storage: prevent unbounded raft log growth without quorum

Fixes cockroachdb#27772.

This change adds safeguards to prevent cases where a raft log
would grow without bound during loss of quorum scenarios. It
also adds a new test that demonstrates that the raft log does
not grow without bound in these cases.

There are two cases that need to be handled to prevent the
unbounded raft log growth observed in cockroachdb#27772.
1. When the leader proposes a command and cannot establish a
   quorum. In this case, we know the leader has the entry in
   its log, so there's no need to refresh it with `reasonTicks`.
   To avoid this, we no longer use `refreshTicks` as a leader.
2. When a follower proposes a command that is forwarded to the
   leader who cannot establish a quorum. In this case, the
   follower can't be sure (currently) that the leader got the
   proposal, so it needs to refresh using `reasonTicks`. However,
   the leader now detects duplicate forwarded proposals and
   avoids appending redundant entries to its log. It does so
   by maintaining a set of in-flight forwarded proposals that
   it has received during its term as leader. This set is reset
   after every leadership change.

Both of these cases are tested against in the new
TestLogGrowthWhenRefreshingPendingCommands. Without both of
the safeguards introduced in this commit, the test fails.

Release note (bug fix): Prevent unbounded growth of the raft log
caused by a loss of quorum.

pkg/storage/client_raft_test.go

@@ -40,6 +40,7 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/storage/storagebase"
 	"github.com/cockroachdb/cockroach/pkg/testutils"
 	"github.com/cockroachdb/cockroach/pkg/util/hlc"
+	"github.com/cockroachdb/cockroach/pkg/util/humanizeutil"
 	"github.com/cockroachdb/cockroach/pkg/util/leaktest"
 	"github.com/cockroachdb/cockroach/pkg/util/log"
 	"github.com/cockroachdb/cockroach/pkg/util/protoutil"
@@ -1124,6 +1125,144 @@ func TestRefreshPendingCommands(t *testing.T) {
 	}
 }
 
+// Test that when a Raft group is not able to establish a quorum, its Raft log
+// does not grow without bound. It tests two different scenerios where this used
+// to be possible (see #27772):
+// 1. The leader proposes a command and cannot establish a quorum. The leader
+//    continually re-proposes the command.
+// 2. The follower proposes a command and forwards it to the leader, who cannot
+//    establish a quorum. The follower continually re-proposes and forwards the
+//    command to the leader.
+func TestLogGrowthWhenRefreshingPendingCommands(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+
+	sc := storage.TestStoreConfig(nil)
+	// Drop the raft tick interval so the Raft group is ticked more.
+	sc.RaftTickInterval = 10 * time.Millisecond
+	// Don't timeout raft leader. We don't want leadership moving.
+	sc.RaftElectionTimeoutTicks = 1000000
+	// Disable leader transfers during leaseholder changes so that we
+	// can easily create leader-not-leaseholder scenarios.
+	sc.TestingKnobs.DisableLeaderFollowsLeaseholder = true
+	// Refresh pending commands on every Raft group tick instead of
+	// every RaftElectionTimeoutTicks.
+	sc.TestingKnobs.RefreshReasonTicksPeriod = 1
+	// Disable periodic gossip tasks which can move the range 1 lease
+	// unexpectedly.
+	sc.TestingKnobs.DisablePeriodicGossips = true
+	mtc := &multiTestContext{storeConfig: &sc}
+	defer mtc.Stop()
+	mtc.Start(t, 5)
+
+	const rangeID = roachpb.RangeID(1)
+	mtc.replicateRange(rangeID, 1, 2, 3, 4)
+
+	// Raft leadership is kept on node 0.
+	leaderRepl, err := mtc.Store(0).GetReplica(rangeID)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Put some data in the range so we'll have something to test for.
+	incArgs := incrementArgs([]byte("a"), 5)
+	if _, err := client.SendWrapped(context.Background(), mtc.stores[0].TestSender(), incArgs); err != nil {
+		t.Fatal(err)
+	}
+
+	// Wait for all nodes to catch up.
+	mtc.waitForValues(roachpb.Key("a"), []int64{5, 5, 5, 5, 5})
+
+	// Test proposing on leader and proposing on follower. Neither should result
+	// in unbounded raft log growth.
+	testutils.RunTrueAndFalse(t, "proposeOnFollower", func(t *testing.T, proposeOnFollower bool) {
+		// Restart any nodes that are down.
+		for _, s := range []int{2, 3, 4} {
+			if mtc.Store(s) == nil {
+				mtc.restartStore(s)
+			}
+		}
+
+		// Determine which node to propose on. Transfer lease to that node.
+		var propIdx, otherIdx int
+		if !proposeOnFollower {
+			propIdx, otherIdx = 0, 1
+		} else {
+			propIdx, otherIdx = 1, 0
+		}
+		propNode := mtc.stores[propIdx].TestSender()
+		mtc.transferLease(context.TODO(), rangeID, otherIdx, propIdx)
+		testutils.SucceedsSoon(t, func() error {
+			// Lease transfers may not be immediately observed by the new
+			// leaseholder. Wait until the new leaseholder is aware.
+			repl, err := mtc.Store(propIdx).GetReplica(rangeID)
+			if err != nil {
+				t.Fatal(err)
+			}
+			repDesc, err := repl.GetReplicaDescriptor()
+			if err != nil {
+				t.Fatal(err)
+			}
+			if lease, _ := repl.GetLease(); lease.Replica != repDesc {
+				return errors.Errorf("lease not transferred yet; found %v", lease)
+			}
+			return nil
+		})
+
+		// Stop enough nodes to prevent a quorum.
+		for _, s := range []int{2, 3, 4} {
+			mtc.stopStore(s)
+		}
+
+		// Determine the current raft log size.
+		initLogSize := leaderRepl.GetRaftLogSize()
+
+		// While a majority nodes are down, write some data.
+		putRes := make(chan *roachpb.Error)
+		go func() {
+			putArgs := putArgs([]byte("b"), make([]byte, 8<<10 /* 8 KB */))
+			_, err := client.SendWrapped(context.Background(), propNode, putArgs)
+			putRes <- err
+		}()
+
+		// Wait for a bit and watch for Raft log growth.
+		wait := time.After(500 * time.Millisecond)
+		ticker := time.Tick(50 * time.Millisecond)
+	Loop:
+		for {
+			select {
+			case <-wait:
+				break Loop
+			case <-ticker:
+				// Verify that the leader is node 0.
+				status := leaderRepl.RaftStatus()
+				if status == nil || status.RaftState != raft.StateLeader {
+					t.Fatalf("raft leader should be node 0, but got status %+v", status)
+				}
+
+				// Check raft log size.
+				const logSizeLimit = 64 << 10 // 64 KB
+				curlogSize := leaderRepl.GetRaftLogSize()
+				logSize := curlogSize - initLogSize
+				logSizeStr := humanizeutil.IBytes(logSize)
+				if logSize > logSizeLimit {
+					t.Fatalf("raft log size grew to %s", logSizeStr)
+				}
+				t.Logf("raft log size grew to %s", logSizeStr)
+			case err := <-putRes:
+				t.Fatalf("write finished with quorum unavailable; err=%v", err)
+			}
+		}
+
+		// Start enough nodes to establish a quorum.
+		mtc.restartStore(2)
+
+		// The write should now succeed.
+		if err := <-putRes; err != nil {
+			t.Fatal(err)
+		}
+	})
+}
+
 // TestStoreRangeUpReplicate verifies that the replication queue will notice
 // under-replicated ranges and replicate them. Also tests that preemptive
 // snapshots which contain sideloaded proposals don't panic the receiving end.

pkg/storage/replica.go

@@ -392,7 +392,12 @@ type Replica struct {
 		// map must only be referenced while Replica.mu is held, except if the
 		// element is removed from the map first. The notable exception is the
 		// contained RaftCommand, which we treat as immutable.
-		proposals         map[storagebase.CmdIDKey]*ProposalData
+		proposals map[storagebase.CmdIDKey]*ProposalData
+		// remoteProposals is maintained by Raft leaders and stores in-flight
+		// commands that were forwarded to the leader during its current term.
+		// The set allows leaders to detect duplicate forwarded commands and
+		// avoid re-proposing the same forwarded command multiple times.
+		remoteProposals   map[storagebase.CmdIDKey]struct{}
 		internalRaftGroup *raft.RawNode
 		// The ID of the replica within the Raft group. May be 0 if the replica has
 		// been created from a preemptive snapshot (i.e. before being added to the
@@ -852,6 +857,7 @@ func (r *Replica) cancelPendingCommandsLocked() {
 		p.finishRaftApplication(resp)
 	}
 	r.mu.proposals = map[storagebase.CmdIDKey]*ProposalData{}
+	r.mu.remoteProposals = nil
 }
 
 // setTombstoneKey writes a tombstone to disk to ensure that replica IDs never
@@ -3333,6 +3339,39 @@ func (r *Replica) stepRaftGroup(req *RaftMessageRequest) error {
 		// other replica is not quiesced, so we don't need to wake the leader.
 		r.unquiesceLocked()
 		r.refreshLastUpdateTimeForReplicaLocked(req.FromReplica.ReplicaID)
+		if req.Message.Type == raftpb.MsgProp {
+			// A proposal was forwarded to this replica.
+			if r.mu.replicaID == r.mu.leaderID {
+				// This replica is the leader. Record that the proposal
+				// was seen and drop the proposal if it was already seen.
+				// This prevents duplicate forwarded proposals from each
+				// being appended to a leader's raft log.
+				allSeen := true
+				for _, e := range req.Message.Entries {
+					switch e.Type {
+					case raftpb.EntryNormal:
+						cmdID, _ := DecodeRaftCommand(e.Data)
+						if r.mu.remoteProposals == nil {
+							r.mu.remoteProposals = map[storagebase.CmdIDKey]struct{}{}
+						}
+						if _, ok := r.mu.remoteProposals[cmdID]; !ok {
+							r.mu.remoteProposals[cmdID] = struct{}{}
+							allSeen = false
+						}
+					case raftpb.EntryConfChange:
+						// We could peek into the EntryConfChange to find the
+						// command ID, but we don't expect follower-initiated
+						// conf changes.
+						allSeen = false
+					default:
+						log.Fatalf(context.TODO(), "unexpected Raft entry: %v", e)
+					}
+				}
+				if allSeen {
+					return false /* unquiesceAndWakeLeader */, nil
+				}
+			}
+		}
 		err := raftGroup.Step(req.Message)
 		if err == raft.ErrProposalDropped {
 			// A proposal was forwarded to this replica but we couldn't propose it.
@@ -3441,6 +3480,9 @@ func (r *Replica) handleRaftReadyRaftMuLocked(
 		if !r.store.TestingKnobs().DisableRefreshReasonNewLeader {
 			refreshReason = reasonNewLeader
 		}
+		// Clear the remote proposal set. Would have been nil already if not
+		// previously the leader.
+		r.mu.remoteProposals = nil
 		leaderID = roachpb.ReplicaID(rd.SoftState.Lead)
 	}
 
@@ -3831,17 +3873,28 @@ func (r *Replica) tick() (bool, error) {
 
 	r.mu.ticks++
 	r.mu.internalRaftGroup.Tick()
+
+	refreshAtDelta := r.store.cfg.RaftElectionTimeoutTicks
+	if knob := r.store.TestingKnobs().RefreshReasonTicksPeriod; knob > 0 {
+		refreshAtDelta = knob
+	}
 	if !r.store.TestingKnobs().DisableRefreshReasonTicks &&
-		r.mu.ticks%r.store.cfg.RaftElectionTimeoutTicks == 0 {
+		r.mu.replicaID != r.mu.leaderID &&
+		r.mu.ticks%refreshAtDelta == 0 {
 		// RaftElectionTimeoutTicks is a reasonable approximation of how long we
 		// should wait before deciding that our previous proposal didn't go
 		// through. Note that the combination of the above condition and passing
 		// RaftElectionTimeoutTicks to refreshProposalsLocked means that commands
 		// will be refreshed when they have been pending for 1 to 2 election
 		// cycles.
-		r.refreshProposalsLocked(
-			r.store.cfg.RaftElectionTimeoutTicks, reasonTicks,
-		)
+		//
+		// However, we don't refresh proposals if we are the leader because
+		// doing so would be useless. The commands tracked by a leader replica
+		// were either all proposed when the replica was a leader or were
+		// re-proposed when the replica became a leader. Either way, they are
+		// guaranteed to be in the leader's Raft log so re-proposing won't do
+		// anything.
+		r.refreshProposalsLocked(refreshAtDelta, reasonTicks)
 	}
 	return true, nil
 }
@@ -3948,6 +4001,9 @@ func (r *Replica) maybeTransferRaftLeader(
 	if !r.isLeaseValidRLocked(l, now) {
 		return
 	}
+	if r.store.TestingKnobs().DisableLeaderFollowsLeaseholder {
+		return
+	}
 	if pr, ok := status.Progress[uint64(l.Replica.ReplicaID)]; ok && pr.Match >= status.Commit {
 		log.VEventf(ctx, 1, "transferring raft leadership to replica ID %v", l.Replica.ReplicaID)
 		r.store.metrics.RangeRaftLeaderTransfers.Inc(1)
@@ -4609,6 +4665,9 @@ func (r *Replica) processRaftCommand(
 		delete(r.mu.proposals, idKey)
 	}
 
+	// Delete the entry for a forwarded proposal set.
+	delete(r.mu.remoteProposals, idKey)
+
 	leaseIndex, proposalRetry, forcedErr := r.checkForcedErrLocked(ctx, idKey, raftCmd, proposal, proposedLocally)
 
 	r.mu.Unlock()

pkg/storage/replica_proposal.go

@@ -274,7 +274,8 @@ func (r *Replica) leasePostApply(ctx context.Context, newLease roachpb.Lease) {
 		r.txnWaitQueue.Clear(true /* disable */)
 	}
 
-	if !iAmTheLeaseHolder && r.IsLeaseValid(newLease, r.store.Clock().Now()) {
+	if !iAmTheLeaseHolder && r.IsLeaseValid(newLease, r.store.Clock().Now()) &&
+		!r.store.TestingKnobs().DisableLeaderFollowsLeaseholder {
 		// If this replica is the raft leader but it is not the new lease holder,
 		// then try to transfer the raft leadership to match the lease. We like it
 		// when leases and raft leadership are collocated because that facilitates

pkg/storage/replica_test.go

@@ -7835,8 +7835,14 @@ func TestReplicaRefreshPendingCommandsTicks(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	electionTicks := tc.store.cfg.RaftElectionTimeoutTicks
+	// Only followers refresh pending commands during tick events. Change the
+	// replica that the range thinks is the leader so that the replica thinks
+	// it's a follower.
+	r.mu.Lock()
+	r.mu.leaderID = 2
+	r.mu.Unlock()
 
+	electionTicks := tc.store.cfg.RaftElectionTimeoutTicks
 	{
 		// The verifications of the reproposal counts below rely on r.mu.ticks
 		// starting with a value of 0 (modulo electionTicks). Move the replica into

pkg/storage/store.go

@@ -733,6 +733,9 @@ type StoreTestingKnobs struct {
 	DisableScanner bool
 	// DisablePeriodicGossips disables periodic gossiping.
 	DisablePeriodicGossips bool
+	// DisableLeaderFollowsLeaseholder disables attempts to transfer raft
+	// leadership when it diverges from the range's leaseholder.
+	DisableLeaderFollowsLeaseholder bool
 	// DisableRefreshReasonTicks disables refreshing pending commands when a new
 	// leader is discovered.
 	DisableRefreshReasonNewLeader bool
@@ -742,6 +745,10 @@ type StoreTestingKnobs struct {
 	// DisableRefreshReasonTicks disables refreshing pending commands
 	// periodically.
 	DisableRefreshReasonTicks bool
+	// RefreshReasonTicksPeriod overrides the default period over which
+	// pending commands are refreshed. The period is specified as a multiple
+	// of Raft group ticks.
+	RefreshReasonTicksPeriod int
 	// DisableProcessRaft disables the process raft loop.
 	DisableProcessRaft bool
 	// DisableLastProcessedCheck disables checking on replica queue last processed times.