reconcilable_result: use chunked_vector to hold partitions

Usually, a reconcilable_result holds very few partitions (1 is common),
since the page size is limited by 1MB. But if we have paging disabled or
if we are reconciling a range full of tombstones, we may see many more.
This can cause large allocations.

Change to chunked_vector to prevent those large allocations, as they
can be quite expensive.

Fixes #4780.

(cherry picked from commit 093d2cd)

idl/reconcilable_result.idl.hh

@@ -26,6 +26,6 @@ class partition {
 
 class reconcilable_result {
     uint32_t row_count();
-    std::vector<partition> partitions();
+    utils::chunked_vector<partition> partitions();
     query::short_read is_short_read() [[version 1.6]] = query::short_read::no;
 };

mutation_query.cc

@@ -31,19 +31,19 @@ reconcilable_result::reconcilable_result()
     : _row_count(0)
 { }
 
-reconcilable_result::reconcilable_result(uint32_t row_count, std::vector<partition> p, query::short_read short_read,
+reconcilable_result::reconcilable_result(uint32_t row_count, utils::chunked_vector<partition> p, query::short_read short_read,
                                          query::result_memory_tracker memory_tracker)
     : _row_count(row_count)
     , _short_read(short_read)
     , _memory_tracker(std::move(memory_tracker))
     , _partitions(std::move(p))
 { }
 
-const std::vector<partition>& reconcilable_result::partitions() const {
+const utils::chunked_vector<partition>& reconcilable_result::partitions() const {
     return _partitions;
 }
 
-std::vector<partition>& reconcilable_result::partitions() {
+utils::chunked_vector<partition>& reconcilable_result::partitions() {
     return _partitions;
 }
 

mutation_query.hh

@@ -27,6 +27,7 @@
 #include "frozen_mutation.hh"
 #include "db/timeout_clock.hh"
 #include "querier.hh"
+#include "utils/chunked_vector.hh"
 #include <seastar/core/execution_stage.hh>
 
 class reconcilable_result;
@@ -72,17 +73,17 @@ class reconcilable_result {
     uint32_t _row_count;
     query::short_read _short_read;
     query::result_memory_tracker _memory_tracker;
-    std::vector<partition> _partitions;
+    utils::chunked_vector<partition> _partitions;
 public:
     ~reconcilable_result();
     reconcilable_result();
     reconcilable_result(reconcilable_result&&) = default;
     reconcilable_result& operator=(reconcilable_result&&) = default;
-    reconcilable_result(uint32_t row_count, std::vector<partition> partitions, query::short_read short_read,
+    reconcilable_result(uint32_t row_count, utils::chunked_vector<partition> partitions, query::short_read short_read,
                         query::result_memory_tracker memory_tracker = { });
 
-    const std::vector<partition>& partitions() const;
-    std::vector<partition>& partitions();
+    const utils::chunked_vector<partition>& partitions() const;
+    utils::chunked_vector<partition>& partitions();
 
     uint32_t row_count() const {
         return _row_count;
@@ -112,7 +113,7 @@ class reconcilable_result_builder {
     const schema& _schema;
     const query::partition_slice& _slice;
 
-    std::vector<partition> _result;
+    utils::chunked_vector<partition> _result;
     uint32_t _live_rows{};
 
     bool _has_ck_selector{};

service/storage_proxy.cc

@@ -2751,8 +2751,8 @@ class data_read_resolver : public abstract_read_resolver {
 
         // build reconcilable_result from reconciled data
         // traverse backwards since large keys are at the start
-        std::vector<partition> vec;
-        auto r = boost::accumulate(reconciled_partitions | boost::adaptors::reversed, std::ref(vec), [] (std::vector<partition>& a, const mutation_and_live_row_count& m_a_rc) {
+        utils::chunked_vector<partition> vec;
+        auto r = boost::accumulate(reconciled_partitions | boost::adaptors::reversed, std::ref(vec), [] (utils::chunked_vector<partition>& a, const mutation_and_live_row_count& m_a_rc) {
             a.emplace_back(partition(m_a_rc.live_row_count, freeze(m_a_rc.mut)));
             return std::ref(a);
         });