
 
-    for (const auto& [uuid, seqNo]: disk.FinishedMigrations) {
+    for (const auto& [uuid, seqNo, _]: disk.FinishedMigrations) {
         Y_UNUSED(seqNo);
 
         if (DeviceList.ReleaseDevice(uuid)) {
@@ -4874,7 +4874,7 @@ NProto::TDiskConfig TDiskRegistryState::BuildDiskConfig(
     config.SetStorageMediaKind(diskState.MediaKind);
     config.SetMigrationStartTs(diskState.MigrationStartTs.MicroSeconds());
 
-    for (const auto& [uuid, seqNo]: diskState.FinishedMigrations) {
+    for (const auto& [uuid, seqNo, _]: diskState.FinishedMigrations) {
         Y_UNUSED(seqNo);
         auto& m = *config.AddFinishedMigrations();
         m.SetDeviceId(uuid);
@@ -5023,19 +5023,16 @@ void TDiskRegistryState::ApplyAgentStateChange(
         }
 
         if (agent.GetState() == NProto::AGENT_STATE_WARNING) {
-            if (disk.MigrationSource2Target.contains(deviceId)) {
-                // migration already started
-                continue;
-            }
-
             if (Find(disk.Devices, deviceId) == disk.Devices.end()) {
                 ReportDiskRegistryWrongMigratedDeviceOwnership(
                     TStringBuilder() << "ApplyAgentStateChange: device "
                                      << deviceId << " not found");
                 continue;
             }
 
-            AddMigration(disk, diskId, deviceId);
+            if (MigrationCanBeStarted(disk, deviceId)) {
+                AddMigration(disk, diskId, deviceId);
+            }
         } else {
             if (agent.GetState() == NProto::AGENT_STATE_UNAVAILABLE
                     && disk.MasterDiskId)
@@ -6019,7 +6016,7 @@ void TDiskRegistryState::ApplyDeviceStateChange(
         return;
     }
 
-    if (!disk->MigrationSource2Target.contains(uuid)) {
+    if (MigrationCanBeStarted(*disk, uuid)) {
         AddMigration(*disk, diskId, uuid);
     }
 }
@@ -6092,10 +6089,8 @@ void TDiskRegistryState::CancelDeviceMigration(
 
     const ui64 seqNo = AddReallocateRequest(db, diskId);
 
-    disk.FinishedMigrations.push_back({
-        .DeviceId = targetId,
-        .SeqNo = seqNo
-    });
+    disk.FinishedMigrations.push_back(
+        {.DeviceId = targetId, .SeqNo = seqNo, .IsCanceled = true});
 
     NProto::TDiskHistoryItem historyItem;
     historyItem.SetTimestamp(now.MicroSeconds());
@@ -6164,7 +6159,9 @@ NProto::TError TDiskRegistryState::FinishDeviceMigration(
 
     const ui64 seqNo = AddReallocateRequest(db, diskId);
     *devIt = targetId;
-    disk.FinishedMigrations.push_back({.DeviceId = sourceId, .SeqNo = seqNo});
+
+    disk.FinishedMigrations.push_back(
+        {.DeviceId = sourceId, .SeqNo = seqNo, .IsCanceled = false});
 
     if (disk.MasterDiskId) {
         const bool replaced =
@@ -7561,4 +7558,24 @@ std::optional<ui64> TDiskRegistryState::GetDiskBlockCount(
     return diskInfo.GetBlocksCount();
 }
 
+// static
+bool TDiskRegistryState::MigrationCanBeStarted(
+    const TDiskState& disk,
+    const TString& deviceUUID)
+{
+    if (disk.MigrationSource2Target.contains(deviceUUID)) {
+        // migration already started
+        return false;
+    }
+
+    for (const auto& m: disk.FinishedMigrations) {
+        if (m.DeviceId == deviceUUID && !m.IsCanceled) {
+            // there is a finished migration for the device
+            return false;
+        }
+    }
+
+    return true;
+}
+
 }   // namespace NCloud::NBlockStore::NStorage
diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_state.h b/cloud/blockstore/libs/storage/disk_registry/disk_registry_state.h
@@ -1312,6 +1312,10 @@ class TDiskRegistryState
     void CleanupAgentConfig(
         TDiskRegistryDatabase& db,
         const NProto::TAgentConfig& agent);
+
+    static bool MigrationCanBeStarted(
+        const TDiskState& disk,
+        const TString& deviceUUID);
 };
 
 }   // namespace NCloud::NBlockStore::NStorage