Skip to content

Fix BSC reporting for Bridge mode #21101

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 15, 2025
Merged

Fix BSC reporting for Bridge mode #21101

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
101 changes: 100 additions & 1 deletion ydb/core/mind/bscontroller/bsc.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,103 @@ void TBlobStorageController::TGroupInfo::CalculateLayoutStatus(TBlobStorageContr
}
}

void TBlobStorageController::TGroupInfo::FillInGroupParameters(
NKikimrBlobStorage::TEvControllerSelectGroupsResult::TGroupParameters *params,
TBlobStorageController *self) const {
if (GroupMetrics) {
params->MergeFrom(GroupMetrics->GetGroupParameters());
} else if (BridgeGroupInfo) {
for (const auto& groupId : BridgeGroupInfo->GetBridgeGroupIds()) {
if (TGroupInfo *groupInfo = self->FindGroup(TGroupId::FromValue(groupId))) {
if (self->BridgeInfo && groupInfo->BridgePileId &&
self->BridgeInfo->GetPile(*groupInfo->BridgePileId)->State == NKikimrBridge::TClusterState::DISCONNECTED) {
continue; // ignore groups from disconnected piles
}
groupInfo->FillInGroupParameters(params, self);
} else {
Y_DEBUG_ABORT();
}
}
} else {
FillInResources(params->MutableAssuredResources(), true);
FillInResources(params->MutableCurrentResources(), false);
FillInVDiskResources(params);
}
}

void TBlobStorageController::TGroupInfo::FillInResources(
NKikimrBlobStorage::TEvControllerSelectGroupsResult::TGroupParameters::TResources *pb, bool countMaxSlots) const {
// count minimum params for each of slots assuming they are shared fairly between all the slots (expected or currently created)
std::optional<ui64> size;
std::optional<double> iops;
std::optional<ui64> readThroughput;
std::optional<ui64> writeThroughput;
std::optional<double> occupancy;
for (const TVSlotInfo *vslot : VDisksInGroup) {
const TPDiskInfo *pdisk = vslot->PDisk;
const auto& metrics = pdisk->Metrics;
const ui32 shareFactor = countMaxSlots ? pdisk->ExpectedSlotCount : pdisk->NumActiveSlots;
ui64 vdiskSlotSize = 0;
if (metrics.HasEnforcedDynamicSlotSize()) {
vdiskSlotSize = metrics.GetEnforcedDynamicSlotSize();
} else if (metrics.GetTotalSize()) {
vdiskSlotSize = metrics.GetTotalSize() / shareFactor;
}
if (vdiskSlotSize) {
size = Min(size.value_or(Max<ui64>()), vdiskSlotSize);
}
if (metrics.HasMaxIOPS()) {
iops = Min(iops.value_or(Max<double>()), metrics.GetMaxIOPS() * 100 / shareFactor * 0.01);
}
if (metrics.HasMaxReadThroughput()) {
readThroughput = Min(readThroughput.value_or(Max<ui64>()), metrics.GetMaxReadThroughput() / shareFactor);
}
if (metrics.HasMaxWriteThroughput()) {
writeThroughput = Min(writeThroughput.value_or(Max<ui64>()), metrics.GetMaxWriteThroughput() / shareFactor);
}
if (const auto& vm = vslot->Metrics; vm.HasOccupancy()) {
occupancy = Max(occupancy.value_or(0), vm.GetOccupancy());
}
}

// and recalculate it to the total size of the group according to the erasure
TBlobStorageGroupType type(ErasureSpecies);
const double factor = (double)VDisksInGroup.size() * type.DataParts() / type.TotalPartCount();
if (size) {
pb->SetSpace(Min<ui64>(pb->HasSpace() ? pb->GetSpace() : Max<ui64>(), *size * factor));
}
if (iops) {
pb->SetIOPS(Min<double>(pb->HasIOPS() ? pb->GetIOPS() : Max<double>(), *iops * VDisksInGroup.size() / type.TotalPartCount()));
}
if (readThroughput) {
pb->SetReadThroughput(Min<ui64>(pb->HasReadThroughput() ? pb->GetReadThroughput() : Max<ui64>(), *readThroughput * factor));
}
if (writeThroughput) {
pb->SetWriteThroughput(Min<ui64>(pb->HasWriteThroughput() ? pb->GetReadThroughput() : Max<ui64>(), *writeThroughput * factor));
}
if (occupancy) {
pb->SetOccupancy(Max<double>(pb->HasOccupancy() ? pb->GetOccupancy() : Min<double>(), *occupancy));
}
}

void TBlobStorageController::TGroupInfo::FillInVDiskResources(
NKikimrBlobStorage::TEvControllerSelectGroupsResult::TGroupParameters *pb) const {
TBlobStorageGroupType type(ErasureSpecies);
const double f = (double)VDisksInGroup.size() * type.DataParts() / type.TotalPartCount();
for (const TVSlotInfo *vslot : VDisksInGroup) {
const auto& m = vslot->Metrics;
if (m.HasAvailableSize()) {
pb->SetAvailableSize(Min<ui64>(pb->HasAvailableSize() ? pb->GetAvailableSize() : Max<ui64>(), f * m.GetAvailableSize()));
}
if (m.HasAllocatedSize()) {
pb->SetAllocatedSize(Max<ui64>(pb->HasAllocatedSize() ? pb->GetAllocatedSize() : 0, f * m.GetAllocatedSize()));
}
if (m.HasSpaceColor()) {
pb->SetSpaceColor(pb->HasSpaceColor() ? Max(pb->GetSpaceColor(), m.GetSpaceColor()) : m.GetSpaceColor());
}
}
}

NKikimrBlobStorage::TGroupStatus::E TBlobStorageController::DeriveStatus(const TBlobStorageGroupInfo::TTopology *topology,
const TBlobStorageGroupInfo::TGroupVDisks& failed) {
auto& checker = *topology->QuorumChecker;
Expand Down Expand Up @@ -189,7 +286,9 @@ void TBlobStorageController::Handle(TEvNodeWardenStorageConfig::TPtr ev) {
StaticVDiskMap.emplace(TVDiskID(vdiskId.GroupID, 0, vdiskId), vslotId);
++StaticPDisks.at(pdiskId).StaticSlotUsage;
SysViewChangedVSlots.insert(vslotId);
SysViewChangedGroups.insert(vdiskId.GroupID);
}
for (const auto& group : ss.GetGroups()) {
SysViewChangedGroups.insert(TGroupId::FromProto(&group, &NKikimrBlobStorage::TGroupInfo::GetGroupID));
}
} else {
Y_FAIL("no storage configuration provided");
Expand Down
2 changes: 2 additions & 0 deletions ydb/core/mind/bscontroller/cmds_storage_pool.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -169,6 +169,8 @@ namespace NKikimr::NBsController {
GroupContentChanged.insert(it->second);
}
}
// retain some fields
storagePool.BridgeMode = cur.BridgeMode;
cur = std::move(storagePool); // update existing storage pool
} else {
// enable bridge mode by default for new pools (when bridge mode is enabled cluster-wide)
Expand Down
11 changes: 5 additions & 6 deletions ydb/core/mind/bscontroller/config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -712,6 +712,9 @@ namespace NKikimr::NBsController {
}
for (const auto& [base, overlay] : state.Groups.Diff()) {
SysViewChangedGroups.insert(overlay->first);
if (overlay->second && overlay->second->BridgeProxyGroupId) {
SysViewChangedGroups.insert(*overlay->second->BridgeProxyGroupId);
}
}
for (const auto& [prev, cur] : Diff(&StoragePools, &state.StoragePools.Get())) {
SysViewChangedStoragePools.insert(cur ? cur->first : prev->first);
Expand Down Expand Up @@ -1109,10 +1112,7 @@ namespace NKikimr::NBsController {
pb->SetExpectedStatus(status.ExpectedStatus);

if (group.BridgeGroupInfo) {
NKikimrBlobStorage::TGroupInfo groupInfoPb;
bool success = groupInfoPb.ParseFromString(*group.BridgeGroupInfo);
Y_DEBUG_ABORT_UNLESS(success);
pb->SetIsProxyGroup(groupInfoPb.BridgeGroupIdsSize() != 0);
pb->SetIsProxyGroup(group.BridgeGroupInfo->BridgeGroupIdsSize() != 0);
}

if (group.DecommitStatus != NKikimrBlobStorage::TGroupDecommitStatus::NONE || group.VirtualGroupState) {
Expand Down Expand Up @@ -1241,8 +1241,7 @@ namespace NKikimr::NBsController {
group->SetGroupSizeInUnits(groupInfo.GroupSizeInUnits);

if (groupInfo.BridgeGroupInfo) {
const bool success = group->MergeFromString(*groupInfo.BridgeGroupInfo);
Y_DEBUG_ABORT_UNLESS(success);
group->MergeFrom(*groupInfo.BridgeGroupInfo);
}
}

Expand Down
34 changes: 14 additions & 20 deletions ydb/core/mind/bscontroller/config_fit_groups.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,7 @@ namespace NKikimr {
false, /* down */
false, /* seenOperational */
0, /* groupSizeInUnits */
std::nullopt, /* bridgePileId */
StoragePoolId, /* storagePoolId */
0, /* numFailRealms */
0, /* numFailDomainsPerFailRealm */
Expand All @@ -120,22 +121,19 @@ namespace NKikimr {
auto& index = State.IndexGroupSpeciesToGroup.Unshare();
index[species].push_back(mainGroupId);

NKikimrBlobStorage::TGroupInfo mainGroup;
NKikimrBlobStorage::TGroupInfo& mainGroup = groupInfo->BridgeGroupInfo.emplace();
const auto& bridgeInfo = State.BridgeInfo;
Y_ABORT_UNLESS(bridgeInfo);
bridgeInfo->ForEachPile([&](TBridgePileId bridgePileId) {
const TGroupId groupId = CreateGroup(bridgePileId);
const TGroupId groupId = CreateGroup(bridgePileId, mainGroupId);
groupId.CopyToProto(&mainGroup, &NKikimrBlobStorage::TGroupInfo::AddBridgeGroupIds);
});

const bool success = mainGroup.SerializeToString(&groupInfo->BridgeGroupInfo.ConstructInPlace());
Y_DEBUG_ABORT_UNLESS(success);
} else {
CreateGroup(std::nullopt); // regular single group
CreateGroup(std::nullopt, std::nullopt); // regular single group
}
}

TGroupId CreateGroup(std::optional<TBridgePileId> bridgePileId) {
TGroupId CreateGroup(std::optional<TBridgePileId> bridgePileId, std::optional<TGroupId> bridgeProxyGroupId) {
////////////////////////////////////////////////////////////////////////////////////////////
// ALLOCATE GROUP ID FOR THE NEW GROUP
////////////////////////////////////////////////////////////////////////////////////////////
Expand Down Expand Up @@ -187,12 +185,10 @@ namespace NKikimr {
TGroupInfo *groupInfo = State.Groups.ConstructInplaceNewEntry(groupId, groupId, 1,
0, Geometry.GetErasure(), desiredPDiskCategory.GetOrElse(0), StoragePool.VDiskKind,
StoragePool.EncryptionMode.GetOrElse(0), lifeCyclePhase, mainKeyId, encryptedGroupKey,
groupKeyNonce, MainKeyVersion, false, false, groupSizeInUnits, StoragePoolId, Geometry.GetNumFailRealms(),
Geometry.GetNumFailDomainsPerFailRealm(), Geometry.GetNumVDisksPerFailDomain());
groupKeyNonce, MainKeyVersion, false, false, groupSizeInUnits, bridgePileId, StoragePoolId,
Geometry.GetNumFailRealms(), Geometry.GetNumFailDomainsPerFailRealm(), Geometry.GetNumVDisksPerFailDomain());

if (bridgePileId) {
groupInfo->BridgePileId = *bridgePileId;
}
groupInfo->BridgeProxyGroupId = bridgeProxyGroupId;

// bind group to storage pool
State.StoragePoolGroups.Unshare().emplace(StoragePoolId, groupId);
Expand All @@ -216,10 +212,6 @@ namespace NKikimr {
throw TExFitGroupError() << "GroupId# " << groupId << " not found";
}

std::optional<TBridgePileId> bridgePileId = groupInfo->BridgePileId
? std::make_optional(*groupInfo->BridgePileId)
: std::nullopt;

TGroupMapper::TGroupDefinition group;
TGroupMapper::TGroupConstraintsDefinition softConstraints, hardConstraints;
bool layoutIsValid = true;
Expand Down Expand Up @@ -376,7 +368,7 @@ namespace NKikimr {
STLOG(PRI_INFO, BS_CONTROLLER, BSCFG01, "Attempt to sanitize group layout", (GroupId, groupId));
// Use group layout sanitizing algorithm on direct requests or when initial group layout is invalid
auto result = AllocateOrSanitizeGroup(groupId, group, {}, std::move(forbid), groupSizeInUnits, requiredSpace,
AllowUnusableDisks, bridgePileId, &TGroupGeometryInfo::SanitizeGroup);
AllowUnusableDisks, groupInfo->BridgePileId, &TGroupGeometryInfo::SanitizeGroup);

if (replacedSlots.empty()) {
// update information about replaced disks
Expand All @@ -398,10 +390,10 @@ namespace NKikimr {
try {
TGroupMapper::MergeTargetDiskConstraints(hardConstraints, softConstraints);
AllocateOrSanitizeGroup(groupId, group, softConstraints, replacedDisks, std::move(forbid), groupSizeInUnits, requiredSpace,
AllowUnusableDisks, bridgePileId, &TGroupGeometryInfo::AllocateGroup);
AllowUnusableDisks, groupInfo->BridgePileId, &TGroupGeometryInfo::AllocateGroup);
} catch (const TExFitGroupError& ex) {
AllocateOrSanitizeGroup(groupId, group, hardConstraints, replacedDisks, std::move(forbid), groupSizeInUnits, requiredSpace,
AllowUnusableDisks, bridgePileId, &TGroupGeometryInfo::AllocateGroup);
AllowUnusableDisks, groupInfo->BridgePileId, &TGroupGeometryInfo::AllocateGroup);
}
}
if (!IgnoreVSlotQuotaCheck) {
Expand Down Expand Up @@ -797,7 +789,9 @@ namespace NKikimr {

enumerateGroups([&](TGroupId groupId) {
fitter.CheckExistingGroup(groupId);
++numActualGroups;
if (const TGroupInfo *group = state.Groups.Find(groupId); group && !group->BridgePileId) {
++numActualGroups;
}
});
if (createNewGroups) {
if (numActualGroups < storagePool.NumGroups) {
Expand Down
2 changes: 1 addition & 1 deletion ydb/core/mind/bscontroller/group_metrics_exchange.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@ namespace NKikimr::NBsController {
if (TGroupInfo *group = Self->FindGroup(TGroupId::FromValue(groupId))) {
auto *item = outRecord.AddGroupMetrics();
item->SetGroupId(group->ID.GetRawId());
group->FillInGroupParameters(item->MutableGroupParameters());
group->FillInGroupParameters(item->MutableGroupParameters(), Self);
}
}
}
Expand Down
Loading
Loading