feat(pulsar): support distributed indexing

docs/configuration/source-config.md

@@ -223,7 +223,7 @@ will be decided by the control plane.
 
 :::info
 
-Note that distributing the indexing load of partitioned sources like Kafka is done by assigning the different partitions to different pipelines. As a result, it is important to ensure that the number of partitions is a multiple of `num_pipelines`.
+Note that distributing the indexing load of partitioned sources like Kafka and Pulsar is done by assigning the different partitions to different pipelines. As a result, it is important to ensure that the number of partitions is a multiple of `num_pipelines`. For Pulsar specifically, the topic(s) must be partitioned: a non-partitioned Pulsar topic is served by a single active consumer, so extra pipelines stay idle and provide no parallelism.
 
 Also, assuming you are only indexing a single Kafka source in your Quickwit cluster, you should set the number of pipelines to a multiple of the number of indexers. Finally, if your indexing throughput is high, you should provision between 2 and 4 vCPUs per pipeline.
 

docs/ingest-data/pulsar.md

@@ -158,6 +158,29 @@ As soon as the Pulsar source is created, Quickwit control plane will ask an inde
 INFO spawn_pipeline{index=stackoverflow gen=0}:pulsar-consumer{subscription_name="quickwit-stackoverflow-pulsar-source" params=PulsarSourceParams { topics: ["stackoverflow"], address: "pulsar://localhost:6650", consumer_name: "quickwit", authentication: None } current_positions={}}: quickwit_indexing::source::pulsar_source: Seeking to last checkpoint positions. positions={}
 ```
 
+### Distributed indexing
+
+To parallelize indexing across several pipelines (and indexers), set `num_pipelines` on the source config to the number of pipelines you want:
+
+```yaml
+version: 0.8
+source_id: pulsar-source
+source_type: pulsar
+num_pipelines: 4
+params:
+  topics:
+    - stackoverflow
+  address: pulsar://localhost:6650
+```
+
+The pipelines share a single Pulsar `Failover` subscription, and the broker assigns each topic partition to exactly one active pipeline — so each partition is indexed by a single pipeline, with no duplicates and no loss. When a pipeline dies, Pulsar promotes another and it resumes from the subscription cursor.
+
+:::info
+
+Distribution happens **per partition**, so the topic(s) must be partitioned. A non-partitioned topic is served by a single active consumer: extra pipelines stay idle as standbys and provide no parallelism. For balanced load, make the number of partitions a multiple of `num_pipelines`.
+
+:::
+
 ## Create and populate a Pulsar topic
 
 We will use the Pulsar's default tenant/namespace `public/default`. To populate the topic, we will use a python script:

quickwit/quickwit-config/src/source_config/mod.rs

@@ -566,7 +566,10 @@ pub struct PulsarSourceParams {
     pub address: String,
     #[schema(default = "quickwit")]
     #[serde(default = "default_consumer_name")]
-    /// The name to register with the pulsar source.
+    /// The consumer name prefix to register with the pulsar source. With
+    /// distributed indexing (`num_pipelines > 1`), each pipeline appends its
+    /// pipeline UID to this prefix so it connects as a distinct consumer under
+    /// the shared subscription.
     pub consumer_name: String,
     // Serde yaml has some specific behaviour when deserializing
     // enums (see https://github.com/dtolnay/serde-yaml/issues/342)
@@ -955,15 +958,14 @@ mod tests {
                 "source_type": "pulsar",
                 "params": {
                     "topics": ["my-topic"],
-                    "address": "http://localhost:6650"
+                    "address": "pulsar://localhost:6650"
                 }
             }
             "#;
-            load_source_config_from_user_config(ConfigFormat::Json, content.as_bytes())
-                .unwrap_err();
-            // TODO: uncomment asserts once distributed indexing is activated for pulsar.
-            // assert_eq!(source_config.num_pipelines(), 3);
-            // assert_eq!(source_config.max_num_pipelines_per_indexer(), 3);
+            let source_config =
+                load_source_config_from_user_config(ConfigFormat::Json, content.as_bytes())
+                    .unwrap();
+            assert_eq!(source_config.num_pipelines.get(), 3);
         }
     }
 
@@ -986,6 +988,24 @@ mod tests {
                     .unwrap();
             assert_eq!(source_config.num_pipelines.get(), 3);
         }
+        {
+            let content = r#"
+            {
+                "version": "0.8",
+                "source_id": "hdfs-logs-pulsar-source",
+                "num_pipelines": 3,
+                "source_type": "pulsar",
+                "params": {
+                    "topics": ["my-topic"],
+                    "address": "pulsar://localhost:6650"
+                }
+            }
+            "#;
+            let source_config =
+                load_source_config_from_user_config(ConfigFormat::Json, content.as_bytes())
+                    .unwrap();
+            assert_eq!(source_config.num_pipelines.get(), 3);
+        }
     }
 
     #[test]

quickwit/quickwit-config/src/source_config/serialize.rs

@@ -126,11 +126,12 @@ impl SourceConfigForSerialization {
         match &self.source_params {
             SourceParams::PubSub(_)
             | SourceParams::Kafka(_)
+            | SourceParams::Pulsar(_)
             | SourceParams::File(FileSourceParams::Notifications(_)) => {}
             _ => {
                 if self.num_pipelines > 1 {
                     bail!(
-                        "Quickwit currently supports multiple pipelines only for GCP PubSub or Kafka sources. open an issue https://github.com/quickwit-oss/quickwit/issues if you need the feature for other source types"
+                        "Quickwit currently supports multiple pipelines only for GCP PubSub, Kafka or Pulsar sources. open an issue https://github.com/quickwit-oss/quickwit/issues if you need the feature for other source types"
                     );
                 }
             }

quickwit/quickwit-indexing/src/source/mod.rs

@@ -595,6 +595,7 @@ mod tests {
         source_config: SourceConfig,
         metastore_opt: Option<MetastoreServiceClient>,
         queues_dir_path_opt: Option<PathBuf>,
+        pipeline_uid: PipelineUid,
     }
 
     impl SourceRuntimeBuilder {
@@ -604,6 +605,7 @@ mod tests {
                 source_config,
                 metastore_opt: None,
                 queues_dir_path_opt: None,
+                pipeline_uid: PipelineUid::for_test(0u128),
             }
         }
 
@@ -622,7 +624,7 @@ mod tests {
                     node_id: NodeId::from_str("test-node"),
                     index_uid: self.index_uid,
                     source_id: self.source_config.source_id.clone(),
-                    pipeline_uid: PipelineUid::for_test(0u128),
+                    pipeline_uid: self.pipeline_uid,
                 },
                 metastore,
                 ingester_pool: IngesterPool::default(),
@@ -656,6 +658,12 @@ mod tests {
             self
         }
 
+        #[cfg(all(test, feature = "pulsar-broker-tests"))]
+        pub fn with_pipeline_uid(mut self, pipeline_uid: PipelineUid) -> Self {
+            self.pipeline_uid = pipeline_uid;
+            self
+        }
+
         fn setup_mock_metastore(
             &self,
             source_checkpoint_delta_opt: Option<SourceCheckpointDelta>,