Implement `Process.warmup`

[Feature #18885]

For now, the optimizations performed are:

  - Run a major GC
  - Compact the heap
  - Promote all surviving objects to oldgen

Other optimizations may follow.

7 files changed, 94 insertions, 0 deletions

diff --git a/NEWS.md b/NEWS.md
index afcfb374fc..1b9f87c09b 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -35,6 +35,13 @@ Note: We're only listing outstanding class updates.
       The class use equality semantic to lookup keys like a regular hash,
       but it doesn't hold strong references on the keys. [[Feature #18498]]
 
+* Process.warnup
+
+    * Notify the Ruby virtual machine that the boot sequence is finished,
+      and that now is a good time to optimize the application. This is useful
+      for long running applications. The actual optimizations performed are entirely
+      implementation specific and may change in the future without notice. [[Feature #18885]
+
 ## Stdlib updates
 
 The following default gems are updated.
diff --git a/common.mk b/common.mk
index b1cf9b05a2..e186b06d73 100644
--- a/common.mk
+++ b/common.mk
@@ -11024,7 +11024,9 @@ process.$(OBJEXT): {$(VPATH)}thread_$(THREAD_MODEL).h
 process.$(OBJEXT): {$(VPATH)}thread_native.h
 process.$(OBJEXT): {$(VPATH)}util.h
 process.$(OBJEXT): {$(VPATH)}vm_core.h
+process.$(OBJEXT): {$(VPATH)}vm_debug.h
 process.$(OBJEXT): {$(VPATH)}vm_opts.h
+process.$(OBJEXT): {$(VPATH)}vm_sync.h
 ractor.$(OBJEXT): $(CCAN_DIR)/check_type/check_type.h
 ractor.$(OBJEXT): $(CCAN_DIR)/container_of/container_of.h
 ractor.$(OBJEXT): $(CCAN_DIR)/list/list.h
diff --git a/gc.c b/gc.c
index 3dc63433a7..e52e3fc35b 100644
--- a/gc.c
+++ b/gc.c
@@ -9833,6 +9833,26 @@ garbage_collect_with_gvl(rb_objspace_t *objspace, unsigned int reason)
     }
 }
 
+static int
+gc_promote_object_i(void *vstart, void *vend, size_t stride, void *data)
+{
+    rb_objspace_t *objspace = &rb_objspace;
+    VALUE v = (VALUE)vstart;
+    for (; v != (VALUE)vend; v += stride) {
+        switch (BUILTIN_TYPE(v)) {
+          case T_NONE:
+          case T_ZOMBIE:
+            break;
+          default:
+            if (!RVALUE_OLD_P(v) && !RVALUE_WB_UNPROTECTED(v)) {
+                RVALUE_AGE_SET_OLD(objspace, v);
+            }
+        }
+    }
+
+    return 0;
+}
+
 static VALUE
 gc_start_internal(rb_execution_context_t *ec, VALUE self, VALUE full_mark, VALUE immediate_mark, VALUE immediate_sweep, VALUE compact)
 {
@@ -9860,6 +9880,17 @@ gc_start_internal(rb_execution_context_t *ec, VALUE self, VALUE full_mark, VALUE
     return Qnil;
 }
 
+void
+rb_gc_prepare_heap(void)
+{
+    gc_start_internal(NULL, Qtrue, Qtrue, Qtrue, Qtrue, Qtrue);
+
+    /* The transient heap need to be evacuated before we promote objects */
+    rb_transient_heap_start_marking(true);
+    rb_transient_heap_evacuate();
+    rb_objspace_each_objects(gc_promote_object_i, NULL);
+}
+
 static int
 gc_is_moveable_obj(rb_objspace_t *objspace, VALUE obj)
 {
diff --git a/internal/gc.h b/internal/gc.h
index 2b67ca40dc..e0cc3bfa8a 100644
--- a/internal/gc.h
+++ b/internal/gc.h
@@ -214,6 +214,7 @@ extern VALUE *ruby_initial_gc_stress_ptr;
 extern int ruby_disable_gc;
 RUBY_ATTR_MALLOC void *ruby_mimmalloc(size_t size);
 void ruby_mimfree(void *ptr);
+void rb_gc_prepare_heap(void);
 void rb_objspace_set_event_hook(const rb_event_flag_t event);
 VALUE rb_objspace_gc_enable(struct rb_objspace *);
 VALUE rb_objspace_gc_disable(struct rb_objspace *);
diff --git a/process.c b/process.c
index daf3456bfd..9605036c64 100644
--- a/process.c
+++ b/process.c
@@ -115,6 +115,7 @@ int initgroups(const char *, rb_gid_t);
 #include "ruby/thread.h"
 #include "ruby/util.h"
 #include "vm_core.h"
+#include "vm_sync.h"
 #include "ruby/ractor.h"
 
 /* define system APIs */
@@ -8510,6 +8511,37 @@ static VALUE rb_mProcUID;
 static VALUE rb_mProcGID;
 static VALUE rb_mProcID_Syscall;
 
+/*
+ *  call-seq:
+ *     Process.warmup    -> true
+ *
+ *  Notify the Ruby virtual machine that the boot sequence is finished,
+ *  and that now is a good time to optimize the application. This is useful
+ *  for long running applications.
+ *
+ *  This method is expected to be called at the end of the application boot.
+ *  If the application is deployed using a pre-forking model, +Process.warmup+
+ *  should be called in the original process before the first fork.
+ *
+ *  The actual optimizations performed are entirely implementation specific
+ *  and may change in the future without notice.
+ *
+ *  On CRuby, +Process.warmup+:
+ *
+ *  * Perform a major GC.
+ *  * Compacts the heap.
+ *  * Promotes all surviving objects to the old generation.
+ */
+
+static VALUE
+proc_warmup(VALUE _)
+{
+    RB_VM_LOCK_ENTER();
+    rb_gc_prepare_heap();
+    RB_VM_LOCK_LEAVE();
+    return Qtrue;
+}
+
 
 /*
  * Document-module: Process
@@ -8627,6 +8659,8 @@ InitVM_process(void)
     rb_define_module_function(rb_mProcess, "getpriority", proc_getpriority, 2);
     rb_define_module_function(rb_mProcess, "setpriority", proc_setpriority, 3);
 
+    rb_define_module_function(rb_mProcess, "warmup", proc_warmup, 0);
+
 #ifdef HAVE_GETPRIORITY
     /* see Process.setpriority */
     rb_define_const(rb_mProcess, "PRIO_PROCESS", INT2FIX(PRIO_PROCESS));
diff --git a/spec/ruby/core/process/warmup_spec.rb b/spec/ruby/core/process/warmup_spec.rb
new file mode 100644
index 0000000000..fbdfd34848
--- /dev/null
+++ b/spec/ruby/core/process/warmup_spec.rb
@@ -0,0 +1,11 @@
+require_relative '../../spec_helper'
+
+describe "Process.warmup" do
+  ruby_version_is "3.3" do
+    # The behavior is entirely implementation dependant.
+    # Other implementations are free to just make it a noop
+    it "is implemented" do
+      Process.warmup.should == true
+    end
+  end
+end
diff --git a/test/ruby/test_process.rb b/test/ruby/test_process.rb
index 6e8050a54a..fa25d10353 100644
--- a/test/ruby/test_process.rb
+++ b/test/ruby/test_process.rb
@@ -4,6 +4,7 @@ require 'test/unit'
 require 'tempfile'
 require 'timeout'
 require 'rbconfig'
+require 'objspace'
 
 class TestProcess < Test::Unit::TestCase
   RUBY = EnvUtil.rubybin
@@ -2679,4 +2680,11 @@ EOS
       end
     end;
   end if Process.respond_to?(:_fork)
+
+  def test_warmup_promote_all_objects_to_oldgen
+    obj = Object.new
+    refute_includes(ObjectSpace.dump(obj), '"old":true')
+    Process.warmup
+    assert_includes(ObjectSpace.dump(obj), '"old":true')
+  end
 end