@@ -226,10 +226,32 @@
          (fullpath (realpath this-script)))
     fullpath))
 (define *common:this-exe-fullpath* (common:get-this-exe-fullpath))
 (define *common:this-exe-dir*      (pathname-directory *common:this-exe-fullpath*))
 (define *common:this-exe-name*     (pathname-strip-directory *common:this-exe-fullpath*))
+
+;; when called from a wrapper I need sometimes to find the calling
+;; wrapper, this is for dashboard to find the correct megatest.
+;;
+(define (common:find-local-megatest #!optional (progname "megatest"))
+  (let ((res (filter file-exists?
+		     (map (lambda (updir)
+			    (let* ((lm  (car (argv)))
+				   (dir (pathname-directory lm))
+				   (exe (pathname-strip-directory lm)))
+			      (conc (if dir (conc dir "/") "")
+				    (case (string->symbol exe)
+				      ((dboard)    (conc updir progname))
+				      ((mtest)     (conc updir progname))
+				      ((dashboard) progname)
+				      (else exe)))))
+			  '("../../" "../")))))
+    (if (null? res)
+	(begin
+	  (debug:print 0 *current-log-port* "Failed to find this executable! Using what can be found on the path")
+	  progname)
+	(car res))))
 
 (define *common:logpro-exit-code->status-sym-alist*
   '( ( 0 . pass )
      ( 1 . fail )
      ( 2 . warn )
@@ -1758,17 +1780,25 @@
 	(handle-exceptions
 	 exn
 	 #f
 	 (with-output-to-file fullpath (lambda ()(pp dat)))))
       #f))
-  
+
+(define (common:raw-get-remote-host-load remote-host)
+  (handle-exceptions
+   exn
+   #f ;; more specific handling of errors needed
+   (with-input-from-pipe 
+    (conc "ssh " remote-host " cat /proc/loadavg")
+    (lambda ()(list (read)(read)(read))))))
+
 ;; get cpu load by reading from /proc/loadavg, return all three values
 ;;
 (define (common:get-cpu-load remote-host)
   (handle-exceptions
    exn
-   '(99 99 99)
+   '(-99 -99 -99)
    (let* ((actual-hostname (or remote-host (get-host-name) "localhost")))
      (or (common:get-cached-info actual-hostname "cpu-load")
 	 (let ((result (if remote-host
 			   (map (lambda (res)
 				  (if (eof-object? res) 9e99 res))
@@ -1775,12 +1805,21 @@
 			        (with-input-from-pipe 
 				 (conc "ssh " remote-host " cat /proc/loadavg")
 				 (lambda ()(list (read)(read)(read)))))
 			   (with-input-from-file "/proc/loadavg" 
 			     (lambda ()(list (read)(read)(read)))))))
-	   (common:write-cached-info actual-hostname "cpu-load" result)
-	                     result)))))
+	   (match
+	    result
+	    ((l1 l2 l3)
+	     (if (and (number? l1)
+		      (number? l2)
+		      (number? l3))
+		 (begin
+		   (common:write-cached-info actual-hostname "cpu-load" result)
+		   result)
+		 '(-1 -1 -1))) ;; -1 is bad result
+	    (else '(-2 -2 -2))))))))
 
 ;; get normalized cpu load by reading from /proc/loadavg and /proc/cpuinfo return all three values and the number of real cpus and the number of threads
 ;; returns alist '((adj-cpu-load . normalized-proc-load) ... etc.
 ;;  keys: adj-proc-load, adj-core-load, 1m-load, 5m-load, 15m-load
 ;;
@@ -2026,11 +2065,11 @@
 	      (common:wait-for-normalized-load maxload msg remote-host (- rem-tries 1))
 	      #f)))))
 
 ;; DO NOT CALL THIS DIRECTLY. It is called from common:wait-for-normalized-load
 ;;
-(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f))
+(define (common:wait-for-cpuload maxload-in numcpus-in waitdelay #!key (count 1000) (msg #f)(remote-host #f)(force-maxload #f)(num-tries 5))
   (let* ((loadavg (common:get-cpu-load remote-host))
 	 (numcpus (if (<= 1 numcpus-in) ;; not possible to have zero.  If we get 1, it's possible that we got the previous default, and we should check again
 		      (common:get-num-cpus remote-host)
 		      numcpus-in))
 	 (maxload (if force-maxload
@@ -2038,28 +2077,53 @@
 		      (if (number? maxload-in)
 			  (max maxload-in 0.5)
 			  0.5))) ;; so maxload must be greater than 0.5 for now BUG - FIXME?
 	 (first   (car loadavg))
 	 (next    (cadr loadavg))
-	 (adjload (* maxload (max 1 numcpus))) ;; possible bug where numcpus (or could be maxload) is zero, crude fallback is to at least use 1
+	 (adjmaxload (* maxload (max 1 numcpus))) ;; possible bug where
+					       ;; numcpus (or could be
+					       ;; maxload) is zero,
+					       ;; crude fallback is to
+					       ;; at least use 1
 	 (loadjmp (- first next))
-         (adjwait (min (+ 300 (random 10)) (abs (* (+ (random 10)(/ (- 1000 count) 10) waitdelay) (- first adjload) ))  )));; add some randomness to the time to break any alignment where netbatch dumps many jobs to machines simultaneously
-    ;; let's let the user know once in a long while that load checking is happening but not constantly report it
+	 ;; add some randomness to the time to break any alignment
+	 ;; where netbatch dumps many jobs to machines simultaneously
+         (adjwait (min (+ 300 (random 10)) (abs (* (+ (random 10)
+						      (/ (- 1000 count) 10)
+						      waitdelay)
+						   (- first adjmaxload) ))  )))
+    ;; let's let the user know once in a long while that load checking
+    ;; is happening but not constantly report it
     (if (> (random 100) 75) ;; about 25% of the time
 	(debug:print-info 1 *default-log-port* "Checking cpuload on " (or remote-host "localhost") ", maxload: " maxload
-			  ", load: " first ", adjload: " adjload ", loadjmp: " loadjmp))
+			  ", load: " first ", adjmaxload: " adjmaxload ", loadjmp: " loadjmp))
     (cond
-     ((and (> first adjload)
+     ((and (< first 0) ;; this indicates the loadavg data is bad - machine may not be reachable
+	   (> num-tries 0))
+      (debug:print 0 *default-log-port* "WARNING: received bad data from get-cpu-load " first ", we'll sleep 10s and try " num-tries " more times.")
+      (thread-sleep! 10)
+      (common:wait-for-cpuload maxload-in numcpus-in waitdelay
+			       count: count remote-host: remote-host force-maxload: force-maxload num-tries: (- num-tries 1)))
+     ((and (> first adjmaxload)
 	   (> count 0))
-      (debug:print-info 0 *default-log-port* "server start delayed " adjwait " seconds due to load " first " exceeding max of " adjload " on server " (or remote-host (get-host-name)) " (normalized load-limit: " maxload ") " (if msg msg ""))
+      (debug:print-info 0 *default-log-port*
+			"server start delayed " adjwait
+			" seconds due to load " first
+			" exceeding max of " adjmaxload
+			" on server " (or remote-host (get-host-name))
+			" (normalized load-limit: " maxload ") " (if msg msg ""))
       (thread-sleep! adjwait)
       (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
      ((and (> loadjmp numcpus)
 	   (> count 0))
       (debug:print-info 0 *default-log-port* "waiting " adjwait " seconds due to load jump " loadjmp " > numcpus " numcpus (if msg msg ""))
       (thread-sleep! adjwait)
-      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host)))))
+      (common:wait-for-cpuload maxload numcpus waitdelay count: (- count 1) msg: msg remote-host: remote-host))
+     (else
+      (if (> num-tries 0)
+	  (debug:print 0 *default-log-port* "Load on " (or remote-host "localhost") " is acceptable at " first " continuing.")
+	  (debug:print 0 *default-log-port* "Load on " (or remote-host "localhost") ", "first" could not be retrieved. Giving up and continuing."))))))
 
 (define (get-uname . params)
   (let* ((uname-res (process:cmd-run->list (conc "uname " (if (null? params) "-a" (car params)))))
 	 (uname #f))
     (if (null? (car uname-res))