Rr/sc 60366 sparse global order reader merge (#5417)

Sc-60366

Implements "pre-process tile order" mode for the sparse global order reader.
In brief, this does a first pass over the fragment metadata tile MBRs
to create a single unified list of all the (fragment ID, tile ID) pairs arranged
approximately in global order.

When running natively, the tile order is computed a single time the first
time the query is submitted, and then kept in memory.  When running
on the REST server, the tile order is recomputed for each query message.
Evidence indicates that the overhead of this is low; we expect that it is
lower than serializing the tile order would be.

- Add parallel merge algorithm
- Add rapidcheck library and some common scaffolding code
- Add `tiledb_submit_a_b` performance testing binary
- Add preprocess tile order mode to sparse global order reader, on by default

---
TYPE: FEATURE | BUG | IMPROVEMENT
DESC: sparse global order reader determine global order of result tiles

.gitlab-ci.yml

@@ -18,4 +18,4 @@ trigger_pipeline:
     TILEDB_REF: ${CI_COMMIT_REF_NAME}
   trigger:
     project: tiledb-inc/tiledb-internal
-    strategy: depend    
+    strategy: depend

CMakeLists.txt

@@ -447,6 +447,7 @@ target_include_directories(local_install INTERFACE ${CMAKE_SOURCE_DIR})
 enable_testing()
 
 if(TILEDB_TESTS)
+  find_package(rapidcheck CONFIG REQUIRED)
   # Add custom Catch2 entry point that suppresses message boxes on debug assertion
   # failures on Windows CI.
   find_package(Catch2 REQUIRED)

scripts/find_heap_api_violations.py

@@ -104,7 +104,7 @@
     "zstd_compressor.h": ["std::unique_ptr<ZSTD_DCtx, decltype(&ZSTD_freeDCtx)> ctx_;", "std::unique_ptr<ZSTD_CCtx, decltype(&ZSTD_freeCCtx)> ctx_;"],
     "posix.cc": ["std::unique_ptr<DIR, UniqueDIRDeleter>", "static std::unique_ptr<char, decltype(&free)> cwd_(getcwd(nullptr, 0), free);"],
     "curl.h": ["std::unique_ptr<CURL, decltype(&curl_easy_cleanup)>"],
-    "pmr.h": ["std::unique_ptr", "unique_ptr<Tp> make_unique("],
+    "pmr.h": ["std::unique_ptr", "unique_ptr<Tp> make_unique(", "unique_ptr<Tp> emplace_unique("],
 }
 
 

test/CMakeLists.txt

@@ -236,6 +236,7 @@ target_link_libraries(tiledb_unit
     Catch2::Catch2
     tiledb_test_support_lib
     configuration_definitions
+    rapidcheck
 )
 
 target_link_libraries(tiledb_unit PRIVATE $<BUILD_INTERFACE:common>)
@@ -373,9 +374,7 @@ endif()
 
 # CI tests
 add_subdirectory(ci)
-if(WIN32)
   add_subdirectory(performance)
-endif()
 
 add_custom_target(
   check-package

test/performance/CMakeLists.txt

@@ -26,6 +26,8 @@
 # THE SOFTWARE.
 #
 
+if(WIN32)
+
 # These options not exposed in bootstrap script.
 option(TILEDB_TESTS_AWS_S3_CONFIG "Use an S3 config appropriate for AWS in tests" OFF)
 
@@ -94,3 +96,24 @@ if (${CMAKE_SYSTEM_NAME} MATCHES "Linux" AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU
       LINK_FLAGS "-Wl,--no-as-needed -ldl"
     )
 endif()
+
+endif()
+
+
+add_executable(
+  tiledb_submit_a_b EXCLUDE_FROM_ALL
+  $<TARGET_OBJECTS:TILEDB_CORE_OBJECTS>
+  "tiledb_submit_a_b.cc"
+)
+
+target_include_directories(
+  tiledb_submit_a_b BEFORE PRIVATE
+    ${TILEDB_CORE_INCLUDE_DIR}
+    ${TILEDB_EXPORT_HEADER_DIR}
+)
+
+target_link_libraries(tiledb_submit_a_b
+  PUBLIC
+    TILEDB_CORE_OBJECTS_ILIB
+    tiledb_test_support_lib
+)