WIP: the tip of my filterx performance efforts

cmake/syslog-ng-config.h.in

@@ -106,3 +106,4 @@
 #cmakedefine01 SYSLOG_NG_HAVE_SO_MEMINFO
 #cmakedefine01 SYSLOG_NG_ENABLE_AFSOCKET_MEMINFO_METRICS
 #cmakedefine01 SYSLOG_NG_HAVE_IV_WORK_POOL_SUBMIT_CONTINUATION
+#cmakedefine01 SYSLOG_NG_ENABLE_PERF

configure.ac

@@ -266,6 +266,10 @@ AC_ARG_ENABLE(ebpf,
               [  --enable-ebpf           Enable support for loading of eBPF programs (default: no)]
               ,,enable_ebpf="no")
 
+AC_ARG_ENABLE(perf,
+              [  --enable-perf           Enable support for the Linux perf tool (default: auto)]
+              ,,enable_perf="auto")
+
 AC_ARG_ENABLE(gcov,
               [  --enable-gcov           Enable coverage profiling (default: no)]
               ,,enable_gcov="no")
@@ -466,6 +470,7 @@ dnl Checks for programs.
 AC_PROG_CC
 AC_PROG_CC_C99
 AM_PROG_CC_C_O
+AM_PROG_AS
 if test "x$ac_cv_prog_cc_c99" = "xno"; then
 	AC_MSG_ERROR([C99 standard compliant C compiler required. Try GCC 3.x or later.])
 fi
@@ -2060,6 +2065,16 @@ if test "x$enable_ebpf" = "xyes"; then
         AC_SUBST(BPF_CC)
 fi
 
+if test "x$enable_perf" = "xauto"; then
+	uname_s=`uname -s`
+	uname_m=`uname -m`
+	if test "$uname_s" = "Linux" -a "$uname_m" = "x86_64"; then
+		enable_perf="yes"
+	else
+		enable_perf="no"
+	fi
+fi
+
 dnl ***************************************************************************
 dnl check if we have timezone variable in <time.h>
 dnl ***************************************************************************
@@ -2243,6 +2258,7 @@ AC_DEFINE_UNQUOTED(ENABLE_IPV6, `enable_value $enable_ipv6`, [Enable IPv6 suppor
 AC_DEFINE_UNQUOTED(ENABLE_TCP_WRAPPER, `enable_value $enable_tcp_wrapper`, [Enable TCP wrapper support])
 AC_DEFINE_UNQUOTED(ENABLE_LINUX_CAPS, `enable_value $enable_linux_caps`, [Enable Linux capability management support])
 AC_DEFINE_UNQUOTED(ENABLE_EBPF, `enable_value $enable_ebpf`, [Enable Linux eBPF support])
+AC_DEFINE_UNQUOTED(ENABLE_PERF, `enable_value $enable_perf`, [Enable Linux perf support])
 AC_DEFINE_UNQUOTED(ENABLE_ENV_WRAPPER, `enable_value $enable_env_wrapper`, [Enable environment wrapper support])
 AC_DEFINE_UNQUOTED(ENABLE_SYSTEMD, `enable_value $enable_systemd`, [Enable systemd support])
 AC_DEFINE_UNQUOTED(ENABLE_KAFKA, `enable_value $enable_kafka`, [Enable kafka support])
@@ -2259,6 +2275,7 @@ AM_CONDITIONAL(ENABLE_SUN_STREAMS, [test "$enable_sun_streams" = "yes"])
 AM_CONDITIONAL(ENABLE_DARWIN_OSL, [test "$enable_darwin_osl" = "yes"])
 AM_CONDITIONAL(ENABLE_OPENBSD_SYSTEM_SOURCE, [test "$enable_openbsd_system_source" = "yes"])
 AM_CONDITIONAL(ENABLE_EBPF, [test "$enable_ebpf" = "yes"])
+AM_CONDITIONAL(ENABLE_PERF, [test "$enable_perf" = "yes"])
 AM_CONDITIONAL(ENABLE_PACCT, [test "$enable_pacct" = "yes"])
 AM_CONDITIONAL(ENABLE_MONGODB, [test "$enable_mongodb" = "yes"])
 AM_CONDITIONAL(ENABLE_SMTP, [test "$enable_smtp" = "yes"])
@@ -2414,6 +2431,7 @@ echo "  Env wrapper support         : ${enable_env_wrapper:=no}"
 echo "  systemd support             : ${enable_systemd:=no} (unit dir: ${systemdsystemunitdir:=none})"
 echo "  systemd-journal support     : ${with_systemd_journal:=no}"
 echo "  JSON support                : $with_jsonc"
+echo "  perf support                : ${enable_perf:=no}"
 echo " Build options:"
 echo "  Generate manual pages       : ${enable_manpages:=no}"
 echo "  Install manual pages        : ${enable_manpages_install:=no}"

lib/Makefile.am

@@ -29,6 +29,7 @@ include lib/logthrsource/Makefile.am
 include lib/logthrdest/Makefile.am
 include lib/signal-slot-connector/Makefile.am
 include lib/multi-line/Makefile.am
+include lib/perf/Makefile.am
 
 LSNG_RELEASE		= $(shell echo @PACKAGE_VERSION@ | cut -d. -f1,2)
 
@@ -307,7 +308,8 @@ lib_libsyslog_ng_la_SOURCES		= \
 	$(multiline_sources)		\
 	$(logthrsource_sources)		\
 	$(logthrdest_sources) \
-	$(signal_slot_connector_sources)
+	$(signal_slot_connector_sources) \
+	$(perf_sources)
 
 lib_libsyslog_ng_la_CFLAGS		= \
 	$(AM_CFLAGS) \

lib/afinter.c

@@ -414,7 +414,7 @@ static gboolean
 afinter_sd_pre_config_init(LogPipe *s)
 {
   main_loop_worker_allocate_thread_space(1);
-  return TRUE;
+  return log_pipe_pre_config_init_method(s);
 }
 
 static gboolean
@@ -459,7 +459,10 @@ afinter_sd_post_config_init(LogPipe *s)
 {
   AFInterSourceDriver *self = (AFInterSourceDriver *) s;
 
-  return afinter_source_start_thread(self->source);
+  if (!afinter_source_start_thread(self->source))
+    return FALSE;
+
+  return log_pipe_post_config_init_method(s);
 }
 
 static gboolean

lib/cfg-grammar.y

@@ -567,14 +567,14 @@ expr_stmt
 source_stmt
         : KW_SOURCE string '{' source_content '}'
           {
-            $$ = log_expr_node_new_source($2, $4, &@1);
+            $$ = log_expr_node_new_source($2, $4, &@$);
             free($2);
           }
 	;
 dest_stmt
        : KW_DESTINATION string '{' dest_content '}'
           {
-            $$ = log_expr_node_new_destination($2, $4, &@1);
+            $$ = log_expr_node_new_destination($2, $4, &@$);
             free($2);
           }
 	;

lib/cfg-lexer.c

@@ -393,6 +393,7 @@ cfg_lexer_init_include_level_buffer(CfgLexer *self, CfgIncludeLevel *level,
   level->buffer.content = lexer_buffer;
   level->buffer.content_length = lexer_buffer_len;
   level->buffer.original_content = g_strdup(lexer_buffer);
+  level->buffer.original_lines = NULL;
 }
 
 gboolean
@@ -458,6 +459,7 @@ cfg_lexer_include_level_clear(CfgLexer *self, CfgIncludeLevel *level)
     {
       g_free(level->buffer.content);
       g_free(level->buffer.original_content);
+      g_strfreev(level->buffer.original_lines);
     }
   memset(level, 0, sizeof(*level));
 }

lib/cfg-lexer.h

@@ -160,6 +160,7 @@ struct _CfgIncludeLevel
     {
       /* the lexer mutates content, so save it for error reporting */
       gchar *original_content;
+      gchar **original_lines;
       /* buffer for the lexer */
       gchar *content;
       gsize content_length;

lib/cfg-source.c

@@ -173,7 +173,7 @@ _extract_source_from_file_location(GString *result, const gchar *filename, const
   FILE *f;
   gint lineno = 0;
   gint buflen = 65520;
-  gchar *line = g_malloc(buflen);
+  gboolean res = FALSE;
 
   if (yylloc->first_column < 1 || yylloc->last_column < 1 ||
       yylloc->first_column > buflen-1 || yylloc->last_column > buflen-1)
@@ -183,6 +183,7 @@ _extract_source_from_file_location(GString *result, const gchar *filename, const
   if (!f)
     return FALSE;
 
+  gchar *line = g_malloc(buflen);
   while (fgets(line, buflen, f))
     {
       lineno++;
@@ -216,19 +217,23 @@ _extract_source_from_file_location(GString *result, const gchar *filename, const
         }
     }
   fclose(f);
+  res = TRUE;
 
   /* NOTE: do we have the appropriate number of lines? */
   if (lineno <= yylloc->first_line)
-    return FALSE;
-
+    res = FALSE;
   g_free(line);
-  return TRUE;
+  return res;
 }
 
 static gboolean
-_extract_source_from_buffer_location(GString *result, const gchar *buffer_content, const CFG_LTYPE *yylloc)
+_extract_source_from_buffer_location(GString *result, CfgIncludeLevel *level, const CFG_LTYPE *yylloc)
 {
-  gchar **lines = g_strsplit(buffer_content, "\n", yylloc->last_line + 1);
+  const gchar *buffer_content = level->buffer.original_content;
+  gchar **lines = level->buffer.original_lines;
+
+  if (!lines)
+    lines = level->buffer.original_lines = g_strsplit(buffer_content, "\n", 0);
   gint num_lines = g_strv_length(lines);
 
   if (num_lines <= yylloc->first_line)
@@ -244,10 +249,18 @@ _extract_source_from_buffer_location(GString *result, const gchar *buffer_conten
 
       if (lineno == yylloc->first_line)
         {
+          gint token_start = MIN(linelen, yylloc->first_column - 1);
+
           if (yylloc->first_line == yylloc->last_line)
-            g_string_append_len(result, &line[MIN(linelen, yylloc->first_column-1)], yylloc->last_column - yylloc->first_column);
+            {
+              /* both last_column & first_column are 1 based, they cancel that out */
+              gint token_len = yylloc->last_column - yylloc->first_column;
+              if (token_start + token_len > linelen)
+                token_len = linelen - token_start;
+              g_string_append_len(result, &line[token_start], token_len);
+            }
           else
-            g_string_append(result, &line[MIN(linelen, yylloc->first_column-1)]);
+            g_string_append(result, &line[token_start]);
         }
       else if (lineno < yylloc->last_line)
         {
@@ -256,13 +269,16 @@ _extract_source_from_buffer_location(GString *result, const gchar *buffer_conten
         }
       else if (lineno == yylloc->last_line)
         {
+          /* last_column is 1 based */
+          gint token_len = yylloc->last_column - 1;
+          if (token_len > linelen)
+            token_len = linelen;
           g_string_append_c(result, ' ');
-          g_string_append_len(result, line, yylloc->last_column);
+          g_string_append_len(result, line, token_len);
         }
     }
 
 exit:
-  g_strfreev(lines);
   return TRUE;
 }
 
@@ -279,7 +295,7 @@ cfg_source_extract_source_text(CfgLexer *lexer, const CFG_LTYPE *yylloc, GString
       CFG_LTYPE buf_lloc = *yylloc;
       cfg_lexer_undo_set_file_location(lexer, &buf_lloc);
 
-      return _extract_source_from_buffer_location(result, level->buffer.original_content, &buf_lloc);
+      return _extract_source_from_buffer_location(result, level, &buf_lloc);
     }
   else
     g_assert_not_reached();

lib/cfg-tree.c

@@ -26,6 +26,7 @@
 #include "logmpx.h"
 #include "logpipe.h"
 #include "metrics-pipe.h"
+#include "cfg-source.h"
 
 #include <string.h>
 
@@ -282,6 +283,13 @@ log_expr_node_new(gint layout, gint content, const gchar *name, LogExprNode *chi
       self->line = yylloc->first_line;
       self->column = yylloc->first_column;
     }
+  if (yylloc && (debug_flag || 1))
+    {
+      CFG_LTYPE lloc = *yylloc;
+      GString *text = g_string_new("");
+      cfg_source_extract_source_text(configuration->lexer, &lloc, text);
+      self->expr_text = g_string_free(text, FALSE);
+    }
   return self;
 }
 
@@ -308,6 +316,7 @@ _log_expr_node_free(LogExprNode *self)
     self->aux_destroy(self->aux);
   g_free(self->name);
   g_free(self->filename);
+  g_free(self->expr_text);
   g_free(self);
 }
 
@@ -645,6 +654,7 @@ log_expr_node_lookup_flag(const gchar *flag)
   return 0;
 }
 
+/* stores the ref to @pipe */
 static LogPipe *
 cfg_tree_assoc_pipe(CfgTree *self, LogExprNode *node, LogPipe *pipe, const gchar *info)
 {
@@ -1610,6 +1620,40 @@ cfg_tree_stop(CfgTree *self)
   return success;
 }
 
+gboolean
+cfg_tree_optimize(CfgTree *self)
+{
+  gint i;
+
+  g_assert(self->compiled);
+
+  for (i = 0; i < self->initialized_pipes->len; i++)
+    {
+      LogPipe **ppipe = (LogPipe **) &g_ptr_array_index(self->initialized_pipes, i);
+      LogPipe *pipe = *ppipe;
+
+      log_pipe_optimize(&pipe);
+      if (pipe != *ppipe)
+        {
+//          log_pipe_unref(*ppipe);
+          *ppipe = log_pipe_ref(pipe);
+        }
+    }
+
+  for (i = 0; i < self->initialized_pipes->len;)
+    {
+      if (g_ptr_array_index(self->initialized_pipes, i) == NULL)
+        {
+          g_ptr_array_remove_index(self->initialized_pipes, i);
+        }
+      else
+        i++;
+    }
+
+
+  return TRUE;
+}
+
 gboolean
 cfg_tree_pre_config_init(CfgTree *self)
 {
@@ -1658,7 +1702,7 @@ void
 cfg_tree_init_instance(CfgTree *self, GlobalConfig *cfg)
 {
   memset(self, 0, sizeof(*self));
-  self->initialized_pipes = g_ptr_array_new();
+  self->initialized_pipes = g_ptr_array_new_with_free_func((GDestroyNotify) log_pipe_unref);
   self->objects = g_hash_table_new_full(cfg_tree_objects_hash, cfg_tree_objects_equal, NULL,
                                         (GDestroyNotify) log_expr_node_unref);
   self->templates = g_hash_table_new_full(g_str_hash, g_str_equal, NULL, (GDestroyNotify) log_template_unref);
@@ -1670,7 +1714,6 @@ cfg_tree_init_instance(CfgTree *self, GlobalConfig *cfg)
 void
 cfg_tree_free_instance(CfgTree *self)
 {
-  g_ptr_array_foreach(self->initialized_pipes, (GFunc) log_pipe_unref, NULL);
   g_ptr_array_free(self->initialized_pipes, TRUE);
 
   g_ptr_array_foreach(self->rules, (GFunc) log_expr_node_unref, NULL);

lib/cfg-tree.h

@@ -124,6 +124,7 @@ struct _LogExprNode
   gchar *filename;
   gint line, column;
   gint child_id;
+  gchar *expr_text;
 };
 
 gint log_expr_node_lookup_flag(const gchar *flag);
@@ -187,6 +188,7 @@ gchar *cfg_tree_get_rule_name(CfgTree *self, gint content, LogExprNode *node);
 gchar *cfg_tree_get_child_id(CfgTree *self, gint content, LogExprNode *node);
 
 gboolean cfg_tree_compile(CfgTree *self);
+gboolean cfg_tree_optimize(CfgTree *self);
 gboolean cfg_tree_start(CfgTree *self);
 gboolean cfg_tree_stop(CfgTree *self);
 gboolean cfg_tree_pre_config_init(CfgTree *self);

lib/cfg.c

@@ -301,6 +301,8 @@ cfg_init(GlobalConfig *cfg)
     return FALSE;
   if (!cfg_tree_compile(&cfg->tree))
     return FALSE;
+  if (!cfg_tree_optimize(&cfg->tree))
+    return FALSE;
   app_config_pre_pre_init();
   if (!cfg_tree_pre_config_init(&cfg->tree))
     return FALSE;

lib/filterx/Makefile.am

@@ -8,6 +8,7 @@ filterxinclude_HEADERS = 			\
 	lib/filterx/expr-condition.h \
 	lib/filterx/expr-done.h \
 	lib/filterx/expr-drop.h \
+	lib/filterx/expr-break.h \
 	lib/filterx/expr-function.h \
 	lib/filterx/expr-generator.h \
 	lib/filterx/expr-get-subscript.h \
@@ -24,6 +25,7 @@ filterxinclude_HEADERS = 			\
 	lib/filterx/expr-regexp.h \
 	lib/filterx/expr-set-subscript.h \
 	lib/filterx/expr-setattr.h \
+	lib/filterx/expr-switch.h \
 	lib/filterx/expr-template.h \
 	lib/filterx/expr-unset.h \
 	lib/filterx/expr-variable.h \
@@ -75,6 +77,7 @@ filterx_sources = 				\
 	lib/filterx/expr-condition.c \
 	lib/filterx/expr-done.c \
 	lib/filterx/expr-drop.c \
+	lib/filterx/expr-break.c \
 	lib/filterx/expr-function.c \
 	lib/filterx/expr-generator.c \
 	lib/filterx/expr-get-subscript.c \
@@ -91,6 +94,7 @@ filterx_sources = 				\
 	lib/filterx/expr-regexp.c \
 	lib/filterx/expr-set-subscript.c \
 	lib/filterx/expr-setattr.c \
+	lib/filterx/expr-switch.c \
 	lib/filterx/expr-template.c \
 	lib/filterx/expr-unset.c \
 	lib/filterx/expr-variable.c \