io_queue: Scatter requests across shards evenly

All requests that get queued to specific ptiotity class are spread among all shards evenly in plain round-robin manner. Signed-off-by: Pavel Emelyanov <[email protected]>
xemul · Jun 7, 2024 · 23a8769 · 23a8769
1 parent 5236129
commit 23a8769
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 2 deletions.
diff --git a/include/seastar/core/io_queue.hh b/include/seastar/core/io_queue.hh
@@ -107,7 +107,7 @@ private:
 
     priority_class_data& find_or_create_class(internal::priority_class pc);
     future<size_t> queue_request(internal::priority_class pc, internal::io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs) noexcept;
-    future<size_t> queue_one_request(internal::priority_class pc, internal::io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs) noexcept;
+    future<size_t> queue_one_request(internal::priority_class pc, internal::io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs, bool foreign = false) noexcept;
     future<size_t> queue_one_request(priority_class_data& pclass, internal::io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs) noexcept;
 
     // The fields below are going away, they are just here so we can implement deprecated

diff --git a/src/core/io_queue.cc b/src/core/io_queue.cc
@@ -116,6 +116,7 @@ class io_queue::priority_class_data {
     std::chrono::duration<double> _total_execution_time;
     std::chrono::duration<double> _starvation_time;
     io_queue::clock_type::time_point _activated;
+    unsigned _balance_round_robbin = 0;
 
     io_group::priority_class_data& _group;
     size_t _replenish_head;
@@ -212,6 +213,10 @@ class io_queue::priority_class_data {
 
     std::vector<seastar::metrics::impl::metric_definition_impl> metrics();
     metrics::metric_groups metric_groups;
+
+    unsigned balance() noexcept {
+        return _balance_round_robbin++ % smp::count;
+    }
 };
 
 class io_desc_read_write final : public io_completion {
@@ -957,9 +962,18 @@ future<size_t> io_queue::queue_one_request(priority_class_data& pclass, io_direc
     return fut;
 }
 
-future<size_t> io_queue::queue_one_request(internal::priority_class pc, io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs) noexcept {
+future<size_t> io_queue::queue_one_request(internal::priority_class pc, io_direction_and_length dnl, internal::io_request req, io_intent* intent, iovec_keeper iovs, bool foreign) noexcept {
     try {
         auto& pclass = find_or_create_class(pc);
+        if (!foreign) {
+            auto target = pclass.balance();
+            if (target != this_shard_id()) {
+                // FIXME: intent is lost :(
+                return smp::submit_to(target, [pc = std::move(pc), dnl = std::move(dnl), req = std::move(req), iovs = std::move(iovs), g = _group] () mutable {
+                    return g->_io_queues[this_shard_id()]->queue_one_request(std::move(pc), std::move(dnl), std::move(req), nullptr, std::move(iovs), true);
+                });
+            }
+        }
         return queue_one_request(pclass, std::move(dnl), std::move(req), intent, std::move(iovs));
     } catch (...) {
         return make_exception_future<size_t>(std::current_exception());