Skip to content

HC made storage issue RED when no bsc reply #2578

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 7 commits into from
Mar 11, 2024
Merged

HC made storage issue RED when no bsc reply #2578

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
c4ce41e
NBYDB-15 no-bsc-reply
StekPerepolnen Mar 8, 2024
a857e10
remove AuthenticGroups
StekPerepolnen Mar 8, 2024
7e4c254
remove comments
StekPerepolnen Mar 8, 2024
a6a304d
added test
StekPerepolnen Mar 8, 2024
1178349
fixed test
StekPerepolnen Mar 8, 2024
926f823
remove logs
StekPerepolnen Mar 8, 2024
f6dbe62
don't report RED storage
StekPerepolnen Mar 11, 2024
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
131 changes: 40 additions & 91 deletions ydb/core/health_check/health_check.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,21 +219,20 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
};

struct TStoragePoolState {
TString Kind;
TString Name;
THashSet<TGroupId> Groups;
THashSet<TGroupId> AuthenticGroups;
};

struct TDatabaseState {
TTabletId HiveId = {};
TPathId ResourcePathId = {};
TVector<TNodeId> ComputeNodeIds;
THashSet<TString> StoragePoolNames;
THashSet<ui64> StoragePools;
THashMap<std::pair<TTabletId, NNodeWhiteboard::TFollowerId>, const NKikimrHive::TTabletInfo*> MergedTabletState;
THashMap<TNodeId, TNodeTabletState> MergedNodeTabletState;
THashMap<TNodeId, ui32> NodeRestartsPerPeriod;
ui64 StorageQuota;
ui64 StorageUsage;
ui64 StorageQuota = 0;
ui64 StorageUsage = 0;
TMaybeServerlessComputeResourcesMode ServerlessComputeResourcesMode;
};

Expand Down Expand Up @@ -431,6 +430,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
TTabletId RootHiveId;
THashMap<TString, TTenantInfo> TenantByPath;
THashMap<TString, THolder<NSchemeShard::TEvSchemeShard::TEvDescribeSchemeResult>> DescribeByPath;
THashMap<TString, THashSet<TString>> PathsByPoolName;
THashMap<TString, Ydb::Cms::GetDatabaseStatusResult> DatabaseStatusByPath;
THashMap<TString, THolder<NTenantSlotBroker::TEvTenantSlotBroker::TEvTenantState>> TenantStateByPath;
THashMap<TString, THolder<NSchemeCache::TSchemeCacheNavigate>> NavigateResult;
Expand All @@ -442,7 +442,6 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
THolder<TEvBlobStorage::TEvControllerConfigResponse> BaseConfig;

THashSet<TNodeId> NodeIds;
THashSet<TNodeId> StorageNodeIds;
THashSet<TNodeId> ComputeNodeIds;
std::unordered_map<std::pair<TNodeId, int>, ui32> NodeRetries;
ui32 MaxRetries = 20;
Expand All @@ -463,8 +462,8 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
std::unordered_set<TString> ValidPDisks;
std::unordered_set<TGroupId> ValidGroups;

THashMap<TString, TStoragePoolState> StoragePoolState;
THashSet<TString> StoragePoolSeen;
THashMap<ui64, TStoragePoolState> StoragePoolState;
THashSet<ui64> StoragePoolSeen;

THashSet<TNodeId> UnavailableComputeNodes;

Expand Down Expand Up @@ -585,37 +584,10 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {

Send(GetNameserviceActorId(), new TEvInterconnect::TEvListNodes());
++Requests;
Send(MakeBlobStorageNodeWardenID(SelfId().NodeId()), new TEvNodeWardenQueryStorageConfig(false));
++Requests;

Become(&TThis::StateWait, Timeout, new TEvents::TEvWakeup());
}

void Handle(TEvNodeWardenStorageConfig::TPtr ev) {
if (const NKikimrBlobStorage::TStorageConfig& config = *ev->Get()->Config; config.HasBlobStorageConfig()) {
if (const auto& bsConfig = config.GetBlobStorageConfig(); bsConfig.HasServiceSet()) {
const auto& staticConfig = bsConfig.GetServiceSet();
for (const NKikimrBlobStorage::TNodeWardenServiceSet_TPDisk& pDisk : staticConfig.pdisks()) {
RequestStorageNode(pDisk.GetNodeID());
}
for (const NKikimrBlobStorage::TGroupInfo& group : staticConfig.groups()) {
ValidGroups.emplace(group.GetGroupID());
TString storagePoolName = group.GetStoragePoolName();
if (!storagePoolName) {
storagePoolName = STATIC_STORAGE_POOL_NAME;
}
StoragePoolState[storagePoolName].Groups.emplace(group.groupid());

if (!IsSpecificDatabaseFilter()) {
DatabaseState[DomainPath].StoragePoolNames.emplace(storagePoolName);
}
}
}
}

RequestDone("TEvNodeWardenStorageConfig");
}

STATEFN(StateWait) {
switch (ev->GetTypeRewrite()) {
hFunc(TEvents::TEvUndelivered, Handle);
Expand All @@ -627,15 +599,13 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
hFunc(TEvHive::TEvResponseHiveInfo, Handle);
hFunc(NSchemeShard::TEvSchemeShard::TEvDescribeSchemeResult, Handle);
hFunc(TEvTxProxySchemeCache::TEvNavigateKeySetResult, Handle)
hFunc(TEvBlobStorage::TEvControllerSelectGroupsResult, Handle);
hFunc(TEvBlobStorage::TEvControllerConfigResponse, Handle);
hFunc(NNodeWhiteboard::TEvWhiteboard::TEvSystemStateResponse, Handle);
hFunc(TEvInterconnect::TEvNodeDisconnected, Disconnected);
hFunc(TEvTabletPipe::TEvClientDestroyed, Handle);
hFunc(TEvTabletPipe::TEvClientConnected, Handle);
hFunc(TEvPrivate::TEvRetryNodeWhiteboard, Handle);
cFunc(TEvents::TSystem::Wakeup, HandleTimeout);
hFunc(TEvNodeWardenStorageConfig, Handle);
}
}

Expand Down Expand Up @@ -702,16 +672,10 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
RequestTabletPipe(ConsoleId, "TEvListTenantsRequest", request.Release());
}

void RequestSelectGroups(const TString& storagePoolName) {
THolder<TEvBlobStorage::TEvControllerSelectGroups> request = MakeHolder<TEvBlobStorage::TEvControllerSelectGroups>();
request->Record.SetReturnAllMatchingGroups(true);
request->Record.AddGroupParameters()->MutableStoragePoolSpecifier()->SetName(storagePoolName);
RequestTabletPipe(BsControllerId, "TEvControllerSelectGroups:" + storagePoolName, request.Release());
}

void RequestConfig() {
THolder<TEvBlobStorage::TEvControllerConfigRequest> request = MakeHolder<TEvBlobStorage::TEvControllerConfigRequest>();
request->Record.MutableRequest()->AddCommand()->MutableQueryBaseConfig();
request->Record.MutableRequest()->AddCommand()->MutableReadStoragePool()->SetBoxId(Max<ui64>());
RequestTabletPipe(BsControllerId, "TEvControllerConfigRequest", request.Release());
}

Expand Down Expand Up @@ -758,12 +722,6 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
}
}

void RequestStorageNode(TNodeId nodeId) {
if (StorageNodeIds.emplace(nodeId).second) {
RequestGenericNode(nodeId);
}
}

void Handle(TEvPrivate::TEvRetryNodeWhiteboard::TPtr& ev) {
switch (ev->Get()->EventId) {
case NNodeWhiteboard::TEvWhiteboard::EvSystemStateRequest:
Expand Down Expand Up @@ -851,44 +809,27 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
void Handle(TEvBlobStorage::TEvControllerConfigResponse::TPtr& ev) {
TabletRequests.CompleteRequest(ev->Cookie);
const NKikimrBlobStorage::TEvControllerConfigResponse& pbRecord(ev->Get()->Record);
if (pbRecord.HasResponse() && pbRecord.GetResponse().StatusSize() > 0) {
if (pbRecord.HasResponse() && pbRecord.GetResponse().StatusSize() > 1) {
const NKikimrBlobStorage::TConfigResponse::TStatus& pbStatus(pbRecord.GetResponse().GetStatus(0));
if (pbStatus.HasBaseConfig()) {
const NKikimrBlobStorage::TBaseConfig& pbConfig(pbStatus.GetBaseConfig());
for (const NKikimrBlobStorage::TBaseConfig::TPDisk& pDisk : pbConfig.GetPDisk()) {
RequestStorageNode(pDisk.GetNodeId());
}
BaseConfig = ev->Release();
}
}
RequestDone("TEvControllerConfigResponse");
}

void Handle(TEvBlobStorage::TEvControllerSelectGroupsResult::TPtr& ev) {
TabletRequests.CompleteRequest(ev->Cookie);
for (const auto& matchingGroups : ev->Get()->Record.matchinggroups()) {
for (const auto& group : matchingGroups.groups()) {
TString storagePoolName = group.storagepoolname();
StoragePoolState[storagePoolName].Groups.emplace(group.groupid());
StoragePoolState[storagePoolName].AuthenticGroups.emplace(group.groupid());
}
}
RequestDone("TEvControllerSelectGroupsResult");
}

void Handle(NSchemeShard::TEvSchemeShard::TEvDescribeSchemeResult::TPtr& ev) {
TabletRequests.CompleteRequest(ev->Cookie);
if (ev->Get()->GetRecord().status() == NKikimrScheme::StatusSuccess) {
TString path = ev->Get()->GetRecord().path();
TDatabaseState& state(DatabaseState[path]);
for (const auto& storagePool : ev->Get()->GetRecord().pathdescription().domaindescription().storagepools()) {
TString storagePoolName = storagePool.name();
state.StoragePoolNames.emplace(storagePoolName);
StoragePoolState[storagePoolName].Kind = storagePool.kind();
RequestSelectGroups(storagePoolName);
PathsByPoolName[storagePoolName].emplace(path); // no poolId in TEvDescribeSchemeResult, so it's neccesary to keep poolNames instead
}
if (path == DomainPath) {
state.StoragePoolNames.emplace(STATIC_STORAGE_POOL_NAME);
state.StoragePools.emplace(0); // static group has poolId = 0
StoragePoolState[0].Name = STATIC_STORAGE_POOL_NAME;
}
state.StorageUsage = ev->Get()->GetRecord().pathdescription().domaindescription().diskspaceusage().tables().totalsize();
state.StorageQuota = ev->Get()->GetRecord().pathdescription().domaindescription().databasequotas().data_size_hard_quota();
Expand Down Expand Up @@ -1106,14 +1047,26 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
}
for (const NKikimrBlobStorage::TBaseConfig::TGroup& group : pbConfig.GetGroup()) {
auto groupId = group.GetGroupId();
auto poolId = group.GetStoragePoolId();
ValidGroups.emplace(groupId);
BSConfigGroups.emplace(groupId, &group);
StoragePoolState[poolId].Groups.emplace(group.groupid());
}
for (const NKikimrBlobStorage::TBaseConfig::TNode& node : pbConfig.GetNode()) {
auto nodeId = node.GetNodeId();
BSConfigNodes.emplace(nodeId, &node);
}
}
const NKikimrBlobStorage::TConfigResponse::TStatus& spStatus(pbRecord.GetResponse().GetStatus(1));
for (const NKikimrBlobStorage::TDefineStoragePool& pool : spStatus.GetStoragePool()) { // there is no specific pool for static group here
ui64 poolId = pool.GetStoragePoolId();
TString storagePoolName = pool.GetName();
StoragePoolState[poolId].Name = storagePoolName;

for (const TString& path : PathsByPoolName[storagePoolName]) {
DatabaseState[path].StoragePools.emplace(poolId);
}
}
}
for (auto itPDisk = BSConfigPDisks.begin(); itPDisk != BSConfigPDisks.end();) {
if (ValidPDisks.count(itPDisk->first)) {
Expand Down Expand Up @@ -1380,10 +1333,6 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
return TStringBuilder() << vSlot.vslotid().nodeid() << "-" << vSlot.vslotid().pdiskid();
}

static TString GetPDiskId(const NKikimrBlobStorage::TNodeWardenServiceSet_TPDisk& pDisk) {
return TStringBuilder() << pDisk.nodeid() << "-" << pDisk.pdiskid();
}

void FillPDiskStatus(const TString& pDiskId, Ydb::Monitoring::StoragePDiskStatus& storagePDiskStatus, TSelfCheckContext context) {
context.Location.clear_database(); // PDisks are shared between databases
context.Location.mutable_storage()->mutable_pool()->clear_name(); // PDisks are shared between pools
Expand Down Expand Up @@ -1943,9 +1892,9 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
mergeContext.FillRecords(records);
}

void FillPoolStatus(const TString& poolName, const TStoragePoolState& pool, Ydb::Monitoring::StoragePoolStatus& storagePoolStatus, TSelfCheckContext context) {
context.Location.mutable_storage()->mutable_pool()->set_name(poolName);
storagePoolStatus.set_id(poolName);
void FillPoolStatus(const TStoragePoolState& pool, Ydb::Monitoring::StoragePoolStatus& storagePoolStatus, TSelfCheckContext context) {
context.Location.mutable_storage()->mutable_pool()->set_name(pool.Name);
storagePoolStatus.set_id(pool.Name);
for (auto groupId : pool.Groups) {
FillGroupStatus(groupId, *storagePoolStatus.add_groups(), {&context, "STORAGE_GROUP"});
}
Expand All @@ -1970,17 +1919,17 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
}

void FillStorage(TDatabaseState& databaseState, Ydb::Monitoring::StorageStatus& storageStatus, TSelfCheckContext context) {
if (databaseState.StoragePoolNames.empty()) {
if (!BaseConfig) {
// it will be reported RED BSController System Tablet in this case, we don't need report RED Storage
context.ReportStatus(Ydb::Monitoring::StatusFlag::GREEN, "System tablet BSC didn't provide information", ETags::StorageState);
} else if (databaseState.StoragePools.empty()) {
context.ReportStatus(Ydb::Monitoring::StatusFlag::RED, "There are no storage pools", ETags::StorageState);
} else {
for (const TString& poolName : databaseState.StoragePoolNames) {
auto itStoragePoolState = StoragePoolState.find(poolName);
for (const ui64 poolId : databaseState.StoragePools) {
auto itStoragePoolState = StoragePoolState.find(poolId);
if (itStoragePoolState != StoragePoolState.end()) {
if (!itStoragePoolState->second.AuthenticGroups.empty()) {
itStoragePoolState->second.Groups = itStoragePoolState->second.AuthenticGroups;
}
FillPoolStatus(poolName, itStoragePoolState->second, *storageStatus.add_pools(), {&context, "STORAGE_POOL"});
StoragePoolSeen.emplace(poolName);
FillPoolStatus(itStoragePoolState->second, *storageStatus.add_pools(), {&context, "STORAGE_POOL"});
StoragePoolSeen.emplace(poolId);
}
}
switch (context.GetOverallStatus()) {
Expand Down Expand Up @@ -2142,14 +2091,14 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
context.UpdateMaxStatus(tabletContext.GetOverallStatus());
}
}
if (!FilterDatabase) {
if (!FilterDatabase && BaseConfig) {
TDatabaseState unknownDatabase;
for (auto& [name, pool] : StoragePoolState) {
if (StoragePoolSeen.count(name) == 0) {
unknownDatabase.StoragePoolNames.insert(name);
for (auto& [id, pool] : StoragePoolState) {
if (StoragePoolSeen.count(id) == 0) {
unknownDatabase.StoragePools.insert(id);
}
}
if (!unknownDatabase.StoragePoolNames.empty()) {
if (!unknownDatabase.StoragePools.empty()) {
Ydb::Monitoring::DatabaseStatus& databaseStatus(*context.Result->add_database_status());
TSelfCheckResult storageContext;
FillStorage(unknownDatabase, *databaseStatus.mutable_storage(), {&storageContext, "STORAGE"});
Expand Down
Loading