1 |
tdb |
1.1 |
//---PACKAGE DECLARATION--- |
2 |
|
|
package uk.ac.ukc.iscream.client.monitors; |
3 |
|
|
|
4 |
|
|
//---IMPORTS--- |
5 |
|
|
import java.util.HashMap; |
6 |
|
|
import java.util.Iterator; |
7 |
|
|
import uk.ac.ukc.iscream.client.*; |
8 |
|
|
import uk.ac.ukc.iscream.core.*; |
9 |
|
|
import uk.ac.ukc.iscream.util.*; |
10 |
|
|
import uk.ac.ukc.iscream.componentmanager.*; |
11 |
|
|
|
12 |
|
|
/** |
13 |
|
|
* This Monitor watches heartbeats |
14 |
|
|
* |
15 |
tdb |
1.2 |
* @author $Author: tdb1 $ |
16 |
tdb |
1.4 |
* @version $Id: Heartbeat__Monitor.java,v 1.3 2001/03/05 23:14:30 tdb1 Exp $ |
17 |
tdb |
1.1 |
*/ |
18 |
tdb |
1.3 |
public class Heartbeat__Monitor implements PluginMonitor, Runnable { |
19 |
tdb |
1.1 |
|
20 |
|
|
//---FINAL ATTRIBUTES--- |
21 |
|
|
|
22 |
|
|
/** |
23 |
|
|
* The current CVS revision of this class |
24 |
|
|
*/ |
25 |
tdb |
1.4 |
public final String REVISION = "$Revision: 1.3 $"; |
26 |
tdb |
1.1 |
|
27 |
|
|
public final String DESC = "Monitors Heartbeats."; |
28 |
|
|
|
29 |
tdb |
1.3 |
public final int DEFAULT_CHECK_PERIOD = 60; |
30 |
|
|
|
31 |
tdb |
1.1 |
//---STATIC METHODS--- |
32 |
|
|
|
33 |
|
|
//---CONSTRUCTORS--- |
34 |
|
|
|
35 |
tdb |
1.2 |
public Heartbeat__Monitor() { |
36 |
tdb |
1.3 |
new Thread(this).start(); |
37 |
tdb |
1.2 |
} |
38 |
|
|
|
39 |
tdb |
1.1 |
//---PUBLIC METHODS--- |
40 |
|
|
|
41 |
|
|
public void run() { |
42 |
|
|
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
43 |
|
|
while(true) { |
44 |
tdb |
1.3 |
// this cycle period of this monitor's checks |
45 |
|
|
int checkPeriod = 0; |
46 |
|
|
try { |
47 |
|
|
checkPeriod = Integer.parseInt(cp.getProperty(_name, "Monitor.Heartbeat.checkPeriod")); |
48 |
|
|
} catch (PropertyNotFoundException e) { |
49 |
|
|
checkPeriod = DEFAULT_CHECK_PERIOD; |
50 |
|
|
_logger.write(toString(), Logger.WARNING, "Monitor.Heartbeat.checkPeriod value unavailable using default of " + checkPeriod + " seconds"); |
51 |
|
|
} catch (NumberFormatException e) { |
52 |
|
|
checkPeriod = DEFAULT_CHECK_PERIOD; |
53 |
|
|
_logger.write(toString(), Logger.WARNING, "Erronous Monitor.Heartbeat.checkPeriod value in configuration using default of " + checkPeriod + " seconds"); |
54 |
|
|
} |
55 |
tdb |
1.1 |
|
56 |
|
|
// perform the checks (use HB hash, although they *should* be the same) |
57 |
|
|
Iterator i = _hostsHB.keySet().iterator(); |
58 |
|
|
while(i.hasNext()) { |
59 |
|
|
// get host |
60 |
|
|
String source = (String) i.next(); |
61 |
|
|
// check it |
62 |
|
|
analyseHB(source); |
63 |
|
|
} |
64 |
|
|
|
65 |
|
|
// wait a while |
66 |
|
|
try {Thread.sleep(checkPeriod * 1000);} catch (InterruptedException e) {} |
67 |
|
|
} |
68 |
|
|
} |
69 |
|
|
|
70 |
|
|
// only use attribute num 0 :) |
71 |
|
|
public void analyseHB(String source) { |
72 |
|
|
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
73 |
|
|
Register reg = (Register) _hostsReg.get(source); |
74 |
|
|
|
75 |
|
|
// get host's HB interval (seconds) |
76 |
tdb |
1.3 |
// this should always exist, thus we set to 0 |
77 |
tdb |
1.2 |
int hostHBinterval = 0; |
78 |
tdb |
1.3 |
try { |
79 |
tdb |
1.4 |
hostHBinterval = Integer.parseInt(cp.getProperty("Host."+source, "Host.TCPUpdateTime")); |
80 |
tdb |
1.3 |
} catch (PropertyNotFoundException e) { |
81 |
|
|
hostHBinterval = 0; |
82 |
|
|
_logger.write(toString(), Logger.WARNING, "TCPUpdateTime value unavailable using default of " + hostHBinterval + " seconds"); |
83 |
|
|
} catch (NumberFormatException e) { |
84 |
|
|
hostHBinterval = 0; |
85 |
|
|
_logger.write(toString(), Logger.WARNING, "Erronous TCPUpdateTime value in configuration using default of " + hostHBinterval + " seconds"); |
86 |
tdb |
1.2 |
} |
87 |
|
|
|
88 |
tdb |
1.1 |
// get host's last HB time (seconds) |
89 |
tdb |
1.2 |
long lastHeartbeat = ((HeartbeatHolder) _hostsHB.get(source)).getLastHeartbeat(); |
90 |
tdb |
1.1 |
// time since last heartbeat (seconds) |
91 |
tdb |
1.2 |
long timeSinceLastHB = (System.currentTimeMillis()/1000) - lastHeartbeat; |
92 |
tdb |
1.1 |
|
93 |
|
|
// find out the threshold level we're at |
94 |
|
|
int result = checkAttributeThreshold(timeSinceLastHB, reg); |
95 |
|
|
|
96 |
|
|
// decide what threshold level we're on, if we've changed, record that |
97 |
|
|
if (result != reg.getLastThresholdLevel(0)) { |
98 |
|
|
reg.setLastThresholdLevel(0, result); |
99 |
|
|
} |
100 |
|
|
|
101 |
|
|
// as long as this isn't a normal level |
102 |
|
|
if(reg.getLastThresholdLevel(0) != Alert.thresholdNORMAL) { |
103 |
|
|
// if the time since the last alert is more than the time for |
104 |
|
|
// its timeout, fire an alert, escalate the alert |
105 |
|
|
long timeout = reg.getLastAlertTimeout(0); |
106 |
|
|
if ((timeout > 0) && (reg.getTimeLastSent(0) > 0)) { |
107 |
|
|
if((System.currentTimeMillis() - reg.getTimeLastSent(0)) > timeout) { |
108 |
|
|
int lastAlert = reg.getLastAlertLevel(0); |
109 |
|
|
reg.escalateAlert(0); |
110 |
|
|
reg.setTimeLastSent(0, System.currentTimeMillis()); |
111 |
|
|
reg.setLastAlertTimeout(0, reg.getAlertTimeout(reg.getLastAlertLevel(0), 0)); |
112 |
|
|
// -- SEND |
113 |
|
|
fireAlert(source, timeSinceLastHB, reg, lastAlert); |
114 |
|
|
} |
115 |
|
|
// if we don't have a timeout configured...we got STRAIGHT to the next level |
116 |
|
|
} else { |
117 |
|
|
int lastAlert = reg.getLastAlertLevel(0); |
118 |
|
|
reg.escalateAlert(0); |
119 |
|
|
reg.setTimeLastSent(0, System.currentTimeMillis()); |
120 |
|
|
reg.setLastAlertTimeout(0, reg.getAlertTimeout(reg.getLastAlertLevel(0), 0)); |
121 |
|
|
// -- SEND |
122 |
|
|
fireAlert(source, timeSinceLastHB, reg, lastAlert); |
123 |
|
|
} |
124 |
|
|
|
125 |
|
|
// we must be on ok, check the timeout value for this |
126 |
|
|
} else { |
127 |
|
|
// if we were on an OK alert before, then we don't do anything |
128 |
|
|
// but if we weren't we only set OK, once the timeout of the last |
129 |
|
|
// alert has occourd |
130 |
|
|
if (reg.getLastAlertLevel(0) != Alert.alertOK) { |
131 |
|
|
long timeout = reg.getLastAlertTimeout(0); |
132 |
|
|
if ((timeout > 0) && (reg.getTimeLastSent(0) > 0)) { |
133 |
|
|
if ((System.currentTimeMillis() - reg.getTimeLastSent(0)) > timeout) { |
134 |
|
|
int lastAlert = reg.getLastAlertLevel(0); |
135 |
|
|
reg.setLastAlertLevel(0, Alert.alertOK); |
136 |
|
|
reg.setTimeLastSent(0, System.currentTimeMillis()); |
137 |
|
|
reg.setLastAlertTimeout(0, timeout); |
138 |
|
|
// -- SEND |
139 |
|
|
fireAlert(source, timeSinceLastHB, reg, lastAlert); |
140 |
|
|
} |
141 |
|
|
} |
142 |
|
|
} |
143 |
|
|
} |
144 |
|
|
} |
145 |
|
|
|
146 |
|
|
public void analysePacket(XMLPacket packet) { |
147 |
|
|
if (packet.getParam("packet.attributes.type").equals("heartbeat")) { |
148 |
|
|
String source = packet.getParam("packet.attributes.machine_name"); |
149 |
|
|
if (!_hostsHB.containsKey(source)) { |
150 |
|
|
_hostsReg.put(source, new Register(source, _name, 1)); |
151 |
|
|
_hostsHB.put(source, new HeartbeatHolder()); |
152 |
|
|
} |
153 |
tdb |
1.2 |
HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hostsHB.get(source); |
154 |
|
|
lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000); |
155 |
tdb |
1.1 |
} |
156 |
|
|
} |
157 |
|
|
|
158 |
|
|
/** |
159 |
|
|
* Overrides the {@link java.lang.Object#toString() Object.toString()} |
160 |
|
|
* method to provide clean logging (every class should have this). |
161 |
|
|
* |
162 |
|
|
* This uses the uk.ac.ukc.iscream.util.NameFormat class |
163 |
|
|
* to format the toString() |
164 |
|
|
* |
165 |
|
|
* @return the name of this class and its CVS revision |
166 |
|
|
*/ |
167 |
|
|
public String toString() { |
168 |
|
|
return FormatName.getName( |
169 |
|
|
_name, |
170 |
|
|
getClass().getName(), |
171 |
|
|
REVISION); |
172 |
|
|
} |
173 |
|
|
|
174 |
|
|
/** |
175 |
|
|
* return the String representation of what the monitor does |
176 |
|
|
*/ |
177 |
|
|
public String getDescription(){ |
178 |
|
|
return DESC; |
179 |
|
|
} |
180 |
|
|
|
181 |
|
|
//---PRIVATE METHODS--- |
182 |
|
|
|
183 |
tdb |
1.2 |
private int checkAttributeThreshold(long timeSinceLastHB, Register reg) { |
184 |
tdb |
1.1 |
for(int thresholdLevel = Alert.thresholdLevels.length - 1; thresholdLevel >= 0; thresholdLevel--) { |
185 |
|
|
if (reg.getThreshold(thresholdLevel) != -1.0) { |
186 |
tdb |
1.2 |
if (((long) reg.getThreshold(thresholdLevel)) < timeSinceLastHB) { |
187 |
tdb |
1.1 |
return thresholdLevel; |
188 |
|
|
} |
189 |
|
|
} |
190 |
|
|
} |
191 |
|
|
return 0; |
192 |
|
|
} |
193 |
|
|
|
194 |
tdb |
1.2 |
private void fireAlert(String source, long timeSinceLastHB, Register reg, int lastAlert) { |
195 |
tdb |
1.1 |
int alertLevel = reg.getLastAlertLevel(0); |
196 |
|
|
int thresholdLevel = reg.getLastThresholdLevel(0); |
197 |
|
|
String currentValue = String.valueOf(timeSinceLastHB); |
198 |
|
|
String attributeName = "Heartbeat"; |
199 |
|
|
String thresholdValue = String.valueOf(reg.getThreshold(thresholdLevel)); |
200 |
|
|
String time = Long.toString(reg.getAlertTimeout(reg.getLastAlertLevel(0), 0) / 1000); |
201 |
|
|
if (thresholdLevel == Alert.thresholdNORMAL) { |
202 |
|
|
thresholdValue = "-"; |
203 |
|
|
} |
204 |
|
|
if (alertLevel == Alert.alertOK) { |
205 |
|
|
time = "0"; |
206 |
|
|
} |
207 |
|
|
Alert alert = new Alert(alertLevel, lastAlert, thresholdLevel, source, thresholdValue, currentValue, attributeName, time); |
208 |
|
|
_alerterQueue.add(alert); |
209 |
|
|
_logger.write(toString(), Logger.DEBUG, "Fired alert for source:" + source + " at alert level:" + Alert.alertLevels[alertLevel] + " on:" + attributeName + " for threshold level:" + Alert.thresholdLevels[thresholdLevel] + " at:" + currentValue + " exceeding threshold of:" +thresholdValue + " next alert sent in:" + time + "secs"); |
210 |
|
|
} |
211 |
|
|
|
212 |
|
|
//---ACCESSOR/MUTATOR METHODS--- |
213 |
|
|
|
214 |
|
|
//---ATTRIBUTES--- |
215 |
|
|
|
216 |
|
|
/** |
217 |
|
|
* This is the friendly identifier of the |
218 |
|
|
* component this class is running in. |
219 |
|
|
* eg, a Filter may be called "filter1", |
220 |
|
|
* If this class does not have an owning |
221 |
|
|
* component, a name from the configuration |
222 |
|
|
* can be placed here. This name could also |
223 |
|
|
* be changed to null for utility classes. |
224 |
|
|
*/ |
225 |
|
|
private String _name = "Heartbeat"; |
226 |
|
|
|
227 |
|
|
/** |
228 |
|
|
* This holds a reference to the |
229 |
|
|
* system logger that is being used. |
230 |
|
|
*/ |
231 |
|
|
private Logger _logger = ReferenceManager.getInstance().getLogger(); |
232 |
|
|
|
233 |
|
|
private Queue _alerterQueue = ClientMain._alerterQueue; |
234 |
|
|
|
235 |
|
|
/** |
236 |
|
|
* A reference to the configuration proxy in use |
237 |
|
|
*/ |
238 |
|
|
private ConfigurationProxy _cp = ConfigurationProxy.getInstance(); |
239 |
|
|
|
240 |
|
|
private HashMap _hostsHB = new HashMap(); |
241 |
|
|
private HashMap _hostsReg = new HashMap(); |
242 |
|
|
|
243 |
|
|
//---STATIC ATTRIBUTES--- |
244 |
|
|
|
245 |
|
|
//---INNER CLASSES--- |
246 |
|
|
|
247 |
|
|
private class HeartbeatHolder { |
248 |
|
|
|
249 |
tdb |
1.2 |
public void setLastHeartbeat(long lastHeartbeat) { |
250 |
tdb |
1.1 |
_lastHeartbeat = lastHeartbeat; |
251 |
|
|
} |
252 |
|
|
|
253 |
tdb |
1.2 |
public long getLastHeartbeat() { |
254 |
tdb |
1.1 |
return _lastHeartbeat; |
255 |
|
|
} |
256 |
|
|
|
257 |
tdb |
1.2 |
private long _lastHeartbeat; |
258 |
tdb |
1.1 |
} |
259 |
|
|
|
260 |
|
|
} |