made rsyslog tickless in the (usual and default) case that repeated

message reduction is turned off.

More info: http://blog.gerhards.net/2008/06/coding-to-save-environment.html

3 files changed, 22 insertions, 11 deletions

diff --git a/ChangeLog b/ChangeLog
index 2cc995bd..0304f3ef 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -6,6 +6,9 @@ Version 3.19.7 (rgerhards), 2008-06-??
   check before sending. Credits to Martin Schuette for providing the
   idea. Details are available at
   http://blog.gerhards.net/2008/06/reliable-plain-tcp-syslog-once-again.html
+- made rsyslog tickless in the (usual and default) case that repeated
+  message reduction is turned off. More info:
+  http://blog.gerhards.net/2008/06/coding-to-save-environment.html
 ---------------------------------------------------------------------------
 Version 3.19.6 (rgerhards), 2008-06-06
 - enhanced property replacer to support multiple regex matches
diff --git a/doc/rsyslog_ng_comparison.html b/doc/rsyslog_ng_comparison.html
index 72fee210..600875a8 100644
--- a/doc/rsyslog_ng_comparison.html
+++ b/doc/rsyslog_ng_comparison.html
@@ -120,7 +120,7 @@ based framing on syslog/tcp connections</td>
 </tr>
 <tr>
 <td valign="top">syslog over RELP<br>
-truly reliable message delivery (<a href="http://rgerhards.blogspot.com/2008/04/on-unreliability-of-plain-tcp-syslog.html">Why
+truly reliable message delivery (<a href="http://blog.gerhards.net/2008/05/why-you-cant-build-reliable-tcp.html">Why
 is plain tcp syslog not reliable?</a>)</td>
 <td valign="top">yes</td>
 <td valign="top">no</td>
@@ -583,4 +583,4 @@ the mean time, you may want to read it in parallel. It is available at
 <a href="http://www.balabit.com/network-security/syslog-ng/features/detailed/">Balabit's
 site</a>.</p>
 
-</body></html>
\ No newline at end of file
+</body></html>
diff --git a/tools/syslogd.c b/tools/syslogd.c
index 8b829771..7f415f3c 100644
--- a/tools/syslogd.c
+++ b/tools/syslogd.c
@@ -2491,15 +2491,22 @@ mainloop(void)
 	struct timeval tvSelectTimeout;
 
 	BEGINfunc
-	while(!bFinished){
-		/* first check if we have any internal messages queued and spit them out */
-		/* TODO: do we need this any longer? I doubt it, but let's care about it
- 		 * later -- rgerhards, 2007-12-21
- 		 */
-		processImInternal();
+	/* first check if we have any internal messages queued and spit them out. We used
+	 * to do that on any loop iteration, but that is no longer necessry. The reason
+	 * is that once we reach this point here, we always run on multiple threads and
+	 * thus the main queue is properly initialized. -- rgerhards, 2008-06-09
+	 */
+	processImInternal();
 
-		/* this is now just a wait */
-		tvSelectTimeout.tv_sec = TIMERINTVL;
+	while(!bFinished){
+		/* this is now just a wait - please note that we do use a near-"eternal"
+		 * timeout of 1 day if we do not have repeated message reduction turned on
+		 * (which it is not by default). This enables us to help safe the environment
+		 * by not unnecessarily awaking rsyslog on a regular tick (just think
+		 * powertop, for example). In that case, we primarily wait for a signal,
+		 * but a once-a-day wakeup should be quite acceptable. -- rgerhards, 2008-06-09
+		 */
+		tvSelectTimeout.tv_sec = (bReduceRepeatMsgs == 1) ? TIMERINTVL : 86400 /*1 day*/;
 		tvSelectTimeout.tv_usec = 0;
 		select(1, NULL, NULL, NULL, &tvSelectTimeout);
 		if(bFinished)
@@ -2526,7 +2533,8 @@ mainloop(void)
 		 * for the time being, I think the remaining risk can be accepted.
 		 * rgerhards, 2008-01-10
  		 */
-		doFlushRptdMsgs();
+		if(bReducedRepeatMsgs == 1)
+			doFlushRptdMsgs();
 
 		if(restart) {
 			dbgprintf("\nReceived SIGHUP, reloading rsyslogd.\n");