added host disk usage redbox and stats.

Hostdb.h

@@ -53,7 +53,7 @@ enum {
 #define PFLAG_RECOVERYMODE   0x80
 
 // added slow disk reads to it, 4 bytes (was 52)
-#define MAX_PING_SIZE (44+4)
+#define MAX_PING_SIZE (44+4+4)
 
 #define HT_GRUNT   0x01
 #define HT_SPARE   0x02
@@ -144,6 +144,8 @@ class Host {
 	// cpu usage
 	float          m_cpuUsage;
 
+	float          m_diskUsage;
+
 	long          m_slowDiskReads;
 
 	// doc count

PageHosts.cpp

@@ -23,6 +23,7 @@ static int dgramsFromSort ( const void *i1, const void *i2 );
 //static int loadAvgSort    ( const void *i1, const void *i2 );
 static int memUsedSort    ( const void *i1, const void *i2 );
 static int cpuUsageSort   ( const void *i1, const void *i2 );
+static int diskUsageSort  ( const void *i1, const void *i2 );
 
 long generatePingMsg( Host *h, long long nowms, char *buffer );
 
@@ -222,6 +223,9 @@ skipReplaceHost:
 		  "<td><a href=\"/master/hosts?c=%s&sort=10\">"
 		  "<b>cpu</a></td>"
 
+		  "<td><a href=\"/master/hosts?c=%s&sort=17\">"
+		  "<b>disk</a></td>"
+
 		  "<td><a href=\"/master/hosts?c=%s&sort=14\">"
 		  "<b>max ping1</a></td>"
 
@@ -257,6 +261,7 @@ skipReplaceHost:
 		  coll,
 		  coll,
 		  coll,
+		  coll,
 		  shotcol    );
 
 	// loop through each host we know and print it's stats
@@ -295,6 +300,7 @@ skipReplaceHost:
 	case 14:gbsort ( hostSort, nh, sizeof(long), pingMaxSort    ); break;
 	case 15:gbsort ( hostSort, nh, sizeof(long), slowDiskSort    ); break;
 	case 16:gbsort ( hostSort, nh, sizeof(long), defaultSort    ); break;
+	case 17:gbsort ( hostSort, nh, sizeof(long), diskUsageSort   ); break;
 	}
 
 	// we are the only one that uses these flags, so set them now
@@ -379,6 +385,12 @@ skipReplaceHost:
 		if ( cpu > 100.0 ) cpu = 100.0;
 		if ( cpu < 0.0   ) cpu = -1.0;
 
+		char diskUsageMsg[64];
+		sprintf(diskUsageMsg,"%.1f%%",h->m_diskUsage);
+		if ( h->m_diskUsage < 0.0 )
+			sprintf(diskUsageMsg,"???");
+
+
 		// split time, don't divide by zero!
 		long splitTime = 0;
 		if ( h->m_splitsDone ) 
@@ -494,6 +506,8 @@ skipReplaceHost:
 			  "<td>%s%.1f%%%s</td>"
 			  // cpu usage
 			  "<td>%.1f%%</td>"
+			  // disk usage
+			  "<td>%s</td>"
 
 			  // ping max
 			  "<td>%s</td>"
@@ -547,6 +561,7 @@ skipReplaceHost:
 			  h->m_percentMemUsed, // float
 			  fontTagBack,
 			  cpu, // float
+			  diskUsageMsg,
 
 			  // ping max
 			  pms,
@@ -1156,3 +1171,11 @@ int cpuUsageSort ( const void *i1, const void *i2 ) {
 	if ( h1->m_cpuUsage < h2->m_cpuUsage ) return  1;
 	return 0;
 }
+
+int diskUsageSort ( const void *i1, const void *i2 ) {
+	Host *h1 = g_hostdb.getHost ( *(long*)i1 );
+	Host *h2 = g_hostdb.getHost ( *(long*)i2 );
+	if ( h1->m_diskUsage > h2->m_diskUsage ) return -1;
+	if ( h1->m_diskUsage < h2->m_diskUsage ) return  1;
+	return 0;
+}

Pages.cpp

@@ -2586,7 +2586,7 @@ bool printRedBox ( SafeBuf *mb ) {
 
 
 	if ( g_conf.m_numConnectIps == 0 && g_conf.m_numMasterPwds == 0 ) {
-		if ( adds ) mb->safePrintf("<br>");
+		if ( adds ) mb->safePrintf("<br><br>");
 		adds++;
 		mb->safePrintf("URGENT. Please specify a password "
 			       "or IP address in the "
@@ -2595,9 +2595,26 @@ bool printRedBox ( SafeBuf *mb ) {
 			       "to access the Gigablast admin controls.");
 	}
 
+	// out of disk space?
+	long out = 0;
+	for ( long i = 0 ; i < g_hostdb.m_numHosts ; i++ ) {
+		Host *h = &g_hostdb.m_hosts[i];
+		if ( h->m_diskUsage < 98.0 ) continue;
+		out++;
+	}
+	if ( out > 0 ) {
+		if ( adds ) mb->safePrintf("<br><br>");
+		adds++;
+		char *s = "s are";
+		if ( out == 1 ) s = " is";
+		mb->safePrintf("%li host%s over 98%% disk usage. "
+			       "See the <a href=/admin/hosts>"
+			       "hosts</a> table.",out,s);
+	}
+
 
 	if ( g_pingServer.m_hostsConfInDisagreement ) {
-		if ( adds ) mb->safePrintf("<br>");
+		if ( adds ) mb->safePrintf("<br><br>");
 		adds++;
 		mb->safePrintf("The hosts.conf or localhosts.conf file "
 			      "is not the same over all hosts.");

PingServer.cpp

@@ -416,7 +416,7 @@ void PingServer::pingHost ( Host *h , uint32_t ip , uint16_t port ) {
 	//*(long*)p = (long)g_test.m_urlsIndexed;
 	//p += sizeof(long);
 	// our num recs, eventsIndexed
-	//*(long*)p = g_timedb.getNumTotalEvents();//g_coldb.m_numEventsAllColls;
+	//*(long*)p = g_timedb.getNumTotalEvents();//g_coldb.m_numEventsAllColl
 	//*(long *)p = 0;
 	//p += sizeof(long);
 	// slow disk reads
@@ -428,6 +428,8 @@ void PingServer::pingHost ( Host *h , uint32_t ip , uint16_t port ) {
 	// ensure crc is legit
 	if ( g_hostdb.getCRC() == 0 ) { char *xx=NULL;*xx=0; }
 
+	// disk usage (df -ka)
+	*(float *)p = g_process.m_diskUsage; p += 4;
 
 	// flags indicating our state
 	long flags = 0;
@@ -895,6 +897,10 @@ void handleRequest11 ( UdpSlot *slot , long niceness ) {
 		// sanity
 		if ( h->m_hostsConfCRC == 0 ) { char *xx=NULL;*xx=0; }
 
+		// disk usage
+		h->m_diskUsage = *(float *)p;
+		p += sizeof(float);
+
 		// put the state flags
 		h->m_flags = *(long *)(p);
 		p += sizeof(long);

Process.cpp

@@ -386,7 +386,7 @@ static void powerMonitorWrapper ( int fd , void *state ) ;
 static void fanSwitchCheckWrapper ( int fd , void *state ) ;
 static void gotPowerWrapper ( void *state , TcpSocket *s ) ;
 static void doneCmdWrapper        ( void *state ) ;
-//static void  hdtempWrapper        ( int fd , void *state ) ;
+static void  hdtempWrapper        ( int fd , void *state ) ;
 static void  hdtempDoneWrapper    ( void *state , ThreadEntry *t ) ;
 static void *hdtempStartWrapper_r ( void *state , ThreadEntry *t ) ;
 static void heartbeatWrapper    ( int fd , void *state ) ;
@@ -400,6 +400,8 @@ Process::Process ( ) {
 }
 
 bool Process::init ( ) {
+	// -1 means unknown
+	m_diskUsage = -1.0;
 	// we do not know if the fans are turned off or on
 	m_currentFanState = -1;
 	m_threadOut = false;
@@ -492,8 +494,9 @@ bool Process::init ( ) {
 
 	// . hard drive temperature
 	// . now that we use intel ssds that do not support smart, ignore this
-	//if ( ! g_loop.registerSleepCallback(10000,NULL,hdtempWrapper,0))
-	//	return false;
+	// . well use it for disk usage i guess
+	if ( ! g_loop.registerSleepCallback(10000,NULL,hdtempWrapper,0))
+		return false;
 
 	// power monitor, every 30 seconds
 	if ( ! g_loop.registerSleepCallback(30000,NULL,powerMonitorWrapper,0))
@@ -871,10 +874,65 @@ void hdtempDoneWrapper ( void *state , ThreadEntry *t ) {
 	s_lasttime = now;
 }
 
+
+// set Process::m_diskUsage
+float getDiskUsage ( ) {
+
+	// first get disk usage now
+	char cmd[10048];
+	char *out = "/tmp/diskusage";
+	snprintf(cmd,10000,"df -ka %s | tail -1 | awk '{print $5}' > %s",
+		 g_hostdb.m_dir,
+		 out);
+	int err = system ( cmd );
+	if ( err == 127 ) {
+		log("build: /bin/sh does not exist. can not get disk usage.");
+		return -1.0; // unknown
+	}
+	// this will happen if you don't upgrade glibc to 2.2.4-32 or above
+	if ( err != 0 ) {
+		log("build: Call to system(\"%s\") had error.",cmd);
+		return -1.0; // unknown
+	}
+
+	// read in temperatures from file
+	int fd = open ( "/tmp/diskusage" , O_RDONLY );
+	if ( fd < 0 ) {
+		//m_errno = errno;
+		log("build: Could not open %s for reading: %s.",
+		    out,mstrerror(errno));
+		return -1.0; // unknown
+	}
+	char buf[2000];
+	long r = read ( fd , buf , 2000 );
+	// did we get an error
+	if ( r <= 0 ) {
+		//m_errno = errno;
+		log("build: Error reading %s: %s.",out,mstrerror(errno));
+		close ( fd );
+		return -1.0; // unknown
+	}
+	// clean up shop
+	close ( fd );
+
+	float usage;
+	sscanf(buf,"%f",&usage);
+	return usage;
+}
+
+
 // . sets m_errno on error
 // . taken from Msg16.cpp
 void *hdtempStartWrapper_r ( void *state , ThreadEntry *t ) {
 
+	// run the df -ka cmd
+	g_process.m_diskUsage = getDiskUsage();
+
+
+	// ignore temps now. ssds don't have it
+	return NULL;
+	
+
 	static char *s_parm = "ata";
 	// make a system call to /usr/sbin/hddtemp /dev/sda,b,c,d
 	//char *cmd = 
@@ -884,9 +942,9 @@ void *hdtempStartWrapper_r ( void *state , ThreadEntry *t ) {
 	//	"/usr/sbin/hddtemp /dev/sdd >> /tmp/hdtemp  ";
  retry:
 	// linux 2.4 does not seem to like hddtemp
-	char cmd[10048];
 	char *path = g_hostdb.m_dir;
 	//char *path = "/usr/sbin/";
+	char cmd[10048];
 	sprintf ( cmd ,
 		  "%ssmartctl -Ad %s /dev/sda | grep Temp | awk '{print $10}' >  /tmp/hdtemp2;"
 		  "%ssmartctl -Ad %s /dev/sdb | grep Temp | awk '{print $10}' >> /tmp/hdtemp2;"
@@ -912,8 +970,8 @@ void *hdtempStartWrapper_r ( void *state , ThreadEntry *t ) {
 		//m_errno = EBADENGINEER;
 		log("build: Call to system(\"%s\") had error.",cmd);
 		//s_flag = 1;
-		// wait an hour
-		s_nextTime = getTime() + 3600;
+		// wait 5 minutes
+		s_nextTime = getTime() + 300; // 3600;
 		return NULL;
 	}
 	// read in temperatures from file

Process.h

@@ -89,6 +89,7 @@ class Process {
 	float m_roofTemp;
 	long  m_currentFanState;
 	long  m_desiredFanState;
+	float m_diskUsage;
 };
 
 extern Process g_process;