ports/54835: inefficient timer in healthd

[Andriy Gapon]

>Number:         54835
>Category:       ports
>Synopsis:       inefficient timer in healthd
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-ports-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Thu Jul 24 22:20:08 PDT 2003
>Closed-Date:
>Last-Modified:
>Originator:     Andriy Gapon
>Release:        FreeBSD 4.8-RELEASE i386
>Organization:
>Environment:
System: FreeBSD edge.foundation.invalid 4.8-RELEASE FreeBSD 4.8-RELEASE #5: Wed Jun 25 19:32:36 EDT 2003
healthd-0.6.9

	
>Description:

healthd has a very inefficient way of timing interval between measurements

>How-To-Repeat:

run healthd in a daemon mode with 10sec interval between measurements,
observe how much CPU time it consumes,
make conclusions

>Fix:

unfortunately I have not received any response from the author, so I am
submitting a patch here.

--- 2-healthd.c.patch begins here ---
--- healthd.c.orig	Mon Jul  7 20:59:46 2003
+++ healthd.c	Mon Jul  7 21:46:38 2003
@@ -636,16 +636,15 @@
         FD_SET(sock, &ready);
       }
     }
-    to.tv_sec = 0;
-    if (sec <= 3) {
-      /* For quick reading we need quicker timeouts */
-      to.tv_usec = 100;
-    } else if (sec <= 10) {
-      /* Still want reasonable accuracy */
-      to.tv_usec = 500;
-    } else {
-      /* Let us reduce CPU ussage */
-      to.tv_usec = 1000;
+    gettimeofday(&to, 0);
+    to.tv_sec = now + sec - to.tv_sec;
+    if (to.tv_usec > 0) {/* most usual case */
+      to.tv_sec -= 1;
+      to.tv_usec = 1000000 - to.tv_usec;
+    }
+    if (to.tv_sec < 0) {/* somehow we are already late */
+      to.tv_sec = 0;
+      to.tv_usec = 0;
     }
     if (select(50, &ready, 0, 0, &to) < 0) {
       if (ExitProgram) {
--- 2-healthd.c.patch ends here ---


>Release-Note:
>Audit-Trail:
>Unformatted: