storage: indicate when a split causes a Raft snapshot

Touches cockroachdb#31409.

Release note: None

Author: tbg

pkg/storage/allocator_test.go

@@ -306,6 +306,7 @@ func replicas(storeIDs ...roachpb.StoreID) []roachpb.ReplicaDescriptor {
 	for i, storeID := range storeIDs {
 		res[i].NodeID = roachpb.NodeID(storeID)
 		res[i].StoreID = storeID
+		res[i].ReplicaID = roachpb.ReplicaID(i + 1)
 	}
 	return res
 }

pkg/storage/replica_command.go

@@ -22,9 +22,6 @@ import (
 	"strings"
 	"time"
 
-	"github.com/pkg/errors"
-	"go.etcd.io/etcd/raft/raftpb"
-
 	"github.com/cockroachdb/cockroach/pkg/base"
 	"github.com/cockroachdb/cockroach/pkg/internal/client"
 	"github.com/cockroachdb/cockroach/pkg/keys"
@@ -40,6 +37,9 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/util/log"
 	"github.com/cockroachdb/cockroach/pkg/util/protoutil"
 	"github.com/cockroachdb/cockroach/pkg/util/retry"
+	"github.com/pkg/errors"
+	"go.etcd.io/etcd/raft"
+	"go.etcd.io/etcd/raft/raftpb"
 )
 
 // evaluateCommand delegates to the eval method for the given
@@ -197,6 +197,23 @@ func maybeDescriptorChangedError(desc *roachpb.RangeDescriptor, err error) (stri
 	return "", false
 }
 
+func splitSnapshotWarningStr(rangeID roachpb.RangeID, status *raft.Status) string {
+	var s string
+	if status != nil && status.RaftState == raft.StateLeader {
+		for replicaID, pr := range status.Progress {
+			if replicaID == status.Lead {
+				// TODO(tschottdorf): remove this line once we have picked up
+				// https://github.com/etcd-io/etcd/pull/10279
+				continue
+			}
+			if pr.State != raft.ProgressStateReplicate {
+				s += fmt.Sprintf("; may cause Raft snapshot to r%d/%d: %v", rangeID, replicaID, &pr)
+			}
+		}
+	}
+	return s
+}
+
 // adminSplitWithDescriptor divides the range into into two ranges, using
 // either args.SplitKey (if provided) or an internally computed key that aims
 // to roughly equipartition the range by size. The split is done inside of a
@@ -295,8 +312,10 @@ func (r *Replica) adminSplitWithDescriptor(
 	}
 	leftDesc.EndKey = splitKey
 
-	log.Infof(ctx, "initiating a split of this range at key %s [r%d]",
-		splitKey, rightDesc.RangeID)
+	extra := splitSnapshotWarningStr(r.RangeID, r.RaftStatus())
+
+	log.Infof(ctx, "initiating a split of this range at key %s [r%d]%s",
+		splitKey, rightDesc.RangeID, extra)
 
 	if err := r.store.DB().Txn(ctx, func(ctx context.Context, txn *client.Txn) error {
 		log.Event(ctx, "split closure begins")

pkg/storage/replica_test.go

@@ -33,6 +33,7 @@ import (
 	"github.com/gogo/protobuf/proto"
 	"github.com/kr/pretty"
 	"github.com/pkg/errors"
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"go.etcd.io/etcd/raft"
 	"go.etcd.io/etcd/raft/raftpb"
@@ -134,6 +135,22 @@ func leaseExpiry(repl *Replica) int64 {
 	return l.Expiration.WallTime + 1
 }
 
+// Create a Raft status that shows everyone fully up to date.
+func upToDateRaftStatus(repls []roachpb.ReplicaDescriptor) *raft.Status {
+	prs := make(map[uint64]raft.Progress)
+	for _, repl := range repls {
+		prs[uint64(repl.ReplicaID)] = raft.Progress{
+			State: raft.ProgressStateReplicate,
+			Match: 100,
+		}
+	}
+	return &raft.Status{
+		HardState: raftpb.HardState{Commit: 100},
+		SoftState: raft.SoftState{Lead: 1, RaftState: raft.StateLeader},
+		Progress:  prs,
+	}
+}
+
 // testContext contains all the objects necessary to test a Range.
 // In most cases, simply call Start(t) (and later Stop()) on a zero-initialized
 // testContext{}. Any fields which are initialized to non-nil values
@@ -10856,3 +10873,21 @@ func TestRollbackMissingTxnRecordNoError(t *testing.T) {
 		t.Errorf("expected %s; got %v", expErr, pErr)
 	}
 }
+
+func TestSplitSnapshotWarningStr(t *testing.T) {
+	defer leaktest.AfterTest(t)()
+
+	status := upToDateRaftStatus(replicas(1, 3, 5))
+	assert.Equal(t, "", splitSnapshotWarningStr(12, status))
+
+	pr := status.Progress[2]
+	pr.State = raft.ProgressStateProbe
+	status.Progress[2] = pr
+
+	assert.Equal(
+		t,
+		"; may cause Raft snapshot to r12/2: next = 0, match = 100, state = ProgressStateProbe,"+
+			" waiting = false, pendingSnapshot = 0",
+		splitSnapshotWarningStr(12, status),
+	)
+}