diff --git a/be/CMakeLists.txt b/be/CMakeLists.txt
index 0741d0a17dd551..b2f1b8129ce89c 100644
--- a/be/CMakeLists.txt
+++ b/be/CMakeLists.txt
@@ -409,7 +409,14 @@ if (COMPILER_CLANG)
                         -Wthread-safety)
     add_compile_options(-Wno-gnu-statement-expression
                         -Wno-implicit-float-conversion
-                        -Wno-sign-conversion)
+                        -Wno-sign-conversion
+                        # libstdc++-15 puts `#pragma GCC unroll` on loops in its own
+                        # headers (bits/stl_algobase.h); when a sanitizer/coverage
+                        # build compiles at low optimization the unroller punts and
+                        # clang's -Wpass-failed turns that missed *hint* into an
+                        # -Werror failure. The diagnostic carries no correctness
+                        # signal, so drop it.
+                        -Wno-pass-failed)
     if (USE_LIBCPP)
         add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-stdlib=libc++>)
         add_definitions(-DUSE_LIBCPP)
@@ -981,6 +988,24 @@ function(pch_reuse target)
     endif()
 endfunction(pch_reuse target)
 
+# Every unity opt-out goes through this helper. set_source_files_properties()
+# silently ignores paths that do not exist, so a skip entry going stale after
+# a rename or move would quietly re-join its unity batch -- slower builds, or
+# new file-scope clashes, with no diagnostic anywhere. Fail the configure
+# instead. Entries under GENSRC_DIR are exempt: they are emitted by
+# add_custom_command at build time (wkt_lex.l.cpp, the generated
+# thrift/protobuf sources) and legitimately do not exist on a fresh configure.
+function(doris_skip_unity_inclusion)
+    foreach(entry IN LISTS ARGN)
+        string(FIND "${entry}" "${GENSRC_DIR}" gensrc_prefix_pos)
+        if(NOT gensrc_prefix_pos EQUAL 0 AND NOT EXISTS "${entry}")
+            message(FATAL_ERROR
+                "unity skip entry does not exist (renamed or moved?): ${entry}")
+        endif()
+    endforeach()
+    set_source_files_properties(${ARGN} PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+endfunction()
+
 
 add_subdirectory(${SRC_DIR}/agent)
 add_subdirectory(${SRC_DIR}/common)
diff --git a/be/src/agent/CMakeLists.txt b/be/src/agent/CMakeLists.txt
index c680c41d6b209c..423b9c5e1fbbdf 100644
--- a/be/src/agent/CMakeLists.txt
+++ b/be/src/agent/CMakeLists.txt
@@ -29,4 +29,9 @@ endif()
 
 add_library(Agent STATIC ${AGENT_SOURCES})
 
-pch_reuse(Agent)
\ No newline at end of file
+pch_reuse(Agent)
+
+# Unity build: nine homogeneous task-worker glue TUs sharing the agent header
+# closure; batch 0 merges them into a single unity TU.
+set_target_properties(Agent PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                       UNITY_BUILD_BATCH_SIZE 0)
\ No newline at end of file
diff --git a/be/src/agent/cgroup_cpu_ctl.h b/be/src/agent/cgroup_cpu_ctl.h
index 262be57b499dee..3e1933fb4d6492 100644
--- a/be/src/agent/cgroup_cpu_ctl.h
+++ b/be/src/agent/cgroup_cpu_ctl.h
@@ -15,6 +15,8 @@
 // specific language governing permissions and limitations
 // under the License.
 
+#pragma once
+
 #include <fcntl.h>
 #include <sys/stat.h>
 #include <sys/types.h>
diff --git a/be/src/cloud/CMakeLists.txt b/be/src/cloud/CMakeLists.txt
index dbe8160c2f8d8c..e81a097b1ff676 100644
--- a/be/src/cloud/CMakeLists.txt
+++ b/be/src/cloud/CMakeLists.txt
@@ -22,3 +22,15 @@ file(GLOB_RECURSE SRC_FILES CONFIGURE_DEPENDS *.cpp)
 add_library(Cloud STATIC ${SRC_FILES})
 
 pch_reuse(Cloud)
+
+# Unity build: homogeneous cloud glue TUs sharing the CloudStorageEngine /
+# meta-service RPC header closure. Batch 12 bounds jumbo-TU size and memory.
+# The two http action TUs stay individual: sibling TUs' file-scope
+# `using namespace ErrorCode` makes the unscoped HttpStatus enum constants
+# (OK, NOT_FOUND, ...) shadow ErrorCode variables under -Wshadow -Werror.
+set(CLOUD_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/cloud_compaction_action.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/injection_point_action.cpp)
+doris_skip_unity_inclusion(${CLOUD_UNITY_SKIP})
+set_target_properties(Cloud PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                       UNITY_BUILD_BATCH_SIZE 12)
diff --git a/be/src/cloud/cloud_internal_service.cpp b/be/src/cloud/cloud_internal_service.cpp
index 7dca1d1517f680..c903b49961fc15 100644
--- a/be/src/cloud/cloud_internal_service.cpp
+++ b/be/src/cloud/cloud_internal_service.cpp
@@ -925,12 +925,7 @@ bvar::Adder<uint64_t> g_file_cache_warm_up_rowset_wait_for_compaction_num(
 bvar::Adder<uint64_t> g_file_cache_warm_up_rowset_wait_for_compaction_timeout_num(
         "file_cache_warm_up_rowset_wait_for_compaction_timeout_num");
 
-// Per-job windowed metrics for target BE
-// bvar::Window enforces MAX_SECONDS_LIMIT = 3600, so the longest window is 1h.
-static constexpr int WINDOW_5M = 300;
-static constexpr int WINDOW_30M = 1800;
-static constexpr int WINDOW_1H = 3600;
-
+// Per-job windowed metrics for target BE (window spans shared via bvar_windowed_adder.h)
 MBvarWindowedAdder g_warmup_ed_finish_segment_num("warmup_ed_finish_segment_num", {"job_id"},
                                                   {WINDOW_5M, WINDOW_30M, WINDOW_1H}, false);
 MBvarWindowedAdder g_warmup_ed_finish_segment_size("warmup_ed_finish_segment_size", {"job_id"},
diff --git a/be/src/cloud/cloud_warm_up_manager.cpp b/be/src/cloud/cloud_warm_up_manager.cpp
index 4590d96fb4ca80..68db44f0b1cdf6 100644
--- a/be/src/cloud/cloud_warm_up_manager.cpp
+++ b/be/src/cloud/cloud_warm_up_manager.cpp
@@ -118,12 +118,7 @@ bvar::Adder<int64_t> g_file_cache_warm_up_job_num("file_cache_warm_up_job_num");
 bvar::LatencyRecorder g_file_cache_warm_up_rowset_wait_for_compaction_latency(
         "file_cache_warm_up_rowset_wait_for_compaction_latency");
 
-// Per-job windowed metrics for source BE
-// bvar::Window enforces MAX_SECONDS_LIMIT = 3600, so the longest window is 1h.
-static constexpr int WINDOW_5M = 300;
-static constexpr int WINDOW_30M = 1800;
-static constexpr int WINDOW_1H = 3600;
-
+// Per-job windowed metrics for source BE (window spans shared via bvar_windowed_adder.h)
 MBvarWindowedAdder g_warmup_ed_requested_segment_num("warmup_ed_requested_segment_num", {"job_id"},
                                                      {WINDOW_5M, WINDOW_30M, WINDOW_1H}, false);
 MBvarWindowedAdder g_warmup_ed_requested_segment_size("warmup_ed_requested_segment_size",
diff --git a/be/src/common/CMakeLists.txt b/be/src/common/CMakeLists.txt
index b1e51e8f30312e..949af9c9723f82 100644
--- a/be/src/common/CMakeLists.txt
+++ b/be/src/common/CMakeLists.txt
@@ -23,6 +23,22 @@ add_library(Common STATIC ${SRC_FILES})
 
 pch_reuse(Common)
 
+# Unity build for the small common TUs. Seven files leak file-scope macros
+# (SET_FIELD/UPDATE_FIELD, MEMORY_SANITIZER, DW_* tables, the DEFINE_*_METRIC
+# families, USE_PHDR_CACHE/UNW_LOCAL_ONLY) and stay individual -- that set
+# also keeps the three same-name metric-hook statics (_s_hook_name) apart.
+set(COMMON_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/config.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/demangle.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/dwarf.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/metrics/doris_metrics.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/metrics/jvm_metrics.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/metrics/system_metrics.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/phdr_cache.cpp)
+doris_skip_unity_inclusion(${COMMON_UNITY_SKIP})
+set_target_properties(Common PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                        UNITY_BUILD_BATCH_SIZE 0)
+
 # Generate env_config.h according to env_config.h.in
 configure_file(${CMAKE_CURRENT_SOURCE_DIR}/env_config.h.in ${GENSRC_DIR}/common/env_config.h)
 target_include_directories(Common PUBLIC ${GENSRC_DIR}/common/)
diff --git a/be/src/core/CMakeLists.txt b/be/src/core/CMakeLists.txt
index 348a375eef0c1d..d5faec6c40cfa2 100644
--- a/be/src/core/CMakeLists.txt
+++ b/be/src/core/CMakeLists.txt
@@ -32,3 +32,41 @@ pch_reuse(Core)
 # instantiations below without definitions. Opt this single TU out of the PCH.
 set_source_files_properties(${CMAKE_CURRENT_SOURCE_DIR}/wide_integer_from_double.cpp
                             PROPERTIES SKIP_PRECOMPILE_HEADERS ON)
+
+# Unity build: column/data_type/serde glue TUs share the core header closure.
+# Opt-outs:
+# - column/column_vector.cpp: the heaviest template-instantiation TU in the
+#   target (~40s alone); it would dominate any batch it joins
+# - three macro-leak files (WRITE_INTEGRAL_COLUMN_TO_ORC,
+#   DECLARE_DECIMAL_COMPARISON, DELARE_DATE_ADD_INTERVAL)
+# - wide_integer_from_double.cpp: macro-leak, and it is the PCH-skipped
+#   explicit-instantiation TU above -- it must keep its own compile
+# - data_type_serde.cpp: holds member-level explicit instantiations of serde
+#   members whose class-level explicit instantiations live in the per-type
+#   serde TUs; merging both forms into one TU is a duplicate explicit
+#   instantiation
+# - the value/variant/ family (9s of slot time in total): copied file-local
+#   helpers with diverging signatures (require_bytes, write_unsigned, ...) --
+#   not worth renaming for the gain
+# - two files that tests compile a second time by #including the .cpp
+#   (column/column_variant.cpp, data_type/convert_field_to_type.cpp): the
+#   test object must shadow a never-pulled archive member, but a unity batch
+#   is pulled in for its siblings and the linker sees a duplicate definition
+set(CORE_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/column/column_vector.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/column/column_variant.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/data_type/convert_field_to_type.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/data_type_serde/data_type_number_serde.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/data_type_serde/data_type_serde.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/field.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/vdatetime_value.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_batch_builder.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_canonical.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_field.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_metadata.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_scalar.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/value/variant/variant_value.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/wide_integer_from_double.cpp)
+doris_skip_unity_inclusion(${CORE_UNITY_SKIP})
+set_target_properties(Core PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                      UNITY_BUILD_BATCH_SIZE 12)
diff --git a/be/src/core/data_type_serde/complex_type_deserialize_util.h b/be/src/core/data_type_serde/complex_type_deserialize_util.h
index ff76d64dcd6542..b1320b4b073fc3 100644
--- a/be/src/core/data_type_serde/complex_type_deserialize_util.h
+++ b/be/src/core/data_type_serde/complex_type_deserialize_util.h
@@ -15,6 +15,8 @@
 // specific language governing permissions and limitations
 // under the License.
 
+#pragma once
+
 #include "core/data_type_serde/data_type_serde.h"
 #include "core/string_ref.h"
 
diff --git a/be/src/core/value/vdatetime_value.h b/be/src/core/value/vdatetime_value.h
index 7419b02175575b..da0bba19bcb1c3 100644
--- a/be/src/core/value/vdatetime_value.h
+++ b/be/src/core/value/vdatetime_value.h
@@ -219,7 +219,9 @@ static constexpr uint32_t DATEV2_YEAR_WIDTH = 23;
 static constexpr uint32_t DATETIMEV2_YEAR_WIDTH = 18;
 static constexpr uint32_t DATETIMEV2_MONTH_WIDTH = 4;
 
-static RE2 time_zone_offset_format_reg(R"(^[+-]{1}\d{2}\:\d{2}$)");
+// One shared instance program-wide (was `static`, i.e. one copy constructed
+// per including TU). Visiting is thread-safe.
+inline RE2 time_zone_offset_format_reg(R"(^[+-]{1}\d{2}\:\d{2}$)");
 
 uint8_t mysql_week_mode(uint32_t mode);
 
diff --git a/be/src/exec/CMakeLists.txt b/be/src/exec/CMakeLists.txt
index 7ec053789111a1..7ccbd60b5ea37b 100644
--- a/be/src/exec/CMakeLists.txt
+++ b/be/src/exec/CMakeLists.txt
@@ -52,7 +52,7 @@ set(EXEC_UNITY_SKIP
     ${CMAKE_CURRENT_SOURCE_DIR}/operator/partitioned_aggregation_sink_operator.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/operator/scan_operator.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/sink/writer/vfile_result_writer.cpp)
-set_source_files_properties(${EXEC_UNITY_SKIP} PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+doris_skip_unity_inclusion(${EXEC_UNITY_SKIP})
 set_target_properties(Exec PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
                                       UNITY_BUILD_BATCH_SIZE 12)
 
diff --git a/be/src/exprs/CMakeLists.txt b/be/src/exprs/CMakeLists.txt
index b7bf6916a901ae..1940ba6d8e4657 100644
--- a/be/src/exprs/CMakeLists.txt
+++ b/be/src/exprs/CMakeLists.txt
@@ -97,7 +97,7 @@ set(EXPRS_UNITY_SKIP
     ${CMAKE_CURRENT_SOURCE_DIR}/math_functions.cpp
     ${GENSRC_DIR}/geo/wkt_lex.l.cpp
     ${GENSRC_DIR}/geo/wkt_yacc.y.cpp)
-set_source_files_properties(${EXPRS_UNITY_SKIP} PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+doris_skip_unity_inclusion(${EXPRS_UNITY_SKIP})
 set_target_properties(Exprs PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
                                        UNITY_BUILD_BATCH_SIZE 8)
 
diff --git a/be/src/format/CMakeLists.txt b/be/src/format/CMakeLists.txt
index f73ef2dbfaab7c..94b7c6bb6d1b4c 100644
--- a/be/src/format/CMakeLists.txt
+++ b/be/src/format/CMakeLists.txt
@@ -29,3 +29,18 @@ list(APPEND SRC_FILES ${FORMAT_V2_SRC_FILES})
 add_library(Format STATIC ${SRC_FILES})
 
 pch_reuse(Format)
+
+# Unity build scoped to format_v2/: the v2 readers are the maintained line and
+# share one reader/expr header closure. The format/ (v1) tree is slated for
+# removal in a future version, so it is not worth the v1/v2 twin-symbol
+# deduplication a joint unity would require: every v1 file stays individual
+# until the tree is deleted. Within v2, adbc_reader.cpp stays individual
+# because its file-scope RETURN_IF_ADBC_ERROR macro must not leak into unity
+# siblings. Batch 8 bounds jumbo-TU size and memory.
+set(FORMAT_UNITY_SKIP ${SRC_FILES})
+list(FILTER FORMAT_UNITY_SKIP EXCLUDE REGEX ".*/format_v2/.*")
+list(APPEND FORMAT_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/../format_v2/table/adbc_reader.cpp)
+doris_skip_unity_inclusion(${FORMAT_UNITY_SKIP})
+set_target_properties(Format PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                        UNITY_BUILD_BATCH_SIZE 8)
diff --git a/be/src/format_v2/column_mapper.cpp b/be/src/format_v2/column_mapper.cpp
index 9f433cc853f61a..e98a0cde9a1e9f 100644
--- a/be/src/format_v2/column_mapper.cpp
+++ b/be/src/format_v2/column_mapper.cpp
@@ -349,10 +349,6 @@ static VExprSPtr create_file_slot_ref(const VSlotRef& slot_ref,
     return ref;
 }
 
-static bool is_cast_expr(const VExprSPtr& expr) {
-    return dynamic_cast<const Cast*>(expr.get()) != nullptr;
-}
-
 static bool is_binary_comparison_predicate(const VExprSPtr& expr) {
     if (expr == nullptr || expr->get_num_children() != 2 ||
         (expr->node_type() != TExprNodeType::BINARY_PRED &&
diff --git a/be/src/format_v2/column_mapper_nested.cpp b/be/src/format_v2/column_mapper_nested.cpp
index fefcdf26f12e1d..9d1980c357984f 100644
--- a/be/src/format_v2/column_mapper_nested.cpp
+++ b/be/src/format_v2/column_mapper_nested.cpp
@@ -37,10 +37,6 @@ namespace doris::format {
 
 namespace {
 
-static bool is_cast_expr(const VExprSPtr& expr) {
-    return dynamic_cast<const Cast*>(expr.get()) != nullptr;
-}
-
 static bool is_signed_integer_type(PrimitiveType type) {
     switch (type) {
     case TYPE_TINYINT:
diff --git a/be/src/format_v2/expr/cast.h b/be/src/format_v2/expr/cast.h
index 22604455e50099..18b3bea676c4fc 100644
--- a/be/src/format_v2/expr/cast.h
+++ b/be/src/format_v2/expr/cast.h
@@ -68,4 +68,9 @@ class Cast final : public VExpr {
     std::string _expr_name;
     FunctionBasePtr _function;
 };
+
+inline bool is_cast_expr(const VExprSPtr& expr) {
+    return dynamic_cast<const Cast*>(expr.get()) != nullptr;
+}
+
 } // namespace doris::format
diff --git a/be/src/format_v2/parquet/parquet_scan.cpp b/be/src/format_v2/parquet/parquet_scan.cpp
index a9154021999a4f..115ea2e373d23f 100644
--- a/be/src/format_v2/parquet/parquet_scan.cpp
+++ b/be/src/format_v2/parquet/parquet_scan.cpp
@@ -49,6 +49,7 @@
 #include "format_v2/parquet/reader/native/column_chunk_reader.h"
 #include "format_v2/parquet/reader/native_column_reader.h"
 #include "format_v2/parquet/reader/row_position_column_reader.h"
+#include "format_v2/parquet/selection_vector.h" // count_range_rows
 #include "runtime/runtime_state.h"
 #include "util/defer_op.h"
 #include "util/time.h"
@@ -823,14 +824,6 @@ Status execute_batch_filters(const format::FileScanRequest& request, int64_t bat
 }
 
 namespace {
-int64_t count_range_rows(const std::vector<RowRange>& ranges) {
-    int64_t rows = 0;
-    for (const auto& range : ranges) {
-        rows += range.length;
-    }
-    return rows;
-}
-
 void append_intersection(const RowRange& left, const RowRange& right,
                          std::vector<RowRange>& result) {
     const int64_t start = std::max(left.start, right.start);
diff --git a/be/src/format_v2/parquet/parquet_statistics.cpp b/be/src/format_v2/parquet/parquet_statistics.cpp
index e6f42c59ce61ea..b67536fd02c225 100644
--- a/be/src/format_v2/parquet/parquet_statistics.cpp
+++ b/be/src/format_v2/parquet/parquet_statistics.cpp
@@ -42,6 +42,7 @@
 #include "format_v2/parquet/parquet_file_context.h"
 #include "format_v2/parquet/reader/native/block_split_bloom_filter.h"
 #include "format_v2/parquet/reader/native_column_reader.h"
+#include "format_v2/parquet/selection_vector.h" // count_range_rows
 #include "format_v2/timestamp_statistics.h"
 #include "runtime/runtime_profile.h"
 #include "storage/index/bloom_filter/bloom_filter.h"
@@ -989,14 +990,6 @@ std::vector<RowRange> intersect_ranges(const std::vector<RowRange>& left,
     return result;
 }
 
-int64_t count_range_rows(const std::vector<RowRange>& ranges) {
-    int64_t rows = 0;
-    for (const auto& range : ranges) {
-        rows += range.length;
-    }
-    return rows;
-}
-
 void append_row_range(const RowRange& range, std::vector<RowRange>* ranges) {
     if (range.length == 0) {
         return;
diff --git a/be/src/format_v2/parquet/selection_vector.h b/be/src/format_v2/parquet/selection_vector.h
index 99c854d8b4e896..a9c8c040e9302e 100644
--- a/be/src/format_v2/parquet/selection_vector.h
+++ b/be/src/format_v2/parquet/selection_vector.h
@@ -31,6 +31,14 @@ struct RowRange {
     int64_t length = 0;
 };
 
+inline int64_t count_range_rows(const std::vector<RowRange>& ranges) {
+    int64_t rows = 0;
+    for (const auto& range : ranges) {
+        rows += range.length;
+    }
+    return rows;
+}
+
 struct ParquetPageSkipPlan {
     int leaf_column_id = -1;
     // Page ordinal is the data-page ordinal in the column chunk. It intentionally excludes
diff --git a/be/src/format_v2/table/schema_history_util.cpp b/be/src/format_v2/table/schema_history_util.cpp
index 96a466fa772837..a3dbad7c1e36ee 100644
--- a/be/src/format_v2/table/schema_history_util.cpp
+++ b/be/src/format_v2/table/schema_history_util.cpp
@@ -27,13 +27,6 @@
 namespace doris::format {
 namespace {
 
-const schema::external::TField* get_field_ptr(const schema::external::TFieldPtr& field_ptr) {
-    if (!field_ptr.__isset.field_ptr || field_ptr.field_ptr == nullptr) {
-        return nullptr;
-    }
-    return field_ptr.field_ptr.get();
-}
-
 const schema::external::TField* find_child_field_by_name(
         const std::vector<schema::external::TFieldPtr>& fields, const std::string& name) {
     for (const auto& field_ptr : fields) {
diff --git a/be/src/format_v2/table/schema_history_util.h b/be/src/format_v2/table/schema_history_util.h
index 3c4a80b5d4c975..49704c5fb98468 100644
--- a/be/src/format_v2/table/schema_history_util.h
+++ b/be/src/format_v2/table/schema_history_util.h
@@ -27,6 +27,14 @@
 
 namespace doris::format {
 
+// Unwrap a thrift TFieldPtr, returning nullptr when the pointer is unset.
+inline const schema::external::TField* get_field_ptr(const schema::external::TFieldPtr& field_ptr) {
+    if (!field_ptr.__isset.field_ptr || field_ptr.field_ptr == nullptr) {
+        return nullptr;
+    }
+    return field_ptr.field_ptr.get();
+}
+
 const schema::external::TSchema* find_history_schema(const TFileScanRangeParams* params,
                                                      int64_t schema_id);
 
diff --git a/be/src/format_v2/table_reader.cpp b/be/src/format_v2/table_reader.cpp
index 40ffe6cb86eff0..b531ff9cf2322a 100644
--- a/be/src/format_v2/table_reader.cpp
+++ b/be/src/format_v2/table_reader.cpp
@@ -50,6 +50,7 @@
 #include "format_v2/native/native_reader.h"
 #include "format_v2/orc/orc_reader.h"
 #include "format_v2/parquet/parquet_reader.h"
+#include "format_v2/table/schema_history_util.h" // get_field_ptr
 #include "runtime/file_scan_profile.h"
 #include "storage/segment/condition_cache.h"
 #include "util/debug_points.h"
@@ -140,13 +141,6 @@ std::string partition_values_debug_string(const std::map<std::string, Field>& pa
     return out.str();
 }
 
-const schema::external::TField* get_field_ptr(const schema::external::TFieldPtr& field_ptr) {
-    if (!field_ptr.__isset.field_ptr || field_ptr.field_ptr == nullptr) {
-        return nullptr;
-    }
-    return field_ptr.field_ptr.get();
-}
-
 const schema::external::TField* find_external_field_by_id(
         const schema::external::TStructField* root, int32_t field_id) {
     if (root == nullptr || !root->__isset.fields) {
diff --git a/be/src/gen_cpp/CMakeLists.txt b/be/src/gen_cpp/CMakeLists.txt
index c4463f75520cb0..abbe3877079d9f 100644
--- a/be/src/gen_cpp/CMakeLists.txt
+++ b/be/src/gen_cpp/CMakeLists.txt
@@ -27,10 +27,26 @@ list(FILTER SRC_FILES EXCLUDE REGEX "_server\\.skeleton\\.cpp$")
 
 add_compile_options(-Wno-return-type)
 
-# keep everything in one library, the object files reference 
+# keep everything in one library, the object files reference
 # each other
 add_library(DorisGen STATIC ${SRC_FILES})
 
+# Unity build for the thrift-generated .cpp half (pre-scanned clean: no
+# file-scope macros or duplicate statics). Generated sources cannot be edited,
+# so anything that clashes can only be opted out, never fixed:
+# - every protobuf .cc stays individual: protoc emits the same file-scope
+#   statics (schemas[], file_default_instances[]) in each one, a structural
+#   clash for any merged pair
+# - the two heaviest thrift TUs stay individual so no batch inherits them
+set(GEN_UNITY_SKIP ${SRC_FILES})
+list(FILTER GEN_UNITY_SKIP INCLUDE REGEX "\\.cc$")
+list(APPEND GEN_UNITY_SKIP
+    ${GEN_CPP_DIR}/FrontendService_types.cpp
+    ${GEN_CPP_DIR}/PlanNodes_types.cpp)
+doris_skip_unity_inclusion(${GEN_UNITY_SKIP})
+set_target_properties(DorisGen PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                          UNITY_BUILD_BATCH_SIZE 8)
+
 if (COMPILER_CLANG)
     target_compile_options(DorisGen PRIVATE
     -Wno-implicit-int-conversion
diff --git a/be/src/io/CMakeLists.txt b/be/src/io/CMakeLists.txt
index 56c2eeb94a3819..3026c82e11fff1 100644
--- a/be/src/io/CMakeLists.txt
+++ b/be/src/io/CMakeLists.txt
@@ -34,11 +34,38 @@ if(ENABLE_TDE)
 endif()
 
 list(REMOVE_ITEM IO_FILES "${CMAKE_CURRENT_SOURCE_DIR}/fs/benchmark/fs_benchmark_tool.cpp")
+# Standalone tool with its own main(); must not be an archive member of libIO.a,
+# or unity builds merge it into an object that the doris_be link actually pulls in.
+list(REMOVE_ITEM IO_FILES "${CMAKE_CURRENT_SOURCE_DIR}/cache/file_cache_lru_tool.cpp")
 
 add_library(IO STATIC ${IO_FILES})
 
 pch_reuse(IO)
 
+# Unity build for the io glue (small fs/cache TUs, parse-cost dominated).
+# Opt-outs:
+# - three filesystem TUs whose CHECK_*_CLIENT file-scope macros must not
+#   leak into unity siblings
+# - http_file_system.cpp: sibling TUs' file-scope `using namespace ErrorCode`
+#   makes the unscoped HttpStatus enum constants shadow ErrorCode variables
+#   under -Wshadow -Werror (same pattern as the cloud http actions)
+# - benchmark sources and ENABLE_TDE module swaps stay individual
+set(IO_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/fs/broker_file_system.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/fs/hdfs_file_system.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/fs/s3_file_system.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/fs/http_file_system.cpp)
+file(GLOB_RECURSE IO_UNITY_BENCH_SRCS CONFIGURE_DEPENDS
+    ${CMAKE_CURRENT_SOURCE_DIR}/fs/benchmark/*.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/tools/*.cpp)
+list(APPEND IO_UNITY_SKIP ${IO_UNITY_BENCH_SRCS})
+if (ENABLE_TDE)
+    list(APPEND IO_UNITY_SKIP ${EXTRA_SOURCES})
+endif()
+doris_skip_unity_inclusion(${IO_UNITY_SKIP})
+set_target_properties(IO PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                    UNITY_BUILD_BATCH_SIZE 16)
+
 if (${BUILD_FS_BENCHMARK} STREQUAL "ON")
     add_executable(fs_benchmark_tool
         fs/benchmark/fs_benchmark_tool.cpp
diff --git a/be/src/io/cache/block_file_cache.cpp b/be/src/io/cache/block_file_cache.cpp
index b44679a94fa150..c4e7ef47853047 100644
--- a/be/src/io/cache/block_file_cache.cpp
+++ b/be/src/io/cache/block_file_cache.cpp
@@ -66,10 +66,6 @@ namespace {
 constexpr std::array<FileCacheType, 4> LRU_LOG_REPLAY_TYPES = {
         FileCacheType::TTL, FileCacheType::INDEX, FileCacheType::NORMAL, FileCacheType::DISPOSABLE};
 
-size_t file_cache_type_index(FileCacheType type) {
-    return static_cast<size_t>(type);
-}
-
 } // namespace
 
 // Insert a block pointer into one shard while swallowing allocation failures.
diff --git a/be/src/io/cache/file_cache_common.h b/be/src/io/cache/file_cache_common.h
index 8b52af8d161b9a..3f8ca5e1ff0161 100644
--- a/be/src/io/cache/file_cache_common.h
+++ b/be/src/io/cache/file_cache_common.h
@@ -43,6 +43,11 @@ enum FileCacheType {
     DISPOSABLE = 0,
     TTL = 3,
 };
+
+inline size_t file_cache_type_index(FileCacheType type) {
+    return static_cast<size_t>(type);
+}
+
 std::string cache_type_to_surfix(FileCacheType type);
 FileCacheType surfix_to_cache_type(const std::string& str);
 
diff --git a/be/src/io/cache/lru_queue_recorder.cpp b/be/src/io/cache/lru_queue_recorder.cpp
index 314a444e232e9e..82edc5f6070003 100644
--- a/be/src/io/cache/lru_queue_recorder.cpp
+++ b/be/src/io/cache/lru_queue_recorder.cpp
@@ -24,13 +24,7 @@
 
 namespace doris::io {
 
-namespace {
-
-size_t file_cache_type_index(FileCacheType type) {
-    return static_cast<size_t>(type);
-}
-
-} // namespace
+namespace {} // namespace
 
 void LRUQueueRecorder::record_queue_event(FileCacheType type, CacheLRULogType log_type,
                                           const UInt128Wrapper hash, const size_t offset,
diff --git a/be/src/load/CMakeLists.txt b/be/src/load/CMakeLists.txt
index 5c2e970f6d9d8d..5faf360734fdef 100644
--- a/be/src/load/CMakeLists.txt
+++ b/be/src/load/CMakeLists.txt
@@ -25,3 +25,13 @@ file(GLOB_RECURSE SRC_FILES CONFIGURE_DEPENDS *.cpp)
 add_library(Load STATIC ${SRC_FILES})
 
 pch_reuse(Load)
+
+# Unity build: load channel/delta-writer/group-commit glue sharing one header
+# closure. Two files leak file-scope macros (UNKNOWN_ID_FOR_TEST,
+# HANDLE_ERROR/HANDLE_MULTI_TABLE_ERROR) and stay individual.
+set(LOAD_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/channel/load_stream.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/routine_load/routine_load_task_executor.cpp)
+doris_skip_unity_inclusion(${LOAD_UNITY_SKIP})
+set_target_properties(Load PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                      UNITY_BUILD_BATCH_SIZE 12)
diff --git a/be/src/runtime/CMakeLists.txt b/be/src/runtime/CMakeLists.txt
index 7d54f9bd76bbff..0d2f3b567d24f6 100644
--- a/be/src/runtime/CMakeLists.txt
+++ b/be/src/runtime/CMakeLists.txt
@@ -34,3 +34,17 @@ add_library(Runtime STATIC
 )
 
 pch_reuse(Runtime)
+
+# Unity build: runtime glue TUs share the exec-env/runtime-state header
+# closure. memory/jemalloc_hook.cpp (a Linux-only member, absent on macOS)
+# stays individual: it defines malloc-replacement hooks and leaks
+# __THROW/ALIAS macros. descriptors.cpp stays individual: its test compiles
+# it a second time by #including the .cpp, and the test object must shadow a
+# never-pulled archive member (a unity batch would get pulled in for its
+# siblings and hand the linker a duplicate definition).
+set(RUNTIME_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/memory/jemalloc_hook.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/descriptors.cpp)
+doris_skip_unity_inclusion(${RUNTIME_UNITY_SKIP})
+set_target_properties(Runtime PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                         UNITY_BUILD_BATCH_SIZE 12)
diff --git a/be/src/runtime/runtime_profile.h b/be/src/runtime/runtime_profile.h
index 54f2e80c89e32d..662a92db11db2e 100644
--- a/be/src/runtime/runtime_profile.h
+++ b/be/src/runtime/runtime_profile.h
@@ -746,9 +746,6 @@ class RuntimeProfile {
     int64_t _metadata;
     bool _is_set_metadata = false;
 
-    bool _is_sink = false;
-    bool _is_set_sink = false;
-
     // The timestamp when the profile was modified, make sure the update is up to date.
     time_t _timestamp;
 
diff --git a/be/src/service/CMakeLists.txt b/be/src/service/CMakeLists.txt
index 28985a2e031a7f..f07b896d4a45f0 100644
--- a/be/src/service/CMakeLists.txt
+++ b/be/src/service/CMakeLists.txt
@@ -37,16 +37,20 @@ add_library(Service STATIC ${SRC_FILES})
 pch_reuse(Service)
 
 # Unity build scoped to the http glue: ~60 small handlers re-parsing the same
-# header closure once per file. Non-http sources (service entry points,
-# arrow_flight) are heterogeneous heavy TUs that gain nothing from merging.
-# http_parser.cpp and be_thread_stack_action.cpp stay individual because their
-# file-scope macros (CR/LF, UNW_LOCAL_ONLY) must not leak into unity siblings.
+# header closure once per file, merged into one unity TU (batch 0). Widening
+# to the whole target at batch 8 was tried and measured slower: splitting the
+# single http jumbo into 8-file batches repays the shared-closure parse per
+# batch, and the heterogeneous non-http TUs gain too little to cover it
+# (service segment 104s -> 132s slot time). So non-http sources stay
+# individual. http_parser.cpp and be_thread_stack_action.cpp stay individual
+# because their file-scope macros (CR/LF, UNW_LOCAL_ONLY) must not leak into
+# unity siblings.
 set(SERVICE_UNITY_SKIP ${SRC_FILES})
 list(FILTER SERVICE_UNITY_SKIP EXCLUDE REGEX ".*/service/http/.*")
 list(APPEND SERVICE_UNITY_SKIP
     ${CMAKE_CURRENT_SOURCE_DIR}/http/http_parser.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/http/action/be_thread_stack_action.cpp)
-set_source_files_properties(${SERVICE_UNITY_SKIP} PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+doris_skip_unity_inclusion(${SERVICE_UNITY_SKIP})
 set_target_properties(Service PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
                                          UNITY_BUILD_BATCH_SIZE 0)
 
diff --git a/be/src/storage/CMakeLists.txt b/be/src/storage/CMakeLists.txt
index 888b16ebe32fc7..21e82183f58dff 100644
--- a/be/src/storage/CMakeLists.txt
+++ b/be/src/storage/CMakeLists.txt
@@ -46,19 +46,36 @@ endif()
 
 pch_reuse(Storage)
 
-# Unity build scoped to storage/index/: 115 homogeneous reader/writer/iterator
-# TUs sharing one (CLucene-heavy) header closure. The rest of Storage keeps
-# compiling individually. Batch 32 bounds jumbo-TU size and memory. Four index
-# files stay individual because their file-scope macros (CL_MAX_PATH and
-# friends, IS_CHINESE_CHAR, APPLY_FOR_PRIMITITYPE) must not leak into unity
-# siblings.
-set(STORAGE_UNITY_SKIP ${SRC_FILES})
-list(FILTER STORAGE_UNITY_SKIP EXCLUDE REGEX ".*/storage/index/.*")
-list(APPEND STORAGE_UNITY_SKIP
+# Unity build for the whole Storage target: segment/rowset/tablet/compaction/
+# index glue TUs re-parse the same storage header closure once per file, so
+# batching them cuts the repeated frontend work. Batch 12 bounds jumbo-TU size
+# and memory. Opted out:
+# - seven files whose file-scope macros (CL_MAX_PATH and friends,
+#   IS_CHINESE_CHAR, APPLY_FOR_PRIMITITYPE, RETURN_IF_NOT_EOF_AND_OK,
+#   FORMAT_*_ADD_JSON_NODE, RETURN_IF_ERROR_) must not leak into unity siblings
+# - the three heaviest template-instantiation TUs (predicate creators) which
+#   would dominate any batch they join
+# - compaction/collection_statistics.cpp: its test compiles it a second time
+#   by #including the .cpp; the test object must shadow a never-pulled archive
+#   member, but a unity batch is pulled in for its siblings and the linker
+#   sees a duplicate definition
+set(STORAGE_UNITY_SKIP
     ${CMAKE_CURRENT_SOURCE_DIR}/index/inverted/inverted_index_compound_reader.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/index/inverted/inverted_index_fs_directory.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/index/inverted/tokenizer/basic/basic_tokenizer.cpp
-    ${CMAKE_CURRENT_SOURCE_DIR}/index/zone_map/zone_map_index.cpp)
-set_source_files_properties(${STORAGE_UNITY_SKIP} PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+    ${CMAKE_CURRENT_SOURCE_DIR}/index/zone_map/zone_map_index.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/iterator/vcollect_iterator.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/tablet/tablet.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/task/engine_clone_task.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/predicate/predicate_creator_comparison.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/predicate/predicate_creator_in_list_in.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/predicate/predicate_creator_in_list_not_in.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/compaction/collection_statistics.cpp)
+if (ENABLE_VARIANT_NESTED_GROUP)
+    # Out-of-tree module sources swapped into this target: unity hygiene
+    # unaudited, keep them individual.
+    list(APPEND STORAGE_UNITY_SKIP ${VARIANT_NESTED_GROUP_SOURCES})
+endif()
+doris_skip_unity_inclusion(${STORAGE_UNITY_SKIP})
 set_target_properties(Storage PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
-                                         UNITY_BUILD_BATCH_SIZE 32)
+                                         UNITY_BUILD_BATCH_SIZE 12)
diff --git a/be/src/storage/index/ann/CMakeLists.txt b/be/src/storage/index/ann/CMakeLists.txt
index a30bfce9773d43..d671cdbf024678 100644
--- a/be/src/storage/index/ann/CMakeLists.txt
+++ b/be/src/storage/index/ann/CMakeLists.txt
@@ -30,3 +30,8 @@ target_link_libraries(ann_index PUBLIC faiss OpenMP::OpenMP_CXX)
 
 # Some header files from faiss are used by doris, they will break compile check.
 target_compile_options(ann_index PRIVATE -Wno-shadow-field)
+
+# Unity build: ten homogeneous TUs sharing the faiss header closure; batch 0
+# merges them into a single unity TU.
+set_target_properties(ann_index PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                           UNITY_BUILD_BATCH_SIZE 0)
diff --git a/be/src/storage/merger.cpp b/be/src/storage/merger.cpp
index d986b367288853..c8abc92e7f7dc3 100644
--- a/be/src/storage/merger.cpp
+++ b/be/src/storage/merger.cpp
@@ -121,8 +121,8 @@ Status Merger::vmerge_rowsets(BaseTabletSPtr tablet, ReaderType reader_type,
         auto tablet_state = tablet->tablet_state();
         if (tablet_state != TABLET_RUNNING && tablet_state != TABLET_NOTREADY) {
             tablet->clear_cache();
-            return Status::Error<INTERNAL_ERROR>("tablet {} is not used any more",
-                                                 tablet->tablet_id());
+            return Status::Error<ErrorCode::INTERNAL_ERROR>("tablet {} is not used any more",
+                                                            tablet->tablet_id());
         }
 
         // Read one block from block reader
@@ -144,8 +144,8 @@ Status Merger::vmerge_rowsets(BaseTabletSPtr tablet, ReaderType reader_type,
         block.clear_column_data();
     }
     if (ExecEnv::GetInstance()->storage_engine().stopped()) {
-        return Status::Error<INTERNAL_ERROR>("tablet {} failed to do compaction, engine stopped",
-                                             tablet->tablet_id());
+        return Status::Error<ErrorCode::INTERNAL_ERROR>(
+                "tablet {} failed to do compaction, engine stopped", tablet->tablet_id());
     }
 
     if (stats_output != nullptr) {
@@ -312,8 +312,8 @@ Status Merger::vertical_compact_one_group(
         auto tablet_state = tablet->tablet_state();
         if (tablet_state != TABLET_RUNNING && tablet_state != TABLET_NOTREADY) {
             tablet->clear_cache();
-            return Status::Error<INTERNAL_ERROR>("tablet {} is not used any more",
-                                                 tablet->tablet_id());
+            return Status::Error<ErrorCode::INTERNAL_ERROR>("tablet {} is not used any more",
+                                                            tablet->tablet_id());
         }
         // Read one block from block reader
         RETURN_NOT_OK_STATUS_WITH_WARN(reader.next_block_with_aggregation(&block, &eof),
@@ -335,8 +335,8 @@ Status Merger::vertical_compact_one_group(
         block.clear_column_data();
     }
     if (ExecEnv::GetInstance()->storage_engine().stopped()) {
-        return Status::Error<INTERNAL_ERROR>("tablet {} failed to do compaction, engine stopped",
-                                             tablet->tablet_id());
+        return Status::Error<ErrorCode::INTERNAL_ERROR>(
+                "tablet {} failed to do compaction, engine stopped", tablet->tablet_id());
     }
 
     if (stats_output != nullptr) {
@@ -391,8 +391,8 @@ Status Merger::vertical_compact_one_group(
         block.clear_column_data();
     }
     if (ExecEnv::GetInstance()->storage_engine().stopped()) {
-        return Status::Error<INTERNAL_ERROR>("tablet {} failed to do compaction, engine stopped",
-                                             tablet_id);
+        return Status::Error<ErrorCode::INTERNAL_ERROR>(
+                "tablet {} failed to do compaction, engine stopped", tablet_id);
     }
 
     if (stats_output != nullptr) {
diff --git a/be/src/storage/predicate/column_predicate.h b/be/src/storage/predicate/column_predicate.h
index 2bcc8359464ce9..1d55531eb5ed60 100644
--- a/be/src/storage/predicate/column_predicate.h
+++ b/be/src/storage/predicate/column_predicate.h
@@ -349,7 +349,8 @@ class ColumnPredicate : public std::enable_shared_from_this<ColumnPredicate> {
         if (_predicate_input_rows_counter == nullptr ||
             _predicate_filtered_rows_counter == nullptr ||
             _predicate_always_true_rows_counter == nullptr) {
-            throw Exception(INTERNAL_ERROR, "Predicate profile counters are not initialized");
+            throw Exception(ErrorCode::INTERNAL_ERROR,
+                            "Predicate profile counters are not initialized");
         }
         COUNTER_UPDATE(_predicate_input_rows_counter, input_rows);
         COUNTER_UPDATE(_predicate_filtered_rows_counter, filter_rows);
@@ -397,7 +398,7 @@ class ColumnPredicate : public std::enable_shared_from_this<ColumnPredicate> {
 protected:
     virtual bool _can_ignore() const { return _runtime_filter_id != -1; }
     virtual uint16_t _evaluate_inner(const IColumn& column, uint16_t* sel, uint16_t size) const {
-        throw Exception(INTERNAL_ERROR, "Not Implemented _evaluate_inner");
+        throw Exception(ErrorCode::INTERNAL_ERROR, "Not Implemented _evaluate_inner");
     }
 
     void reset_judge_selectivity() const { _rf_selectivity.reset_judge_selectivity(); }
diff --git a/be/src/storage/segment/common.h b/be/src/storage/segment/common.h
index 166aa9fca5812f..a35fef0a97e0f0 100644
--- a/be/src/storage/segment/common.h
+++ b/be/src/storage/segment/common.h
@@ -30,5 +30,10 @@ namespace segment_v2 {
 using rowid_t = uint32_t;
 using ordinal_t = uint64_t;
 
+// Magic footer bytes identifying a V2 segment file, written by the segment
+// writers and checked when parsing the footer.
+inline constexpr const char* k_segment_magic = "D0R1";
+inline constexpr uint32_t k_segment_magic_length = 4;
+
 } // namespace segment_v2
 } // namespace doris
diff --git a/be/src/storage/segment/external_col_meta_util.cpp b/be/src/storage/segment/external_col_meta_util.cpp
index 87bc31f1079c33..9bc4bfeba1352d 100644
--- a/be/src/storage/segment/external_col_meta_util.cpp
+++ b/be/src/storage/segment/external_col_meta_util.cpp
@@ -29,8 +29,8 @@ namespace doris::segment_v2 {
 
 namespace {
 
-io::IOContext create_index_io_context(const io::IOContext* source_io_ctx,
-                                      OlapReaderStatistics* stats) {
+io::IOContext create_ext_meta_io_context(const io::IOContext* source_io_ctx,
+                                         OlapReaderStatistics* stats) {
     io::IOContext io_ctx;
     if (source_io_ctx != nullptr) {
         io_ctx = *source_io_ctx;
@@ -145,7 +145,7 @@ Status ExternalColMetaUtil::read_col_meta(const io::FileReaderSPtr& file_reader,
     std::string buf;
     buf.resize(static_cast<size_t>(size));
     size_t meta_read = 0;
-    io::IOContext io_ctx = create_index_io_context(source_io_ctx, stats);
+    io::IOContext io_ctx = create_ext_meta_io_context(source_io_ctx, stats);
     RETURN_IF_ERROR(file_reader->read_at(pos, Slice(buf.data(), buf.size()), &meta_read, &io_ctx));
     if (meta_read != size) {
         return Status::Corruption("short read ColumnMetaPB: expect={}, actual={}", size, meta_read);
diff --git a/be/src/storage/segment/segment.cpp b/be/src/storage/segment/segment.cpp
index 9e58a76e0a16e1..be80f7ae530510 100644
--- a/be/src/storage/segment/segment.cpp
+++ b/be/src/storage/segment/segment.cpp
@@ -73,11 +73,11 @@
 #include "storage/segment/column_meta_accessor.h"
 #include "storage/segment/column_reader.h"
 #include "storage/segment/column_reader_cache.h"
+#include "storage/segment/common.h" // k_segment_magic
 #include "storage/segment/empty_segment_iterator.h"
 #include "storage/segment/page_io.h"
 #include "storage/segment/page_pointer.h"
 #include "storage/segment/segment_iterator.h"
-#include "storage/segment/segment_writer.h" // k_segment_magic_length
 #include "storage/segment/stream_reader.h"
 #include "storage/segment/variant/variant_column_reader.h"
 #include "storage/tablet/tablet_schema.h"
diff --git a/be/src/storage/segment/segment_writer.cpp b/be/src/storage/segment/segment_writer.cpp
index ed2b403c75fee1..2bdefe39ef775e 100644
--- a/be/src/storage/segment/segment_writer.cpp
+++ b/be/src/storage/segment/segment_writer.cpp
@@ -62,6 +62,7 @@
 #include "storage/rowset/rowset_writer_context.h" // RowsetWriterContext
 #include "storage/rowset/segment_creator.h"
 #include "storage/segment/column_writer.h" // ColumnWriter
+#include "storage/segment/common.h"        // k_segment_magic
 #include "storage/segment/encoding_info.h"
 #include "storage/segment/external_col_meta_util.h"
 #include "storage/segment/page_io.h"
@@ -80,9 +81,6 @@ namespace segment_v2 {
 
 using namespace ErrorCode;
 
-const char* k_segment_magic = "D0R1";
-const uint32_t k_segment_magic_length = 4;
-
 inline std::string segment_mem_tracker_name(uint32_t segment_id) {
     return "SegmentWriter:Segment-" + std::to_string(segment_id);
 }
diff --git a/be/src/storage/segment/segment_writer.h b/be/src/storage/segment/segment_writer.h
index be22adcefaa9af..ba5d2f9af4462e 100644
--- a/be/src/storage/segment/segment_writer.h
+++ b/be/src/storage/segment/segment_writer.h
@@ -60,9 +60,6 @@ class FileWriter;
 
 namespace segment_v2 {
 
-extern const char* k_segment_magic;
-extern const uint32_t k_segment_magic_length;
-
 class VariantStatsCaculator;
 
 struct SegmentWriterOptions {
diff --git a/be/src/storage/segment/variant/v2/variant_path_builder.cpp b/be/src/storage/segment/variant/v2/variant_path_builder.cpp
index c7542669fe40fc..d908963ef74bda 100644
--- a/be/src/storage/segment/variant/v2/variant_path_builder.cpp
+++ b/be/src/storage/segment/variant/v2/variant_path_builder.cpp
@@ -849,10 +849,6 @@ size_t dotted_path_depth(const PathInData& path) {
     return path.get_parts().size();
 }
 
-size_t path_allocated_bytes(const PathInData& path) {
-    return path.get_path().capacity() + path.get_parts().capacity() * sizeof(PathInData::Part);
-}
-
 size_t recursive_null_count(const IColumn& column) {
     if (const auto* nullable = check_and_get_column<ColumnNullable>(column)) {
         size_t count = 0;
diff --git a/be/src/storage/segment/variant/v2/variant_shredder.cpp b/be/src/storage/segment/variant/v2/variant_shredder.cpp
index d61b51aed9610e..99c312a83269d9 100644
--- a/be/src/storage/segment/variant/v2/variant_shredder.cpp
+++ b/be/src/storage/segment/variant/v2/variant_shredder.cpp
@@ -35,6 +35,7 @@
 #include "exec/common/variant_util.h"
 #include "exprs/function/parse/variant_jsonb_parse.h"
 #include "storage/tablet/tablet_schema.h"
+#include "util/json/path_in_data.h"
 #include "util/jsonb_writer.h"
 
 namespace doris::segment_v2 {
@@ -52,10 +53,6 @@ PathInData normalize_doc_publication_path(const PathInData& path) {
     return PathInData(path.get_path(), path.get_is_typed());
 }
 
-size_t path_allocated_bytes(const PathInData& path) {
-    return path.get_path().capacity() + path.get_parts().capacity() * sizeof(PathInData::Part);
-}
-
 } // namespace
 
 struct VariantShredder::Impl {
diff --git a/be/src/storage/segment/vertical_segment_writer.cpp b/be/src/storage/segment/vertical_segment_writer.cpp
index abbdb857bc4815..319597a04183bf 100644
--- a/be/src/storage/segment/vertical_segment_writer.cpp
+++ b/be/src/storage/segment/vertical_segment_writer.cpp
@@ -68,6 +68,7 @@
 #include "storage/rowset/rowset_writer_context.h" // RowsetWriterContext
 #include "storage/rowset/segment_creator.h"
 #include "storage/segment/column_writer.h" // ColumnWriter
+#include "storage/segment/common.h"        // k_segment_magic
 #include "storage/segment/encoding_info.h"
 #include "storage/segment/external_col_meta_util.h"
 #include "storage/segment/historical_row_retriever.h"
@@ -88,9 +89,6 @@ namespace doris::segment_v2 {
 
 using namespace ErrorCode;
 
-static constexpr const char* k_segment_magic = "D0R1";
-static constexpr uint32_t k_segment_magic_length = 4;
-
 inline std::string vertical_segment_writer_mem_tracker_name(uint32_t segment_id) {
     return "VerticalSegmentWriter:Segment-" + std::to_string(segment_id);
 }
diff --git a/be/src/udf/CMakeLists.txt b/be/src/udf/CMakeLists.txt
index 0d57e85f5b193d..45e28f93cb40d4 100755
--- a/be/src/udf/CMakeLists.txt
+++ b/be/src/udf/CMakeLists.txt
@@ -26,3 +26,8 @@ file(GLOB PYTHON_UDF_SOURCES "python/*.cpp")
 # can have dependencies on our other libs. The second version is shipped as part
 # of the UDF sdk, which can't use other libs.
 add_library(Udf STATIC ${PYTHON_UDF_SOURCES})
+
+# Unity build: eight homogeneous python-UDF glue TUs sharing the Python/arrow
+# header closure; batch 0 merges them into a single unity TU.
+set_target_properties(Udf PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                     UNITY_BUILD_BATCH_SIZE 0)
diff --git a/be/src/util/CMakeLists.txt b/be/src/util/CMakeLists.txt
index f23c3ad865c89b..8941a7ed57fe8e 100644
--- a/be/src/util/CMakeLists.txt
+++ b/be/src/util/CMakeLists.txt
@@ -41,3 +41,22 @@ add_library(Util STATIC
 )
 
 pch_reuse(Util)
+
+# Unity build for the small util TUs (parse-overhead dominated). Ten files
+# leak file-scope macros and stay individual; hash/city.cc keeps its generic
+# k0..k4 static hash constants away from sibling scopes. perf_counters.cpp
+# is a Linux-only member (macOS builds its _mac variant, which is clean).
+set(UTIL_UNITY_SKIP
+    ${CMAKE_CURRENT_SOURCE_DIR}/bfd_parser.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/hash/city.cc
+    ${CMAKE_CURRENT_SOURCE_DIR}/hash/murmur_hash3.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/network_util.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/os_info.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/os_util.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/perf_counters.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/simd/parquet_kernels.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/thrift_util.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/zlib.cpp)
+doris_skip_unity_inclusion(${UTIL_UNITY_SKIP})
+set_target_properties(Util PROPERTIES UNITY_BUILD ${DORIS_UNITY_BUILD}
+                                      UNITY_BUILD_BATCH_SIZE 24)
diff --git a/be/src/util/bvar_windowed_adder.h b/be/src/util/bvar_windowed_adder.h
index c4e9245b7e3246..f249dc3a198291 100644
--- a/be/src/util/bvar_windowed_adder.h
+++ b/be/src/util/bvar_windowed_adder.h
@@ -33,6 +33,12 @@
 
 namespace doris {
 
+// Shared window spans (seconds) for multi-window metrics.
+// bvar::Window enforces MAX_SECONDS_LIMIT = 3600, so the longest window is 1h.
+inline constexpr int WINDOW_5M = 300;
+inline constexpr int WINDOW_30M = 1800;
+inline constexpr int WINDOW_1H = 3600;
+
 /**
  * Multi-dimension windowed adder.
  *
diff --git a/be/src/util/json/path_in_data.h b/be/src/util/json/path_in_data.h
index 593904dd931462..2e549cb714a83c 100644
--- a/be/src/util/json/path_in_data.h
+++ b/be/src/util/json/path_in_data.h
@@ -135,6 +135,12 @@ class PathInData {
     bool is_typed = false;
 };
 
+// Heap bytes retained beyond sizeof(PathInData): the dotted-path string buffer
+// plus the parts vector.
+inline size_t path_allocated_bytes(const PathInData& path) {
+    return path.get_path().capacity() + path.get_parts().capacity() * sizeof(PathInData::Part);
+}
+
 class PathInDataBuilder {
 public:
     const PathInData::Parts& get_parts() const { return parts; }
diff --git a/be/src/util/sm3.cpp b/be/src/util/sm3.cpp
index 2d4349abc840ec..448f2365e5f12d 100644
--- a/be/src/util/sm3.cpp
+++ b/be/src/util/sm3.cpp
@@ -46,11 +46,11 @@ void SM3Digest::digest() {
 
     char hex_buf[2 * SM3_DIGEST_LENGTH];
 
-    static char dig_vec_lower[] = "0123456789abcdef";
+    static constexpr char sm3_dig_vec_lower[] = "0123456789abcdef";
     char* to = hex_buf;
     for (int i = 0; i < SM3_DIGEST_LENGTH; ++i) {
-        *to++ = dig_vec_lower[buf[i] >> 4];
-        *to++ = dig_vec_lower[buf[i] & 0x0F];
+        *to++ = sm3_dig_vec_lower[buf[i] >> 4];
+        *to++ = sm3_dig_vec_lower[buf[i] & 0x0F];
     }
     _hex.assign(hex_buf, 2 * SM3_DIGEST_LENGTH);
 }
diff --git a/be/src/util/timezone_utils.cpp b/be/src/util/timezone_utils.cpp
index b575ed7a29f8c3..358d8b962cc5fd 100644
--- a/be/src/util/timezone_utils.cpp
+++ b/be/src/util/timezone_utils.cpp
@@ -51,7 +51,7 @@ namespace doris {
 
 using ZoneList = std::unordered_map<std::string, cctz::time_zone>;
 
-RE2 time_zone_offset_format_reg(R"(^[+-]{1}\d{2}\:\d{2}$)"); // visiting is thread-safe
+static RE2 tz_offset_format_reg(R"(^[+-]{1}\d{2}\:\d{2}$)"); // visiting is thread-safe
 
 // for ut, make it never nullptr.
 std::unique_ptr<ZoneList> lower_zone_cache_ = std::make_unique<ZoneList>();
@@ -277,8 +277,8 @@ bool TimezoneUtils::parse_tz_offset_string(const std::string& timezone, cctz::ti
     }
 
     re2::StringPiece value;
-    if (time_zone_offset_format_reg.Match(normalized, 0, normalized.size(), RE2::UNANCHORED, &value,
-                                          1)) [[likely]] {
+    if (tz_offset_format_reg.Match(normalized, 0, normalized.size(), RE2::UNANCHORED, &value, 1))
+            [[likely]] {
         const bool positive = value[0] != '-';
         const int hour = std::stoi(value.substr(1, 2).as_string());
         const int minute = std::stoi(value.substr(4, 2).as_string());
diff --git a/be/test/io/cache/fs_file_cache_storage_leak_cleaner_test.cpp b/be/test/io/cache/fs_file_cache_storage_leak_cleaner_test.cpp
index a4b5035ad457ca..0ab9953cb54309 100644
--- a/be/test/io/cache/fs_file_cache_storage_leak_cleaner_test.cpp
+++ b/be/test/io/cache/fs_file_cache_storage_leak_cleaner_test.cpp
@@ -131,8 +131,10 @@ class FSFileCacheLeakCleanerTest : public BlockFileCacheTest {
             BlockMetaKey mkey(0, hash, offset);
             BlockMeta meta(FileCacheType::NORMAL, 16, 0);
             storage._meta_store->put(mkey, meta);
-            // Wait for async write to complete for test stability
-            for (int i = 0; i < 100 && storage._meta_store->get_write_queue_size() > 0; ++i) {
+            // The async worker dequeues an operation before it lands in rocksdb,
+            // so an empty write queue does not mean the entry is readable yet.
+            // Poll the store itself: leak-scan decisions count rocksdb contents.
+            for (int i = 0; i < 100 && !storage._meta_store->get(mkey).has_value(); ++i) {
                 std::this_thread::sleep_for(std::chrono::milliseconds(50));
             }
         }