Change io-watchdog --exact-timeout option to --method=TYPE, where

TYPE can be sloppy or exact. Similarly, change exact-timeout keyword
in config file to timeout-method = TYPE.

ChangeLog

@@ -3,6 +3,14 @@
 	* src/interposer/io-watchdog-interposer.c : 
 	  Ensure some stdio functions are not #defined, and undef them if so.
 
+	* src/watchdog/io-watchdog.c, src/watchdog/io-watchdog.1.in,
+	  src/conf/conf-parser.l, src/conf/conf-parser.y,
+	  src/conf/io-watchdog.conf.5.in :
+	  Change --exact-timeout option in io-watchdog to --method=exact,
+	  and exact-timeout keywork in io-watchdog.conf to timeout-method
+	  for possible expansion later.
+
+
 2007-07-05  Mark Grondona  <[email protected]>
 
 	* io-watchdog.spec : Add BuildRequires.

doc/example-config

@@ -9,8 +9,8 @@
 # target = glob           Define default target
 # actions = action,...    Define default list of actions
 #
-# exact-timeout           Enable more precise watchdog timeouts
-#                          (incurs some performance impact)
+# timeout-method = TYPE   Specify the watchdog timeout method
+#                          (sloppy or exact, default = sloppy)
 #
 # program pattern         Following settings only apply to programs
 #                          that match [pattern] (globbing pattern).

src/conf/conf-parser.l

@@ -34,7 +34,7 @@ actions         { return ACTIONS; }
 timeout         { return TIMEOUT; }
 rank            { return RANK; }
 target          { return TARGET; }
-exact-timeout   { return EXACT_TIMEOUT; }
+timeout-method  { return METHOD; }
 =               { return '='; }
 
 [^=;, \t\r\n]+  {

src/conf/conf-parser.y

@@ -22,7 +22,7 @@ int yydebug = 0;
 static int cf_action (char * name, char *val);
 static int cf_program (char *program);
 static int cf_timeout (char *timeout);
-static int cf_exact (char *exact);
+static int cf_method (char *method);
 static int cf_rank (char *rank);
 static int cf_target (char *target);
 static int cf_actions ();
@@ -41,7 +41,7 @@ static int stringlist_append (char *string);
 %token RANK
 %token TARGET
 %token SEARCH
-%token EXACT_TIMEOUT
+%token METHOD
 
 
 %%
@@ -62,8 +62,7 @@ stmt    : ACTION STRING '=' STRING   { if (cf_action ($2, $4) < 0)   YYABORT; }
         | TIMEOUT '=' STRING         { if (cf_timeout ($3) < 0)      YYABORT; }
         | RANK '=' STRING            { if (cf_rank ($3) < 0)         YYABORT; }
         | TARGET '=' STRING          { if (cf_target ($3) < 0)       YYABORT; }
-        | EXACT_TIMEOUT              { if (cf_exact ("1") < 0)       YYABORT; }
-        | EXACT_TIMEOUT '=' STRING   { if (cf_exact ($3) < 0)        YYABORT; }
+        | METHOD '=' STRING          { if (cf_method ($3) < 0)       YYABORT; }
         ;
 
 actions : ACTIONS '=' list           { if (cf_actions () < 0)        YYABORT; }
@@ -177,30 +176,23 @@ static int cf_timeout (char *timeout)
     return (rc);
 }
 
-static int cf_exact (char *exact)
+static int cf_method (char *method)
 {
     int val = -1;
 
-    if (exact == NULL)
-        val = 1;
-    else if (  strcmp (exact, "0") == 0
-            || strcasecmp (exact, "false") == 0
-            || strcasecmp (exact, "off") == 0
-            || strcasecmp (exact, "no") == 0)
+    if ( strcasecmp (method, "sloppy") == 0
+      || strcasecmp (method, "default") == 0)
         val = 0;
-    else if (  strcmp (exact, "1") == 0
-            || strcasecmp (exact, "true") == 0
-            || strcasecmp (exact, "yes") == 0
-            || strcasecmp (exact, "on") == 0)
+    else if (strcasecmp (method, "exact") == 0)
         val = 1;
     else 
-        log_err ("%s: %d: Invalid value for exact-timeout \"%s\"\n",
-                cf_file (), cf_line (), exact);
+        log_err ("%s: %d: Invalid value for timeout-method: \"%s\"\n",
+                cf_file (), cf_line (), method);
 
     if (val < 0)
         return (-1);
 
-    log_debug2 ("%s: %d: exact-timeout\n");
+    log_debug2 ("%s: %d: timeout-method = %s\n", method);
 
     io_watchdog_conf_set_exact_timeout (conf, val);
 

src/conf/io-watchdog.conf.5.in

@@ -44,10 +44,13 @@ it can be 's' for seconds (the default), 'm' for minutes, 'h' for
 hours, or 'd' for days. N may be an arbitrary floating point 
 number.
 .TP
-\fBexact-timeout\fR [ = \fI0|1\fR]
-Use a more precise method for the IO watchdog timeout. See the 
-documentation for the \fI\-\-exact-timeout\fR option in the 
-\fBio-watchdog\fR(1) man page for more information.
+\fBtimeout-method\fR = \fITYPE\fR
+Specify an alternate method for the IO watchdog timeout.
+Valid values for \fITYPE\fR are currently \fIsloppy\fR 
+and \fIexact\fR. The \fIexact\fR method uses a more precise
+algorithm for calculating the IO watchdog timeout, but may 
+cause some performance impact. The default value is 
+\fIsloppy\fR.
 .TP
 \fBactions\fR = \fRNAME\fR[,\fINAME\fR,...]
 Set the list of IO watchdog actions to \fINAME\fR. Multiple actions

src/watchdog/io-watchdog.1.in

@@ -68,18 +68,21 @@ as it avoids enabling the io watchdog on the \fBtime\fR process.
 Only target rank N (default = 0) of a SLURM job.
 
 .TP
-.BI "-e, --exact-timeout"
-Use a more precise method for the IO watchdog timeout. By default the
-IO watchdog uses a simple and very lightweight method for tracking IO
-in the interest of impacting the application as little as possible.
-The actual time at which the watchdog may timeout will be anywhere
-from N to 2*N, where N is the timeout period specified by the user.
-With the \fI\-\-exact-timeout\fR option, the IO watchdog will track
-the time of the last application write so that the timeout can be
-as exact as possible. However, this will affect IO performance, since
-a timestamp must be generated on every write call. The magnitude of
-the impact is dependent on the rate of calls to write, not necessarily
-the amount of data written.
+.BI "-m, --method=" TYPE
+Specify an alternate method for the IO watchdog timeout.  Valid values for
+\fITYPE\fR are \fIsloppy\fR or \fIexact\fR.  By default the IO watchdog
+uses the \fIsloppy\fR method, which is a simple and very lightweight
+method for tracking IO, and thus has very little impact on application
+performance. With the \fIsloppy\fR method, however, the actual time at
+which the watchdog will time out may be anywhere from N to 2*N, where N is
+the timeout period selected by the user.  When the \fIexact\fR timeout
+method is selected, the IO watchdog timestamps application IO which
+allows more precise calculation of the next timeout interval. However,
+the \fIexact\fR method will have a greater impact on application write
+performace, because an extra call to \fIgettimeofday\fR(2) is generated
+for every write call.  The performance impact of the \fIexact\fR method
+is dependent on the rate of calls to write, not necessarily the amount
+of data written.
 
 .SH SEE ALSO
 \fBio-watchdog.conf\fR(5) 

src/watchdog/io-watchdog.c

@@ -37,11 +37,11 @@ struct option opt_table [] = {
     { "list-actions", 0, NULL, 'l' },
     { "server",       0, NULL, 'S' },
     { "shared-file",  1, NULL, 'F' },
-    { "exact-timeout",0, NULL, 'e' },
+    { "method",       1, NULL, 'm' },
     { NULL,           0, NULL, 0   }
 };
 
-const char * const opt_string = "hvlSea:t:T:r:F:f:";
+const char * const opt_string = "hvlSm:a:t:T:r:F:f:";
 
 #define USAGE "\
 Usage: %s [OPTIONS] [executable args...]\n\
@@ -56,8 +56,8 @@ Usage: %s [OPTIONS] [executable args...]\n\
                           time(1), as it avoids enabling the io-watchdog on \n\
                           the time process.\n\
   -r, --rank=N           Only target rank [N] (default = 0) of a SLURM job.\n\
-  -e, --exact-timeout    Use a more precise method for the watchdog timeout.\n\
-                          See the io-watchdog(1) man page for details.\n\
+  -m, --method=TYPE      Specify the method used for the watchdog timeout.\n\
+                          TYPE may be `sloppy' or `exact' (default = sloppy)\n\
   \n\
   -f, --config=file      Specify alternate config file [file]\n\
   \n\
@@ -81,6 +81,7 @@ struct io_watchdog_options {
     char *                           config_file;
 
     double                           timeout;
+    char *                           timeout_method;
     unsigned int                     exact_timeout;
     int                              verbose;
     int                              rank;
@@ -108,6 +109,7 @@ static void prog_ctx_init (struct prog_ctx *ctx, int ac, char *av []);
 static void prog_ctx_fini (struct prog_ctx *ctx);
 static void process_env (struct prog_ctx *ctx);
 static void parse_cmdline (struct prog_ctx *ctx, int ac, char *av []);
+static void set_timeout_method (struct prog_ctx *ctx, const char *method);
 static int  io_watchdog_server (struct prog_ctx *ctx);
 static int  check_user_actions (struct prog_ctx *ctx);
 static void list_all_actions (struct prog_ctx *ctx);
@@ -377,8 +379,10 @@ static void apply_config (struct prog_ctx *ctx)
     if (!ctx->opts.target)
         ctx->opts.target = xstrdup (io_watchdog_conf_target (ctx->conf));
 
-    if (!ctx->opts.exact_timeout && io_watchdog_conf_exact_timeout (ctx->conf))
+    if (!ctx->opts.timeout_method 
+       && io_watchdog_conf_exact_timeout (ctx->conf)) {
         ctx->opts.exact_timeout = 1;
+    }
 }
 
 static void parse_cmdline (struct prog_ctx *ctx, int ac, char *av[])
@@ -407,8 +411,8 @@ static void parse_cmdline (struct prog_ctx *ctx, int ac, char *av[])
                                       &ctx->opts.timeout_has_suffix) < 0)
                 log_fatal (1, "Invalid timeout string `%s'\n", optarg);
             break;
-        case 'e':
-            ctx->opts.exact_timeout = 1;
+        case 'm':
+            ctx->opts.timeout_method = strdup (optarg);
             break;
         case 'a':
             ctx->opts.actions = 
@@ -473,6 +477,8 @@ static void parse_cmdline (struct prog_ctx *ctx, int ac, char *av[])
         ctx->opts.actions = list_split_append (ctx->opts.actions, ",:", 
                                                ctx->opts.env_action_string);
 
+    set_timeout_method (ctx, ctx->opts.timeout_method);
+
     apply_config (ctx);
 
     if (ctx->opts.server_only && av [optind])
@@ -514,6 +520,22 @@ static int check_user_actions (struct prog_ctx *ctx)
     return (list_for_each (ctx->opts.actions, (ListForF) check_action, ctx));
 }
 
+static void set_timeout_method (struct prog_ctx *ctx, const char *method)
+{
+    if (!method)
+        return;
+
+    if (strcmp (method, "sloppy") == 0)
+        return;
+
+    if (strcmp (method, "exact") == 0) {
+        ctx->opts.exact_timeout = 1;
+        return;
+    }   
+
+    log_fatal (1, "Invalid timeout method \"%s\" specified.\n", method);
+}
+
 
 /******************************************************************************
  *  Forked io_watchdog server