feat(serverless backfill): proto changes (#17010)

Co-authored-by: ArkBriar <[email protected]> Co-authored-by: Shanicky Chen <[email protected]> Co-authored-by: August <[email protected]>
risingwavelabs · Jul 24, 2024 · ffc6d4a · ffc6d4a
1 parent f2fe044
commit ffc6d4a
Show file tree

Hide file tree

Showing 9 changed files with 79 additions and 2 deletions.
diff --git a/proto/common.proto b/proto/common.proto
@@ -82,6 +82,10 @@ message WorkerNode {
   optional uint64 started_at = 9;
 
   uint32 parallelism = 10;
+
+  // Meta may assign labels to worker nodes to partition workload by label.
+  // This is used for serverless backfilling of materialized views.
+  string node_label = 11;
 }
 
 message Buffer {

diff --git a/proto/ddl_service.proto b/proto/ddl_service.proto
@@ -120,6 +120,16 @@ message DropSubscriptionResponse {
 message CreateMaterializedViewRequest {
   catalog.Table materialized_view = 1;
   stream_plan.StreamFragmentGraph fragment_graph = 2;
+
+  // If SERVERLESS, the materialized view should be created using serverless backfill
+  // For that the controller will create a new compute node, which does backfilling and then is deleted.
+  // May alleviate pressure on the cluster during backfill process.
+  enum BackfillType {
+    UNSPECIFIED = 0;
+    REGULAR = 1;
+    SERVERLESS = 2;
+  }
+  BackfillType backfill = 3;
 }
 
 message CreateMaterializedViewResponse {

diff --git a/proto/meta.proto b/proto/meta.proto
@@ -103,6 +103,13 @@ message TableFragments {
 
   stream_plan.StreamContext ctx = 6;
   TableParallelism parallelism = 7;
+
+  // Actors of a materialize view, sink, or table can only be scheduled on nodes with matching node_label.
+  string node_label = 8;
+
+  // If this is a materialized view: True if backfill is done, else false.
+  // If this is a regular table: Always true.
+  bool backfill_done = 9;
 }
 
 /// Worker slot mapping with fragment id, used for notification.
@@ -466,7 +473,10 @@ message SubscribeResponse {
   }
   common.Status status = 1;
   Operation operation = 2;
+
+  // Catalog version
   uint64 version = 3;
+
   oneof info {
     catalog.Database database = 4;
     catalog.Schema schema = 5;
@@ -548,9 +558,37 @@ message TableParallelism {
   }
 }
 
+// Changes a streaming job in place by overwriting its node_label.
+// This may cause the re-scheduling of the streaming job actors.
+message UpdateStreamingJobNodeLabelsRequest {
+  // Id of the materialized view, table, or sink which we want to update
+  uint32 id = 1;
+
+  // replace the node_label of the streaming job with a given id with below value
+  string node_label = 2;
+}
+
+// We do not need to add an explicit status field here, we can just use the RPC status
+message UpdateStreamingJobNodeLabelsResponse {}
+
+message GetServerlessStreamingJobsStatusRequest {}
+
+// Descriptions of MVs and sinks
+message GetServerlessStreamingJobsStatusResponse {
+  message Status {
+    uint32 table_id = 1;
+    string node_label = 2;
+    bool backfill_done = 3;
+  }
+
+  repeated Status streaming_job_statuses = 1;
+}
+
 service ScaleService {
   rpc GetClusterInfo(GetClusterInfoRequest) returns (GetClusterInfoResponse);
   rpc Reschedule(RescheduleRequest) returns (RescheduleResponse);
+  rpc UpdateStreamingJobNodeLabels(UpdateStreamingJobNodeLabelsRequest) returns (UpdateStreamingJobNodeLabelsResponse);
+  rpc GetServerlessStreamingJobsStatus(GetServerlessStreamingJobsStatusRequest) returns (GetServerlessStreamingJobsStatusResponse);
 }
 
 message MembersRequest {}

diff --git a/src/meta/service/src/scale_service.rs b/src/meta/service/src/scale_service.rs
@@ -24,8 +24,9 @@ use risingwave_meta_model_v2::FragmentId;
 use risingwave_pb::common::WorkerType;
 use risingwave_pb::meta::scale_service_server::ScaleService;
 use risingwave_pb::meta::{
-    GetClusterInfoRequest, GetClusterInfoResponse, PbWorkerReschedule, RescheduleRequest,
-    RescheduleResponse,
+    GetClusterInfoRequest, GetClusterInfoResponse, GetServerlessStreamingJobsStatusRequest,
+    GetServerlessStreamingJobsStatusResponse, PbWorkerReschedule, RescheduleRequest,
+    RescheduleResponse, UpdateStreamingJobNodeLabelsRequest, UpdateStreamingJobNodeLabelsResponse,
 };
 use risingwave_pb::source::{ConnectorSplit, ConnectorSplits};
 use tonic::{Request, Response, Status};
@@ -219,4 +220,18 @@ impl ScaleService for ScaleServiceImpl {
             revision: next_revision.into(),
         }))
     }
+
+    async fn update_streaming_job_node_labels(
+        &self,
+        _request: Request<UpdateStreamingJobNodeLabelsRequest>,
+    ) -> Result<Response<UpdateStreamingJobNodeLabelsResponse>, Status> {
+        todo!()
+    }
+
+    async fn get_serverless_streaming_jobs_status(
+        &self,
+        _request: Request<GetServerlessStreamingJobsStatusRequest>,
+    ) -> Result<Response<GetServerlessStreamingJobsStatusResponse>, Status> {
+        todo!()
+    }
 }
diff --git a/src/meta/src/controller/cluster.rs b/src/meta/src/controller/cluster.rs
@@ -88,6 +88,7 @@ impl From<WorkerInfo> for PbWorkerNode {
             transactional_id: info.0.transaction_id.map(|id| id as _),
             resource: info.2.resource,
             started_at: info.2.started_at,
+            node_label: "".to_string(),
         }
     }
 }
@@ -465,6 +466,7 @@ fn meta_node_info(host: &str, started_at: Option<u64>) -> PbWorkerNode {
             total_cpu_cores: total_cpu_available() as _,
         }),
         started_at,
+        node_label: "".to_string(),
     }
 }
 

diff --git a/src/meta/src/controller/fragment.rs b/src/meta/src/controller/fragment.rs
@@ -340,6 +340,8 @@ impl CatalogController {
                 }
                 .into(),
             ),
+            node_label: "".to_string(),
+            backfill_done: true,
         };
 
         Ok(table_fragments)

diff --git a/src/meta/src/manager/cluster.rs b/src/meta/src/manager/cluster.rs
@@ -215,6 +215,7 @@ impl ClusterManager {
             // resource doesn't need persist
             resource: None,
             started_at: None,
+            node_label: "".to_string(),
         };
 
         let mut worker = Worker::from_protobuf(worker_node.clone());
@@ -771,6 +772,7 @@ fn meta_node_info(host: &str, started_at: Option<u64>) -> WorkerNode {
         }),
         started_at,
         parallelism: 0,
+        node_label: "".to_string(),
     }
 }
 

diff --git a/src/meta/src/model/stream.rs b/src/meta/src/model/stream.rs
@@ -164,6 +164,8 @@ impl MetadataModel for TableFragments {
             actor_splits: build_actor_connector_splits(&self.actor_splits),
             ctx: Some(self.ctx.to_protobuf()),
             parallelism: Some(self.assigned_parallelism.into()),
+            node_label: "".to_string(),
+            backfill_done: true,
         }
     }
 

diff --git a/src/rpc_client/src/meta_client.rs b/src/rpc_client/src/meta_client.rs
@@ -51,6 +51,7 @@ use risingwave_pb::cloud_service::*;
 use risingwave_pb::common::{HostAddress, WorkerNode, WorkerType};
 use risingwave_pb::connector_service::sink_coordination_service_client::SinkCoordinationServiceClient;
 use risingwave_pb::ddl_service::alter_owner_request::Object;
+use risingwave_pb::ddl_service::create_materialized_view_request::PbBackfillType;
 use risingwave_pb::ddl_service::ddl_service_client::DdlServiceClient;
 use risingwave_pb::ddl_service::drop_table_request::SourceId;
 use risingwave_pb::ddl_service::*;
@@ -353,6 +354,7 @@ impl MetaClient {
         let request = CreateMaterializedViewRequest {
             materialized_view: Some(table),
             fragment_graph: Some(graph),
+            backfill: PbBackfillType::Regular as _,
         };
         let resp = self.inner.create_materialized_view(request).await?;
         // TODO: handle error in `resp.status` here