volume: let evacuation proceed on a server in maintenance mode (#11145)

Maintenance mode exists to fence a volume server so it can be evacuated
without taking new writes (#7977), but the gate added in #8115 also
rejected the RPCs evacuation issues against the source: VolumeMarkReadonly
(the first step of every move, and the failure reported in #11066),
VolumeDelete (the last step), and VolumeEcShardsDelete (the last step for
EC shards). volumeServer.evacuate, volume.move and ec.balance therefore
all failed on exactly the server they were meant to drain.

Those three RPCs only remove data or restrict the server further, the same
class as DeleteCollection and the unmount RPCs that were never gated, so
they are exempted from the maintenance check in both the Go and Rust
volume servers. Everything that adds data or reopens the server for
writes (AllocateVolume, WriteNeedleBlob, BatchDelete, VolumeCopy,
ReceiveFile, EC generate/copy/rebuild, vacuum, tiering, VolumeMarkWritable)
stays blocked. A side effect is that scrub can now fence broken volumes
readonly on a server already in maintenance.

Fixes #11066

Generated with [Devin](https://devin.ai)

Co-authored-by: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

test/volume_server/grpc/erasure_coding_test.go

@@ -27,24 +27,27 @@ func TestEcMaintenanceModeRejections(t *testing.T) {
 	conn, grpcClient := framework.DialVolumeServer(t, clusterHarness.VolumeGRPCAddress())
 	defer conn.Close()
 
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	// Encode a volume before entering maintenance mode so the shard delete
+	// below has real shards to remove.
+	const ecVolumeID = uint32(1)
+	framework.AllocateVolume(t, grpcClient, ecVolumeID, "")
+	httpClient := framework.NewHTTPClient()
+	uploadResp := framework.UploadBytes(t, httpClient, clusterHarness.VolumeAdminURL(), framework.NewFileID(ecVolumeID, 990001, 0x11223344), []byte("ec-maintenance-content"))
+	_ = framework.ReadAllAndClose(t, uploadResp)
+	if uploadResp.StatusCode != http.StatusCreated {
+		t.Fatalf("upload expected 201, got %d", uploadResp.StatusCode)
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()
 
-	stateResp, err := grpcClient.GetState(ctx, &volume_server_pb.GetStateRequest{})
-	if err != nil {
-		t.Fatalf("GetState failed: %v", err)
-	}
-	_, err = grpcClient.SetState(ctx, &volume_server_pb.SetStateRequest{
-		State: &volume_server_pb.VolumeServerState{
-			Maintenance: true,
-			Version:     stateResp.GetState().GetVersion(),
-		},
-	})
-	if err != nil {
-		t.Fatalf("SetState maintenance=true failed: %v", err)
+	if _, err := grpcClient.VolumeEcShardsGenerate(ctx, &volume_server_pb.VolumeEcShardsGenerateRequest{VolumeId: ecVolumeID, Collection: ""}); err != nil {
+		t.Fatalf("VolumeEcShardsGenerate before maintenance failed: %v", err)
 	}
 
-	_, err = grpcClient.VolumeEcShardsGenerate(ctx, &volume_server_pb.VolumeEcShardsGenerateRequest{VolumeId: 1, Collection: ""})
+	framework.EnableMaintenanceMode(t, ctx, grpcClient)
+
+	_, err := grpcClient.VolumeEcShardsGenerate(ctx, &volume_server_pb.VolumeEcShardsGenerateRequest{VolumeId: 1, Collection: ""})
 	if err == nil || !strings.Contains(err.Error(), "maintenance mode") {
 		t.Fatalf("VolumeEcShardsGenerate maintenance error mismatch: %v", err)
 	}
@@ -59,13 +62,23 @@ func TestEcMaintenanceModeRejections(t *testing.T) {
 		t.Fatalf("VolumeEcShardsCopy maintenance error mismatch: %v", err)
 	}
 
+	// Deleting shards removes data rather than adding it, and is how evacuating
+	// EC shards off a server in maintenance mode finishes (issue #11066).
 	_, err = grpcClient.VolumeEcShardsDelete(ctx, &volume_server_pb.VolumeEcShardsDeleteRequest{
-		VolumeId:   1,
+		VolumeId:   ecVolumeID,
 		Collection: "",
 		ShardIds:   []uint32{0},
 	})
-	if err == nil || !strings.Contains(err.Error(), "maintenance mode") {
-		t.Fatalf("VolumeEcShardsDelete maintenance error mismatch: %v", err)
+	if err != nil {
+		t.Fatalf("VolumeEcShardsDelete should succeed in maintenance mode, got: %v", err)
+	}
+	_, err = grpcClient.VolumeEcShardsMount(ctx, &volume_server_pb.VolumeEcShardsMountRequest{
+		VolumeId:   ecVolumeID,
+		Collection: "",
+		ShardIds:   []uint32{0},
+	})
+	if err == nil {
+		t.Fatalf("VolumeEcShardsMount should fail once shard 0 has been deleted")
 	}
 
 	_, err = grpcClient.VolumeEcBlobDelete(ctx, &volume_server_pb.VolumeEcBlobDeleteRequest{