SCRIPT FLUSH run truly async, close lua interpreter in bio (#13087)

Even if we have SCRIPT FLUSH ASYNC now, when there are a lot of
lua scripts, SCRIPT FLUSH ASYNC will still block the main thread.
This is because lua_close is executed in the main thread, and lua
heap needs to release a lot of memory.

In this PR, we take the current lua instance on lctx.lua and call
lua_close on it in a background thread, to close it in async way.
This is MeirShpilraien's idea.

src/eval.c

@@ -264,15 +264,31 @@ void scriptingInit(int setup) {
     lctx.lua = lua;
 }
 
+/* Free lua_scripts dict and close lua interpreter. */
+void freeLuaScriptsSync(dict *lua_scripts, lua_State *lua) {
+    dictRelease(lua_scripts);
+    lua_close(lua);
+
+#if !defined(USE_LIBC)
+    /* The lua interpreter may hold a lot of memory internally, and lua is
+     * using libc. libc may take a bit longer to return the memory to the OS,
+     * so after lua_close, we call malloc_trim try to purge it earlier.
+     *
+     * We do that only when Redis itself does not use libc. When Lua and Redis
+     * use different allocators, one won't use the fragmentation holes of the
+     * other, and released memory can take a long time until it is returned to
+     * the OS. */
+    zlibc_trim();
+#endif
+}
+
 /* Release resources related to Lua scripting.
  * This function is used in order to reset the scripting environment. */
 void scriptingRelease(int async) {
     if (async)
-        freeLuaScriptsAsync(lctx.lua_scripts);
+        freeLuaScriptsAsync(lctx.lua_scripts, lctx.lua);
     else
-        dictRelease(lctx.lua_scripts);
-    lctx.lua_scripts_mem = 0;
-    lua_close(lctx.lua);
+        freeLuaScriptsSync(lctx.lua_scripts, lctx.lua);
 }
 
 void scriptingReset(int async) {

src/lazyfree.c

@@ -42,8 +42,9 @@ void lazyFreeTrackingTable(void *args[]) {
 /* Release the lua_scripts dict. */
 void lazyFreeLuaScripts(void *args[]) {
     dict *lua_scripts = args[0];
+    lua_State *lua = args[1];
     long long len = dictSize(lua_scripts);
-    dictRelease(lua_scripts);
+    freeLuaScriptsSync(lua_scripts, lua);
     atomicDecr(lazyfree_objects,len);
     atomicIncr(lazyfreed_objects,len);
 }
@@ -195,13 +196,14 @@ void freeTrackingRadixTreeAsync(rax *tracking) {
     }
 }
 
-/* Free lua_scripts dict, if the dict is huge enough, free it in async way. */
-void freeLuaScriptsAsync(dict *lua_scripts) {
+/* Free lua_scripts dict, if the dict is huge enough, free it in async way.
+ * Close lua interpreter, if there are a lot of lua scripts, close it in async way. */
+void freeLuaScriptsAsync(dict *lua_scripts, lua_State *lua) {
     if (dictSize(lua_scripts) > LAZYFREE_THRESHOLD) {
         atomicIncr(lazyfree_objects,dictSize(lua_scripts));
-        bioCreateLazyFreeJob(lazyFreeLuaScripts,1,lua_scripts);
+        bioCreateLazyFreeJob(lazyFreeLuaScripts,2,lua_scripts,lua);
     } else {
-        dictRelease(lua_scripts);
+        freeLuaScriptsSync(lua_scripts, lua);
     }
 }
 

src/server.h

@@ -3381,7 +3381,8 @@ void ldbKillForkedSessions(void);
 int ldbPendingChildren(void);
 sds luaCreateFunction(client *c, robj *body);
 void luaLdbLineHook(lua_State *lua, lua_Debug *ar);
-void freeLuaScriptsAsync(dict *lua_scripts);
+void freeLuaScriptsSync(dict *lua_scripts, lua_State *lua);
+void freeLuaScriptsAsync(dict *lua_scripts, lua_State *lua);
 void freeFunctionsAsync(functionsLibCtx *lib_ctx);
 int ldbIsEnabled(void);
 void ldbLog(sds entry);

src/zmalloc.c

@@ -50,7 +50,6 @@ void zlibc_free(void *ptr) {
 }
 
 #include <string.h>
-#include <pthread.h>
 #include "zmalloc.h"
 #include "atomicvar.h"
 
@@ -757,6 +756,15 @@ int jemalloc_purge(void) {
 
 #endif
 
+/* This function provides us access to the libc malloc_trim(). */
+void zlibc_trim(void) {
+#if defined(__GLIBC__) && !defined(USE_LIBC)
+    malloc_trim(0);
+#else
+    return;
+#endif
+}
+
 #if defined(__APPLE__)
 /* For proc_pidinfo() used later in zmalloc_get_smap_bytes_by_field().
  * Note that this file cannot be included in zmalloc.h because it includes

src/zmalloc.h

@@ -71,6 +71,7 @@
  */
 #ifndef ZMALLOC_LIB
 #define ZMALLOC_LIB "libc"
+#define USE_LIBC 1
 
 #if !defined(NO_MALLOC_USABLE_SIZE) && \
     (defined(__GLIBC__) || defined(__FreeBSD__) || \
@@ -93,6 +94,11 @@
 #endif
 #endif
 
+/* Includes for malloc_trim(), see zlibc_trim(). */
+#if defined(__GLIBC__) && !defined(USE_LIBC)
+#include <malloc.h>
+#endif
+
 /* We can enable the Redis defrag capabilities only if we are using Jemalloc
  * and the version used is our special version modified for Redis having
  * the ability to return per-allocation fragmentation hints. */
@@ -130,6 +136,7 @@ size_t zmalloc_get_private_dirty(long pid);
 size_t zmalloc_get_smap_bytes_by_field(char *field, long pid);
 size_t zmalloc_get_memory_size(void);
 void zlibc_free(void *ptr);
+void zlibc_trim(void);
 void zmadvise_dontneed(void *ptr);
 
 #ifdef HAVE_DEFRAG