@@ -395,24 +395,37 @@
                                    (delta (abs (- df disk-free))))
                               (if (and (> df 0)
                                        (> (/ delta df) 0.1)) ;; (> delta 200) ;; ignore changes under 200 Meg
                                   df
                                   #f)))
-             (do-sync       (or new-cpu-load new-disk-free over-time)))
+             (do-sync       (or new-cpu-load new-disk-free over-time))
+
+             (test-info   (rmt:get-test-info-by-id run-id test-id))
+             (state       (db:test-get-state test-info))
+             (status      (db:test-get-status test-info))
+             (kill-reason  "no kill reason specified")
+             (kill-job?    #f))
+
+        (cond
+         ((test-get-kill-request run-id test-id)
+          (set! kill-reason "KILLING TEST since received kill request (KILLREQ)")
+          (set! kill-job? #t))
+         ((and runtlim (> (- (current-seconds) start-seconds) runtlim))
+          (set! kill-reason (conc "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim))
+          (set! kill-job? #t))
+         ((equal? status "DEAD")
+          (set! kill-reason "KILLING TEST because it was marked as DEAD by launch:handle-zombie-tests")
+          (set! kill-job? #t)))
+
         (debug:print 4 *default-log-port* "cpu: " new-cpu-load " disk: " new-disk-free " last-sync: " last-sync " do-sync: " do-sync)
-	(set! kill-job? (or (test-get-kill-request run-id test-id) ;; run-id test-name itemdat))
-			    (and runtlim (let* ((run-seconds   (- (current-seconds) start-seconds))
-						(time-exceeded (> run-seconds runtlim)))
-					   (if time-exceeded
-					       (begin
-						 (debug:print-info 0 *default-log-port* "KILLING TEST DUE TO TIME LIMIT EXCEEDED! Runtime=" run-seconds " seconds, limit=" runtlim)
-						 #t)
-					       #f)))))
+        (launch:handle-zombie-tests run-id)
         (if do-sync
             (tests:update-central-meta-info run-id test-id new-cpu-load new-disk-free (calc-minutes) #f #f))
+        
 	(if kill-job? 
 	    (begin
+              (debug:print-info 0 *default-log-port* "proceeding to kill test: "kill-reason)
 	      (mutex-lock! m)
 	      ;; NOTE: The pid can change as different steps are run. Do we need handshaking between this
 	      ;;       section and the runit section? Or add a loop that tries three times with a 1/4 second
 	      ;;       between tries?
 	      (let* ((pid1 (launch:einf-pid exit-info)) ;; (vector-ref exit-info 0))
@@ -442,14 +455,14 @@
 				  #f
 				  (process-signal pid-num signal/kill)))
 			       (process:get-sub-pids pid))))
 		       ;;    (debug:print-info 0 *default-log-port* "not killing process " pid " as it is not alive"))))
 		       pids)
-		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (args:get-arg "-m") #f))
+		      (tests:test-set-status! run-id test-id "KILLED"  "KILLED" (conc (args:get-arg "-m")" "kill-reason) #f))
 		    (begin
 		      (debug:print-error 0 *default-log-port* "Nothing to kill, pid1=" pid1 ", pid2=" pid2)
-		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (args:get-arg "-m") #f)
+		      (tests:test-set-status! run-id test-id "KILLED"  "FAILED TO KILL" (conc (args:get-arg "-m")" "kill-reason) #f)
 		      )))
 	      (mutex-unlock! m)
 	      ;; no point in sticking around. Exit now.
 	      (exit)))
 	(if (hash-table-ref/default misc-flags 'keep-going #f)
@@ -620,11 +633,11 @@
 	      )
 	     (else ;; (member (db:test-get-state test-info) '("REMOVING" "REMOTEHOSTSTART" "RUNNING" "KILLREQ"))
 	      (debug:print-error 0 *default-log-port* "test state is " (db:test-get-state test-info) ", cannot proceed")
 	      (exit))))
 	  
-	  (debug:print 2 *default-log-port* "Exectuing " test-name " (id: " test-id ") on " (get-host-name))
+	  (debug:print 2 *default-log-port* "Executing " test-name " (id: " test-id ") on " (get-host-name))
 	  (set! keys       (rmt:get-keys))
 	  ;; (runs:set-megatest-env-vars run-id inkeys: keys inkeyvals: keyvals) ;; these may be needed by the launching process
 	  ;; one of these is defunct/redundant ...
 	  (if (not (launch:setup force-reread: #t))
 	      (begin
@@ -1332,10 +1345,32 @@
 	    (begin
 	      (debug:print-error 0 *default-log-port* "Failed to create work area at " test-path " with link at " lnktarget ", remaining attempts " remtries)
 	      ;; 
 	      (create-work-area run-id run-info keyvals test-id test-src-path disk-path testname itemdat remtries: (- remtries 1)))
 	    (list #f #f)))))
+
+
+(define (launch:handle-zombie-tests run-id)
+  (let* ((key (conc "zombiescan-runid-"run-id))
+         (now (current-seconds))
+         (threshold (- (current-seconds)  (* 2 (or (configf:lookup-number *configdat* "setup" "deadtime") 120))))
+         (val (rmt:get-var key))
+         (do-scan?
+          (cond
+           ((not val)
+            #t)
+           ((< val threshold)
+            #t)
+           (else #f))))
+    (when do-scan?
+      (debug:print 1 *default-log-port* "INFO: search and mark zombie tests")
+      (rmt:set-var key (current-seconds))
+      (rmt:find-and-mark-incomplete run-id #f))))
+
+
+
+
 
 ;; 1. look though disks list for disk with most space
 ;; 2. create run dir on disk, path name is meaningful
 ;; 3. create link from run dir to megatest runs area 
 ;; 4. remotely run the test on allocated host