Skip to content

added database filter for time difference - merge stable-24-2 #6229

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 3 commits into from
Jul 5, 2024
Merged

added database filter for time difference - merge stable-24-2 #6229

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
81d4eb3
YDB-1470 added database filter for time difference (#2217)
StekPerepolnen Mar 1, 2024
d684c1e
move time difference issue under database level (#5859)
StekPerepolnen Jul 3, 2024
34f2be1
remove peer info from public api
StekPerepolnen Jul 4, 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
6 changes: 4 additions & 2 deletions ydb/core/driver_lib/run/kikimr_services_initializers.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -712,8 +712,10 @@ void TBasicServicesInitializer::InitializeServices(NActors::TActorSystemSetup* s
data.Yellow ? NKikimrWhiteboard::EFlag::Yellow :
data.Orange ? NKikimrWhiteboard::EFlag::Orange :
data.Red ? NKikimrWhiteboard::EFlag::Red : NKikimrWhiteboard::EFlag()));
data.ActorSystem->Send(whiteboardId, new NNodeWhiteboard::TEvWhiteboard::TEvClockSkewUpdate(
data.PeerId, data.ClockSkew));
if (data.ReportClockSkew) {
data.ActorSystem->Send(whiteboardId, new NNodeWhiteboard::TEvWhiteboard::TEvClockSkewUpdate(
data.PeerId, data.ClockSkew));
}
};
}

Expand Down
81 changes: 43 additions & 38 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 @@ -235,6 +235,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
ui64 StorageQuota;
ui64 StorageUsage;
TMaybeServerlessComputeResourcesMode ServerlessComputeResourcesMode;
TNodeId MaxTimeDifferenceNodeId = 0;
};

struct TSelfCheckResult {
Expand Down Expand Up @@ -813,10 +814,10 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
ReplyAndPassAway();
}

bool IsStaticNode(const TEvInterconnect::TNodeInfo& nodeInfo) const {
bool IsStaticNode(const TNodeId nodeId) const {
TAppData* appData = AppData();
if (appData->DynamicNameserviceConfig) {
return nodeInfo.NodeId <= AppData()->DynamicNameserviceConfig->MaxStaticNodeId;
return nodeId <= AppData()->DynamicNameserviceConfig->MaxStaticNodeId;
} else {
return true;
}
Expand All @@ -827,7 +828,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
NodesInfo = ev->Release();
for (const auto& ni : NodesInfo->Nodes) {
MergedNodeInfo[ni.NodeId] = &ni;
if (IsStaticNode(ni) && needComputeFromStaticNodes) {
if (IsStaticNode(ni.NodeId) && needComputeFromStaticNodes) {
DatabaseState[DomainPath].ComputeNodeIds.push_back(ni.NodeId);
RequestComputeNode(ni.NodeId);
}
Expand Down Expand Up @@ -1251,7 +1252,7 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
}
}

void FillComputeNodeStatus(TDatabaseState& databaseState,TNodeId nodeId, Ydb::Monitoring::ComputeNodeStatus& computeNodeStatus, TSelfCheckContext context) {
void FillComputeNodeStatus(TDatabaseState& databaseState, TNodeId nodeId, Ydb::Monitoring::ComputeNodeStatus& computeNodeStatus, TSelfCheckContext context) {
FillNodeInfo(nodeId, context.Location.mutable_compute()->mutable_node());

TSelfCheckContext rrContext(&context, "NODE_UPTIME");
Expand Down Expand Up @@ -1289,6 +1290,32 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
}
loadAverageStatus.set_overall(laContext.GetOverallStatus());
}

if (nodeSystemState.HasMaxClockSkewPeerId()) {
TNodeId peerId = nodeSystemState.GetMaxClockSkewPeerId();
long timeDifferenceUs = nodeSystemState.GetMaxClockSkewWithPeerUs();
TDuration timeDifferenceDuration = TDuration::MicroSeconds(abs(timeDifferenceUs));
Ydb::Monitoring::StatusFlag::Status status;
if (timeDifferenceDuration > MAX_CLOCKSKEW_ORANGE_ISSUE_TIME) {
status = Ydb::Monitoring::StatusFlag::ORANGE;
} else if (timeDifferenceDuration > MAX_CLOCKSKEW_YELLOW_ISSUE_TIME) {
status = Ydb::Monitoring::StatusFlag::YELLOW;
} else {
status = Ydb::Monitoring::StatusFlag::GREEN;
}

if (databaseState.MaxTimeDifferenceNodeId == nodeId) {
TSelfCheckContext tdContext(&context, "NODES_TIME_DIFFERENCE");
if (status == Ydb::Monitoring::StatusFlag::GREEN) {
tdContext.ReportStatus(status);
} else {
tdContext.ReportStatus(status, TStringBuilder() << "Node is "
<< timeDifferenceDuration.MilliSeconds() << " ms "
<< (timeDifferenceUs > 0 ? "behind " : "ahead of ")
<< "peer [" << peerId << "]", ETags::SyncState);
}
}
}
} else {
// context.ReportStatus(Ydb::Monitoring::StatusFlag::RED,
// TStringBuilder() << "Compute node is not available",
Expand Down Expand Up @@ -1320,12 +1347,24 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
if (systemStatus != Ydb::Monitoring::StatusFlag::GREEN && systemStatus != Ydb::Monitoring::StatusFlag::GREY) {
context.ReportStatus(systemStatus, "Compute has issues with system tablets", ETags::ComputeState, {ETags::SystemTabletState});
}
long maxTimeDifferenceUs = 0;
for (TNodeId nodeId : *computeNodeIds) {
auto itNodeSystemState = MergedNodeSystemState.find(nodeId);
if (itNodeSystemState != MergedNodeSystemState.end()) {
if (std::count(computeNodeIds->begin(), computeNodeIds->end(), itNodeSystemState->second->GetMaxClockSkewPeerId()) > 0
&& abs(itNodeSystemState->second->GetMaxClockSkewWithPeerUs()) > maxTimeDifferenceUs) {
maxTimeDifferenceUs = abs(itNodeSystemState->second->GetMaxClockSkewWithPeerUs());
databaseState.MaxTimeDifferenceNodeId = nodeId;
}
}
}
for (TNodeId nodeId : *computeNodeIds) {
auto& computeNode = *computeStatus.add_nodes();
FillComputeNodeStatus(databaseState, nodeId, computeNode, {&context, "COMPUTE_NODE"});
}
context.ReportWithMaxChildStatus("Some nodes are restarting too often", ETags::ComputeState, {ETags::Uptime});
context.ReportWithMaxChildStatus("Compute is overloaded", ETags::ComputeState, {ETags::OverloadState});
context.ReportWithMaxChildStatus("Database has time difference between nodes", ETags::ComputeState, {ETags::SyncState});
Ydb::Monitoring::StatusFlag::Status tabletsStatus = Ydb::Monitoring::StatusFlag::GREEN;
computeNodeIds->push_back(0); // for tablets without node
for (TNodeId nodeId : *computeNodeIds) {
Expand Down Expand Up @@ -2072,39 +2111,6 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
const TDuration MAX_CLOCKSKEW_ORANGE_ISSUE_TIME = TDuration::MicroSeconds(25000);
const TDuration MAX_CLOCKSKEW_YELLOW_ISSUE_TIME = TDuration::MicroSeconds(5000);

void FillNodesSyncStatus(TOverallStateContext& context) {
long maxClockSkewUs = 0;
TNodeId maxClockSkewPeerId = 0;
TNodeId maxClockSkewNodeId = 0;
for (auto& [nodeId, nodeSystemState] : MergedNodeSystemState) {
if (abs(nodeSystemState->GetMaxClockSkewWithPeerUs()) > maxClockSkewUs) {
maxClockSkewUs = abs(nodeSystemState->GetMaxClockSkewWithPeerUs());
maxClockSkewPeerId = nodeSystemState->GetMaxClockSkewPeerId();
maxClockSkewNodeId = nodeId;
}
}
if (!maxClockSkewNodeId) {
return;
}

TSelfCheckResult syncContext;
syncContext.Type = "NODES_TIME_DIFFERENCE";
FillNodeInfo(maxClockSkewNodeId, syncContext.Location.mutable_node());
FillNodeInfo(maxClockSkewPeerId, syncContext.Location.mutable_peer());

TDuration maxClockSkewTime = TDuration::MicroSeconds(maxClockSkewUs);
if (maxClockSkewTime > MAX_CLOCKSKEW_ORANGE_ISSUE_TIME) {
syncContext.ReportStatus(Ydb::Monitoring::StatusFlag::ORANGE, TStringBuilder() << "The nodes have a time difference of " << maxClockSkewTime.MilliSeconds() << " ms", ETags::SyncState);
} else if (maxClockSkewTime > MAX_CLOCKSKEW_YELLOW_ISSUE_TIME) {
syncContext.ReportStatus(Ydb::Monitoring::StatusFlag::YELLOW, TStringBuilder() << "The nodes have a time difference of " << maxClockSkewTime.MilliSeconds() << " ms", ETags::SyncState);
} else {
syncContext.ReportStatus(Ydb::Monitoring::StatusFlag::GREEN);
}

context.UpdateMaxStatus(syncContext.GetOverallStatus());
context.AddIssues(syncContext.IssueRecords);
}

void FillResult(TOverallStateContext context) {
if (IsSpecificDatabaseFilter()) {
FillDatabaseResult(context, FilterDatabase, DatabaseState[FilterDatabase]);
Expand All @@ -2113,7 +2119,6 @@ class TSelfCheckRequest : public TActorBootstrapped<TSelfCheckRequest> {
FillDatabaseResult(context, path, state);
}
}
FillNodesSyncStatus(context);
if (DatabaseState.empty()) {
Ydb::Monitoring::DatabaseStatus& databaseStatus(*context.Result->add_database_status());
TSelfCheckResult tabletContext;
Expand Down
5 changes: 4 additions & 1 deletion ydb/library/actors/interconnect/interconnect_common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,8 +74,10 @@ namespace NActors {
bool Orange;
bool Red;
i64 ClockSkew;
bool ReportClockSkew;

TWhiteboardSessionStatus(TActorSystem* actorSystem, ui32 peerId, const TString& peer, bool connected, bool green, bool yellow, bool orange, bool red, i64 clockSkew)
TWhiteboardSessionStatus(TActorSystem* actorSystem, ui32 peerId, const TString& peer, bool connected,
bool green, bool yellow, bool orange, bool red, i64 clockSkew, bool reportClockSkew)
: ActorSystem(actorSystem)
, PeerId(peerId)
, Peer(peer)
Expand All @@ -85,6 +87,7 @@ namespace NActors {
, Orange(orange)
, Red(red)
, ClockSkew(clockSkew)
, ReportClockSkew(reportClockSkew)
{}
};

Expand Down
7 changes: 6 additions & 1 deletion ydb/library/actors/interconnect/interconnect_tcp_session.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -999,6 +999,10 @@ namespace NActors {
} while (false);
}

// we need track clockskew only if it's one tenant nodes connection
// they have one scope in this case
bool reportClockSkew = Proxy->Common->LocalScopeId.first != 0 && Proxy->Common->LocalScopeId == Params.PeerScopeId;

callback({TlsActivationContext->ExecutorThread.ActorSystem,
Proxy->PeerNodeId,
Proxy->Metrics->GetHumanFriendlyPeerHostName(),
Expand All @@ -1007,7 +1011,8 @@ namespace NActors {
flagState == EFlag::YELLOW,
flagState == EFlag::ORANGE,
flagState == EFlag::RED,
ReceiveContext->ClockSkew_us.load()});
ReceiveContext->ClockSkew_us.load(),
reportClockSkew});
}

if (connected) {
Expand Down
Loading