vdisk evict (#5093)

ydb/core/viewer/json_handlers_vdisk.cpp

@@ -6,14 +6,15 @@
 #include "json_vdiskstat.h"
 #include "json_getblob.h"
 #include "json_blobindexstat.h"
-
+#include "json_vdisk_evict.h"
 
 namespace NKikimr::NViewer {
 
 void InitVDiskJsonHandlers(TJsonHandlers& jsonHandlers) {
     jsonHandlers.AddHandler("/vdisk/vdiskstat", new TJsonHandler<TJsonVDiskStat>);
     jsonHandlers.AddHandler("/vdisk/getblob", new TJsonHandler<TJsonGetBlob>);
     jsonHandlers.AddHandler("/vdisk/blobindexstat", new TJsonHandler<TJsonBlobIndexStat>);
+    jsonHandlers.AddHandler("/vdisk/evict", new TJsonHandler<TJsonVDiskEvict>);
 }
 
 }

ydb/core/viewer/json_pipe_req.h

@@ -174,6 +174,21 @@ class TViewerPipeClient : public TActorBootstrapped<TDerived> {
         SendRequestToPipe(pipeClient, request.Release());
     }
 
+    void RequestBSControllerVDiskEvict(ui32 groupId, ui32 groupGeneration, ui32 failRealmIdx, ui32 failDomainIdx, ui32 vdiskIdx, bool force = false) {
+        TActorId pipeClient = ConnectTabletPipe(GetBSControllerId());
+        THolder<TEvBlobStorage::TEvControllerConfigRequest> request = MakeHolder<TEvBlobStorage::TEvControllerConfigRequest>();
+        auto* evictVDisk = request->Record.MutableRequest()->AddCommand()->MutableReassignGroupDisk();
+        evictVDisk->SetGroupId(groupId);
+        evictVDisk->SetGroupGeneration(groupGeneration);
+        evictVDisk->SetFailRealmIdx(failRealmIdx);
+        evictVDisk->SetFailDomainIdx(failDomainIdx);
+        evictVDisk->SetVDiskIdx(vdiskIdx);
+        if (force) {
+            request->Record.MutableRequest()->SetIgnoreDegradedGroupsChecks(true);
+        }
+        SendRequestToPipe(pipeClient, request.Release());
+    }
+
     void RequestSchemeCacheNavigate(const TString& path) {
         THolder<NSchemeCache::TSchemeCacheNavigate> request = MakeHolder<NSchemeCache::TSchemeCacheNavigate>();
         NSchemeCache::TSchemeCacheNavigate::TEntry entry;

ydb/core/viewer/json_vdisk_evict.h

@@ -0,0 +1,290 @@
+#pragma once
+#include <ydb/library/actors/core/actor_bootstrapped.h>
+#include <ydb/library/actors/core/interconnect.h>
+#include <ydb/library/actors/core/mon.h>
+#include <ydb/library/services/services.pb.h>
+#include <ydb/core/viewer/json/json.h>
+#include <ydb/core/util/wildcard.h>
+#include <library/cpp/json/json_writer.h>
+#include "viewer.h"
+#include "json_pipe_req.h"
+
+namespace NKikimr {
+namespace NViewer {
+
+using namespace NActors;
+
+class TJsonVDiskEvict : public TViewerPipeClient<TJsonVDiskEvict> {
+    enum EEv {
+        EvRetryNodeRequest = EventSpaceBegin(NActors::TEvents::ES_PRIVATE),
+        EvEnd
+    };
+
+    static_assert(EvEnd < EventSpaceEnd(NActors::TEvents::ES_PRIVATE), "expect EvEnd < EventSpaceEnd(TEvents::ES_PRIVATE)");
+
+    struct TEvRetryNodeRequest : NActors::TEventLocal<TEvRetryNodeRequest, EvRetryNodeRequest> {
+        TEvRetryNodeRequest()
+        {}
+    };
+
+protected:
+    using TThis = TJsonVDiskEvict;
+    using TBase = TViewerPipeClient<TThis>;
+    IViewer* Viewer;
+    NMon::TEvHttpInfo::TPtr Event;
+    TJsonSettings JsonSettings;
+    ui32 Timeout = 0;
+    ui32 ActualRetries = 0;
+    ui32 Retries = 0;
+    TDuration RetryPeriod = TDuration::MilliSeconds(500);
+
+    std::unique_ptr<TEvBlobStorage::TEvControllerConfigResponse> Response;
+
+    ui32 GroupId = 0;
+    ui32 GroupGeneration = 0;
+    ui32 FailRealmIdx = 0;
+    ui32 FailDomainIdx = 0;
+    ui32 VdiskIdx = 0;
+    bool Force = false;
+
+public:
+    static constexpr NKikimrServices::TActivity::EType ActorActivityType() {
+        return NKikimrServices::TActivity::VIEWER_HANDLER;
+    }
+
+    TJsonVDiskEvict(IViewer* viewer, NMon::TEvHttpInfo::TPtr& ev)
+        : Viewer(viewer)
+        , Event(ev)
+    {}
+
+    inline ui32 GetRequiredParam(const TCgiParameters& params, const std::string& name, ui32& obj) {
+        if (!TryFromString<ui32>(params.Get(name), obj)) {
+            TBase::Send(Event->Sender, new NMon::TEvHttpInfoRes(
+                Viewer->GetHTTPBADREQUEST(Event->Get(), "text/plain", TStringBuilder() << "field '" << name << "' or 'vdisk_id' are required"),
+                0, NMon::IEvHttpInfoRes::EContentType::Custom));
+            return false;
+        }
+        return true;
+    }
+
+    void Bootstrap() {
+        const auto& params(Event->Get()->Request.GetParams());
+        TString vdisk_id = params.Get("vdisk_id");
+        if (vdisk_id) {
+            TVector<TString> parts = StringSplitter(vdisk_id).Split('-').SkipEmpty();
+            if (parts.size() == 5) {
+                GroupId = FromStringWithDefault<ui32>(parts[0], Max<ui32>());
+                GroupGeneration = FromStringWithDefault<ui32>(parts[1], Max<ui32>());
+                FailRealmIdx = FromStringWithDefault<ui32>(parts[2], Max<ui32>());
+                FailDomainIdx = FromStringWithDefault<ui32>(parts[3], Max<ui32>());
+                VdiskIdx = FromStringWithDefault<ui32>(parts[4], Max<ui32>());
+            }
+            if (parts.size() != 5 || GroupId == Max<ui32>()
+                    || GroupGeneration == Max<ui32>() || FailRealmIdx  == Max<ui32>()
+                    || FailDomainIdx == Max<ui32>() || VdiskIdx == Max<ui32>()) {
+                TBase::Send(Event->Sender, new NMon::TEvHttpInfoRes(
+                    Viewer->GetHTTPBADREQUEST(Event->Get(), "text/plain", TStringBuilder() << "Unable to parse the 'vdisk_id' parameter"),
+                    0, NMon::IEvHttpInfoRes::EContentType::Custom));
+                return PassAway();
+            }
+        } else if (!GetRequiredParam(params, "group_id", GroupId)
+                || !GetRequiredParam(params, "group_generation_id", GroupGeneration)
+                || !GetRequiredParam(params, "fail_realm_idx", FailRealmIdx)
+                || !GetRequiredParam(params, "fail_domain_idx", FailDomainIdx)
+                || !GetRequiredParam(params, "vdisk_idx", VdiskIdx)) {
+            return PassAway();
+        }
+
+        if (Event->Get()->Request.GetMethod() != HTTP_METHOD_POST) {
+            TBase::Send(Event->Sender, new NMon::TEvHttpInfoRes(
+                Viewer->GetHTTPBADREQUEST(Event->Get(), "text/plain", "Only POST method is allowed"),
+                0, NMon::IEvHttpInfoRes::EContentType::Custom));
+            return PassAway();
+        }
+        TBase::InitConfig(params);
+
+        Force = FromStringWithDefault<bool>(params.Get("force"), false);
+        Timeout = FromStringWithDefault<ui32>(params.Get("timeout"), 10000);
+        Retries = FromStringWithDefault<ui32>(params.Get("retries"), 0);
+        RetryPeriod = TDuration::MilliSeconds(FromStringWithDefault<ui32>(params.Get("retry_period"), RetryPeriod.MilliSeconds()));
+
+        SendRequest();
+
+        TBase::Become(&TThis::StateWork, TDuration::MilliSeconds(Timeout), new TEvents::TEvWakeup());
+    }
+
+    STATEFN(StateWork) {
+        switch (ev->GetTypeRewrite()) {
+            hFunc(TEvBlobStorage::TEvControllerConfigResponse, Handle);
+            cFunc(TEvRetryNodeRequest::EventType, HandleRetry);
+            cFunc(TEvents::TEvUndelivered::EventType, Undelivered);
+            cFunc(TEvents::TSystem::Wakeup, HandleTimeout);
+        }
+    }
+
+    void SendRequest() {
+        RequestBSControllerVDiskEvict(GroupId, GroupGeneration, FailRealmIdx, FailDomainIdx, VdiskIdx, Force);
+    }
+
+    bool RetryRequest() {
+        if (Retries) {
+            if (++ActualRetries <= Retries) {
+                TBase::Schedule(RetryPeriod, new TEvRetryNodeRequest());
+                return true;
+            }
+        }
+        return false;
+    }
+
+    void Undelivered() {
+        if (!RetryRequest()) {
+            TBase::RequestDone();
+        }
+    }
+
+    void Handle(TEvBlobStorage::TEvControllerConfigResponse::TPtr& ev) {
+        Response.reset(ev->Release().Release());
+        ReplyAndPassAway();
+    }
+
+    void HandleRetry() {
+        SendRequest();
+    }
+
+    void HandleTimeout() {
+        Send(Event->Sender, new NMon::TEvHttpInfoRes(
+            Viewer->GetHTTPGATEWAYTIMEOUT(Event->Get(), "text/plain", "Timeout receiving response from NodeWarden"),
+            0, NMon::IEvHttpInfoRes::EContentType::Custom));
+    }
+
+    void PassAway() override {
+        TBase::PassAway();
+    }
+
+    void TryToTranslateFromBSC2Human(const NKikimrBlobStorage::TConfigResponse& response, TString& bscError, bool& forceRetryPossible) {
+        if (response.GroupsGetDisintegratedByExpectedStatusSize()) {
+            bscError = TStringBuilder() << "Calling this operation will cause at least groups [" << JoinStrings(response.GetGroupsGetDisintegratedByExpectedStatus().begin(), response.GetGroupsGetDisintegratedByExpectedStatus().end(), ", ") << "] to go into a dead state";
+        } else if (response.GroupsGetDisintegratedSize()) {
+            bscError = TStringBuilder() << "Calling this operation will cause at least groups [" << JoinStrings(response.GetGroupsGetDisintegrated().begin(), response.GetGroupsGetDisintegrated().end(), ", ") << "] to go into a dead state";
+        } else if (response.GroupsGetDegradedSize()) {
+            bscError = TStringBuilder() << "Calling this operation will cause at least groups [" << JoinStrings(response.GetGroupsGetDegraded().begin(), response.GetGroupsGetDegraded().end(), ", ") << "] to go into a degraded state";
+            forceRetryPossible = true;
+        } else if (response.StatusSize()) {
+            const auto& lastStatus = response.GetStatus(response.StatusSize() - 1);
+            TVector<ui32> groups;
+            for (auto& failParam: lastStatus.GetFailParam()) {
+                if (failParam.HasGroupId()) {
+                    groups.emplace_back(failParam.GetGroupId());
+                }
+            }
+            if (lastStatus.GetFailReason() == NKikimrBlobStorage::TConfigResponse::TStatus::kMayGetDegraded) {
+                bscError = TStringBuilder() << "Calling this operation will cause at least groups [" << JoinVectorIntoString(groups, ", ") << "] to go into a degraded state";
+                forceRetryPossible = true;
+            } else if (lastStatus.GetFailReason() == NKikimrBlobStorage::TConfigResponse::TStatus::kMayLoseData) {
+                bscError = TStringBuilder() << "Calling this operation may result in data loss for at least groups [" << JoinVectorIntoString(groups, ", ") << "]";
+            }
+        }
+    }
+
+    void ReplyAndPassAway() {
+        NJson::TJsonValue json;
+        if (Response != nullptr) {
+            if (Response->Record.GetResponse().GetSuccess()) {
+                json["result"] = true;
+            } else {
+                json["result"] = false;
+                TString error = Response->Record.GetResponse().GetErrorDescription();
+                bool forceRetryPossible = false;
+                TryToTranslateFromBSC2Human(Response->Record.GetResponse(), error, forceRetryPossible);
+                json["error"] = error;
+                if (forceRetryPossible) {
+                    json["forceRetryPossible"] = true;
+                }
+            }
+            json["debugMessage"] = Response->Record.ShortDebugString();
+        } else {
+            json["result"] = false;
+            json["error"] = "No response was received from BSC";
+        }
+        TBase::Send(Event->Sender, new NMon::TEvHttpInfoRes(Viewer->GetHTTPOKJSON(Event->Get(), NJson::WriteJson(json)), 0, NMon::IEvHttpInfoRes::EContentType::Custom));
+        PassAway();
+    }
+};
+
+template <>
+YAML::Node TJsonRequestSwagger<TJsonVDiskEvict>::GetSwagger() {
+    return YAML::Load(R"___(
+        post:
+          tags:
+          - vdisk
+          summary: VDisk evict
+          description: VDisk evict
+          parameters:
+          - name: vdisk_id
+            in: query
+            description: vdisk identifier
+            required: false
+            type: string
+          - name: group_id
+            in: query
+            description: group identifier
+            required: false
+            type: integer
+          - name: group_generation_id
+            in: query
+            description: group generation identifier
+            required: false
+            type: integer
+          - name: fail_realm_idx
+            in: query
+            description: fail realm identifier
+            required: false
+            type: integer
+          - name: fail_domain_ids
+            in: query
+            description: fail domain identifier
+            required: false
+            type: integer
+          - name: vdisk_idx
+            in: query
+            description: vdisk idx identifier
+            required: false
+            type: integer
+          - name: timeout
+            in: query
+            description: timeout in ms
+            required: false
+            type: integer
+          - name: force
+            in: query
+            description: attempt forced operation, ignore warnings
+            required: false
+            type: boolean
+          responses:
+            200:
+              description: OK
+              content:
+                application/json:
+                  schema:
+                    type: object
+                    properties:
+                      result:
+                        type: boolean
+                        description: was operation successful or not
+                      error:
+                        type: string
+                        description: details about failed operation
+                      forceRetryPossible:
+                        type: boolean
+                        description: if true, operation can be retried with force flag
+            400:
+              description: Bad Request
+            403:
+              description: Forbidden
+            504:
+              description: Gateway Timeout
+        )___");
+}
+
+}
+}

ydb/core/viewer/ya.make

@@ -48,6 +48,7 @@ SRCS(
     json_topicinfo.h
     json_pqconsumerinfo.h
     json_vdisk_req.h
+    json_vdisk_evict.h
     json_vdiskinfo.h
     json_vdiskstat.h
     json_wb_req.h