Merge pull request #27711 from mmaslankaprv/cl-sg-follow-up

Use shadow linking scheduling group in replicator and topic reconciler

Author: mmaslankaprv

src/v/cluster_link/manager.cc

@@ -585,7 +585,8 @@ ss::future<> manager::start_topic_reconciler() {
           _topic_metadata_cache.get(),
           _registry.get(),
           topic_reconciler_interval,
-          _default_topic_replication);
+          _default_topic_replication,
+          _scheduling_group);
     }
     try {
         co_await _topic_reconciler->start();

src/v/cluster_link/replication/link_replication_mgr.cc

@@ -61,7 +61,7 @@ ss::future<> link_replication_manager::do_start_replicator(
     auto source = _source_factory->make_source(ntp);
     auto sink = _sink_factory->make_sink(ntp);
     auto replicator = std::make_unique<partition_replicator>(
-      ntp, term, std::move(source), std::move(sink));
+      ntp, term, std::move(source), std::move(sink), _sg);
     auto [r_it, _] = _replicators.emplace(ntp, std::move(replicator));
     co_await r_it->second->start();
 }

src/v/cluster_link/replication/partition_replicator.cc

@@ -13,6 +13,8 @@
 #include "cluster_link/logger.h"
 #include "ssx/future-util.h"
 
+#include <seastar/coroutine/switch_to.hh>
+
 namespace cluster_link::replication {
 
 static constexpr std::chrono::seconds base_backoff{1};
@@ -22,16 +24,19 @@ partition_replicator::partition_replicator(
   const model::ntp& ntp,
   model::term_id term,
   std::unique_ptr<data_source> source,
-  std::unique_ptr<data_sink> sink)
+  std::unique_ptr<data_sink> sink,
+  ss::scheduling_group scheduling_group)
   : _term(term)
   , _log(cllog, fmt::format("[{}-term-{}] replicator", ntp, term))
   , _source(std::move(source))
   , _sink(std::move(sink))
+  , _scheduling_group(scheduling_group)
   , _backoff_policy(
       make_exponential_backoff_policy<ss::lowres_clock>(
         base_backoff, max_backoff)) {}
 
 ss::future<> partition_replicator::start() {
+    co_await ss::coroutine::switch_to(_scheduling_group);
     vlog(_log.trace, "Starting replicator");
     co_await _sink->start();
     co_await _source->start(
@@ -48,6 +53,7 @@ ss::future<> partition_replicator::start() {
 }
 
 ss::future<> partition_replicator::stop() {
+    co_await ss::coroutine::switch_to(_scheduling_group);
     vlog(_log.trace, "Stopping replicator");
     _as.request_abort();
     // closing the gate first ensures all the units are returned to the

src/v/cluster_link/replication/partition_replicator.h

@@ -58,7 +58,8 @@ class partition_replicator {
       const ::model::ntp& ntp,
       ::model::term_id,
       std::unique_ptr<data_source> source,
-      std::unique_ptr<data_sink> sink);
+      std::unique_ptr<data_sink> sink,
+      ss::scheduling_group sg = ss::default_scheduling_group());
     ss::future<> start();
     ss::future<> stop();
 
@@ -88,6 +89,7 @@ class partition_replicator {
     ss::abort_source _as;
     std::unique_ptr<data_source> _source;
     std::unique_ptr<data_sink> _sink;
+    ss::scheduling_group _scheduling_group;
     // to pipeline multiple replicate requests in parallel
     static constexpr ssize_t max_in_flight_requests = 5;
     ssx::semaphore _max_requests{

src/v/cluster_link/tests/topic_reconciler_test.cc

@@ -54,7 +54,8 @@ class topic_reconciler_test : public seastar_test {
           _ftmc.get(),
           _link_registry.get(),
           1s,
-          _default_topic_replication.bind());
+          _default_topic_replication.bind(),
+          ss::default_scheduling_group());
         co_await _reconciler->start();
 
         set_required_topic_properties(

src/v/cluster_link/topic_reconciler.cc

@@ -18,6 +18,8 @@
 #include "model/fundamental.h"
 #include "ssx/future-util.h"
 
+#include <seastar/coroutine/switch_to.hh>
+
 #include <exception>
 
 using namespace std::chrono_literals;
@@ -28,12 +30,14 @@ topic_reconciler::topic_reconciler(
   kafka::data::rpc::topic_metadata_cache* topic_metadata_cache,
   link_registry* link_registry,
   ss::lowres_clock::duration run_interval,
-  config::binding<int16_t> default_topic_replication)
+  config::binding<int16_t> default_topic_replication,
+  ss::scheduling_group scheduling_group)
   : _topic_creator(topic_creator)
   , _topic_metadata_cache(topic_metadata_cache)
   , _link_registry(link_registry)
   , _run_interval(run_interval)
-  , _default_topic_replication(std::move(default_topic_replication)) {}
+  , _default_topic_replication(std::move(default_topic_replication))
+  , _scheduling_group(scheduling_group) {}
 
 ss::future<> topic_reconciler::start() {
     vlog(cllog.info, "Starting topic reconciler");
@@ -79,6 +83,7 @@ void topic_reconciler::trigger(model::id_t link_id) {
 }
 
 ss::future<> topic_reconciler::execute(std::optional<model::id_t> link_id) {
+    co_await ss::coroutine::switch_to(_scheduling_group);
     static constexpr auto max_concurrent_reconcilations = 5;
     vlog(cllog.trace, "Executing topic reconciler, link_id: {}", link_id);
     auto units = co_await _reconciler_mutex.get_units(_as);

src/v/cluster_link/topic_reconciler.h

@@ -28,7 +28,8 @@ class topic_reconciler {
       kafka::data::rpc::topic_metadata_cache* topic_metadata_cache,
       link_registry* link_registry,
       ss::lowres_clock::duration run_interval,
-      config::binding<int16_t> default_topic_replication);
+      config::binding<int16_t> default_topic_replication,
+      ss::scheduling_group sg);
     topic_reconciler(const topic_reconciler&) = delete;
     topic_reconciler(topic_reconciler&&) = delete;
     topic_reconciler& operator=(const topic_reconciler&) = delete;
@@ -79,6 +80,7 @@ class topic_reconciler {
     ss::timer<ss::lowres_clock> _reconciler_timer;
     ss::lowres_clock::duration _run_interval;
     config::binding<int16_t> _default_topic_replication;
+    ss::scheduling_group _scheduling_group;
     ss::abort_source _as;
     ss::gate _gate;
 };