1 |
+ |
/* |
2 |
+ |
* i-scream central monitoring system |
3 |
+ |
* http://www.i-scream.org.uk |
4 |
+ |
* Copyright (C) 2000-2002 i-scream |
5 |
+ |
* |
6 |
+ |
* This program is free software; you can redistribute it and/or |
7 |
+ |
* modify it under the terms of the GNU General Public License |
8 |
+ |
* as published by the Free Software Foundation; either version 2 |
9 |
+ |
* of the License, or (at your option) any later version. |
10 |
+ |
* |
11 |
+ |
* This program is distributed in the hope that it will be useful, |
12 |
+ |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 |
+ |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
14 |
+ |
* GNU General Public License for more details. |
15 |
+ |
* |
16 |
+ |
* You should have received a copy of the GNU General Public License |
17 |
+ |
* along with this program; if not, write to the Free Software |
18 |
+ |
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
19 |
+ |
*/ |
20 |
+ |
|
21 |
|
//---PACKAGE DECLARATION--- |
22 |
< |
package uk.ac.ukc.iscream.client.monitors; |
22 |
> |
package uk.org.iscream.cms.server.client.monitors; |
23 |
|
|
24 |
|
//---IMPORTS--- |
25 |
|
import java.util.HashMap; |
26 |
|
import java.util.Iterator; |
27 |
< |
import uk.ac.ukc.iscream.client.*; |
28 |
< |
import uk.ac.ukc.iscream.core.*; |
29 |
< |
import uk.ac.ukc.iscream.util.*; |
30 |
< |
import uk.ac.ukc.iscream.componentmanager.*; |
27 |
> |
import java.util.StringTokenizer; |
28 |
> |
import uk.org.iscream.cms.server.client.*; |
29 |
> |
import uk.org.iscream.cms.server.core.*; |
30 |
> |
import uk.org.iscream.cms.server.util.*; |
31 |
> |
import uk.org.iscream.cms.server.componentmanager.*; |
32 |
|
|
33 |
|
/** |
34 |
< |
* This Monitor watches heartbeats |
34 |
> |
* This Monitor watches heartbeats. |
35 |
> |
* It generates an alert when a heartbeat that is expected |
36 |
> |
* does not arrive. Unlike all the other monitors, this one |
37 |
> |
* is driven by an event *not* occuring, rather than an |
38 |
> |
* event occuring. This means it must be actively checking |
39 |
> |
* for missing heartbeat's, and thus has an extra inner class |
40 |
> |
* thread. |
41 |
|
* |
42 |
|
* @author $Author$ |
43 |
|
* @version $Id$ |
44 |
|
*/ |
45 |
< |
public class Heartbeat__Monitor extends MonitorSkeleton implements Runnable { |
45 |
> |
public class Heartbeat__Monitor extends MonitorSkeleton { |
46 |
|
|
47 |
|
//---FINAL ATTRIBUTES--- |
48 |
|
|
51 |
|
*/ |
52 |
|
public final String REVISION = "$Revision$"; |
53 |
|
|
54 |
+ |
/** |
55 |
+ |
* A description of this monitor |
56 |
+ |
*/ |
57 |
|
public final String DESC = "Monitors Heartbeats."; |
58 |
|
|
59 |
+ |
/** |
60 |
+ |
* The default (used if not configured) period at |
61 |
+ |
* which to check for old heartbeats. (in seconds) |
62 |
+ |
*/ |
63 |
|
public final int DEFAULT_CHECK_PERIOD = 60; |
64 |
|
|
65 |
|
//---STATIC METHODS--- |
66 |
|
|
67 |
|
//---CONSTRUCTORS--- |
68 |
< |
|
68 |
> |
|
69 |
> |
/** |
70 |
> |
* Constructs a new Heartbeat monitor, and starts off |
71 |
> |
* the worker thread. |
72 |
> |
*/ |
73 |
|
public Heartbeat__Monitor() { |
74 |
< |
new Thread(this).start(); |
74 |
> |
super(); |
75 |
> |
createInitialHosts(); |
76 |
> |
new HeartbeatWorker().start(); |
77 |
|
} |
78 |
|
|
79 |
|
//---PUBLIC METHODS--- |
80 |
|
|
81 |
< |
public void run() { |
82 |
< |
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
83 |
< |
while(true) { |
84 |
< |
// this cycle period of this monitor's checks |
85 |
< |
int checkPeriod = 0; |
86 |
< |
try { |
87 |
< |
checkPeriod = Integer.parseInt(cp.getProperty(_name, "Monitor.Heartbeat.checkPeriod")); |
88 |
< |
} catch (PropertyNotFoundException e) { |
89 |
< |
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 |
< |
|
81 |
> |
/** |
82 |
> |
* Analyse a packet of data. In this case, this will just |
83 |
> |
* register the fact that a heartbeat has arrived. |
84 |
> |
* |
85 |
> |
* @param packet The packet of data to analyse |
86 |
> |
*/ |
87 |
> |
public void analysePacket(XMLPacket packet) { |
88 |
> |
String source = packet.getParam("packet.attributes.machine_name"); |
89 |
> |
if (!_hosts.containsKey(source)) { |
90 |
|
synchronized(this) { |
91 |
< |
// perform the checks (use HB hash, although they *should* be the same) |
58 |
< |
Iterator i = _hosts.keySet().iterator(); |
59 |
< |
while(i.hasNext()) { |
60 |
< |
// get host |
61 |
< |
String source = (String) i.next(); |
62 |
< |
// check it |
63 |
< |
boolean remove = analyseHB(source); |
64 |
< |
if(remove) { |
65 |
< |
i.remove(); |
66 |
< |
} |
67 |
< |
} |
91 |
> |
_hosts.put(source, new HeartbeatHolder(new Register(source, _name))); |
92 |
|
} |
69 |
– |
|
70 |
– |
// wait a while |
71 |
– |
try {Thread.sleep(checkPeriod * 1000);} catch (InterruptedException e) {} |
93 |
|
} |
94 |
+ |
HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hosts.get(source); |
95 |
+ |
lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000); |
96 |
|
} |
97 |
|
|
75 |
– |
public void analysePacket(XMLPacket packet) { |
76 |
– |
if (packet.getParam("packet.attributes.type").equals("heartbeat")) { |
77 |
– |
String source = packet.getParam("packet.attributes.machine_name"); |
78 |
– |
if (!_hosts.containsKey(source)) { |
79 |
– |
synchronized(this) { |
80 |
– |
HashMap registerHash = new HashMap(); |
81 |
– |
registerHash.put(source, new Register(source, _name, 1)); |
82 |
– |
_hosts.put(source, new HeartbeatHolder(registerHash)); |
83 |
– |
} |
84 |
– |
} |
85 |
– |
HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hosts.get(source); |
86 |
– |
lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000); |
87 |
– |
} |
88 |
– |
} |
89 |
– |
|
98 |
|
/** |
99 |
|
* Overrides the {@link java.lang.Object#toString() Object.toString()} |
100 |
|
* method to provide clean logging (every class should have this). |
101 |
|
* |
102 |
< |
* This uses the uk.ac.ukc.iscream.util.NameFormat class |
102 |
> |
* This uses the uk.org.iscream.cms.server.util.NameFormat class |
103 |
|
* to format the toString() |
104 |
|
* |
105 |
|
* @return the name of this class and its CVS revision |
119 |
|
} |
120 |
|
|
121 |
|
//---PRIVATE METHODS--- |
114 |
– |
|
115 |
– |
private boolean analyseHB(String source) { |
116 |
– |
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
117 |
– |
HeartbeatHolder hbHolder = (HeartbeatHolder) _hosts.get(source); |
118 |
– |
Register reg = (Register) ((HashMap) hbHolder.getRegisterHash()).get(source); |
119 |
– |
|
120 |
– |
// get host's HB interval (seconds) |
121 |
– |
// this should always exist, thus we set to 0 |
122 |
– |
int hostHBinterval = 0; |
123 |
– |
try { |
124 |
– |
hostHBinterval = Integer.parseInt(cp.getProperty("Host."+source, "Host.TCPUpdateTime")); |
125 |
– |
} catch (PropertyNotFoundException e) { |
126 |
– |
hostHBinterval = 0; |
127 |
– |
_logger.write(toString(), Logger.WARNING, "TCPUpdateTime value unavailable using default of " + hostHBinterval + " seconds"); |
128 |
– |
} catch (NumberFormatException e) { |
129 |
– |
hostHBinterval = 0; |
130 |
– |
_logger.write(toString(), Logger.WARNING, "Erronous TCPUpdateTime value in configuration using default of " + hostHBinterval + " seconds"); |
131 |
– |
} |
132 |
– |
|
133 |
– |
// get host's last HB time (seconds) |
134 |
– |
long lastHeartbeat = hbHolder.getLastHeartbeat(); |
135 |
– |
// time since last heartbeat (seconds) |
136 |
– |
long timeSinceLastHB = (System.currentTimeMillis()/1000) - lastHeartbeat; |
137 |
– |
// time since (or until if negative) the expected heartbeat |
138 |
– |
long timeSinceExpectedHB = timeSinceLastHB - (long) hostHBinterval; |
139 |
– |
|
140 |
– |
// best do a check in case the expected heartbeat is in the future |
141 |
– |
if(timeSinceExpectedHB < 0) { |
142 |
– |
timeSinceExpectedHB = 0; |
143 |
– |
} |
144 |
– |
|
145 |
– |
// find out the threshold level we're at |
146 |
– |
int newThreshold = checkAttributeThreshold(timeSinceExpectedHB, reg); |
147 |
– |
|
148 |
– |
// process the alert |
149 |
– |
processAlert(newThreshold, 0, "Heartbeat", reg, source, String.valueOf(timeSinceExpectedHB)); |
150 |
– |
|
151 |
– |
if(reg.getLastAlertLevel(0) == Alert.alertFINAL) { |
152 |
– |
return true; |
153 |
– |
} |
154 |
– |
return false; |
155 |
– |
} |
122 |
|
|
123 |
+ |
/** |
124 |
+ |
* Checks whether the time since the last heartbeat |
125 |
+ |
* is beyond the threshold(s). |
126 |
+ |
* |
127 |
+ |
* @param timeSinceLastHB a long time since the last heartbeat arrived |
128 |
+ |
* @param reg the Register for this host |
129 |
+ |
* @return the level which has been breached, if any |
130 |
+ |
*/ |
131 |
|
private int checkAttributeThreshold(long timeSinceLastHB, Register reg) { |
132 |
|
for(int thresholdLevel = Alert.thresholdLevels.length - 1; thresholdLevel >= 0; thresholdLevel--) { |
133 |
|
if (reg.getThreshold(thresholdLevel) != -1.0) { |
138 |
|
} |
139 |
|
return Alert.thresholdNORMAL; |
140 |
|
} |
141 |
+ |
|
142 |
+ |
/** |
143 |
+ |
* Gets an initial list of hosts from the config |
144 |
+ |
* and adds a fake set of heartbeats for them. |
145 |
+ |
* If the hosts don't respond within the timeout |
146 |
+ |
* period an alert will be raised. |
147 |
+ |
* |
148 |
+ |
* The effect of this is to allow us to know about |
149 |
+ |
* hosts which weren't on when we started up, and |
150 |
+ |
* will thus never have generated a heartbeat - yet |
151 |
+ |
* will still want to know they're not responding. |
152 |
+ |
*/ |
153 |
+ |
private void createInitialHosts() { |
154 |
+ |
// get the initial list of hosts from the config |
155 |
+ |
String initialHosts = ""; |
156 |
+ |
try { |
157 |
+ |
initialHosts = _cp.getProperty(_name, "Monitor.Heartbeat.initialHosts"); |
158 |
+ |
} catch (PropertyNotFoundException e) { |
159 |
+ |
// just leave initialHosts empty |
160 |
+ |
_logger.write(Heartbeat__Monitor.this.toString(), Logger.DEBUG, "No initial list of hosts set, defaulting to none."); |
161 |
+ |
} |
162 |
+ |
|
163 |
+ |
// parse through the initial hosts adding them |
164 |
+ |
StringTokenizer st = new StringTokenizer(initialHosts, ";"); |
165 |
+ |
while (st.hasMoreTokens()) { |
166 |
+ |
String source = st.nextToken(); |
167 |
+ |
// check if they already exist, don't want to add them twice |
168 |
+ |
if (!_hosts.containsKey(source)) { |
169 |
+ |
synchronized(this) { |
170 |
+ |
_hosts.put(source, new HeartbeatHolder(new Register(source, _name))); |
171 |
+ |
} |
172 |
+ |
} |
173 |
+ |
HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hosts.get(source); |
174 |
+ |
// set a "fake" heartbeat |
175 |
+ |
lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000); |
176 |
+ |
} |
177 |
+ |
} |
178 |
|
|
179 |
|
//---ACCESSOR/MUTATOR METHODS--- |
180 |
< |
|
180 |
> |
|
181 |
> |
/** |
182 |
> |
* Returns a reference to the Queue we're getting data |
183 |
> |
* from. This is specific to this monitor. |
184 |
> |
* |
185 |
> |
* @return a reference to a Queue to get data from |
186 |
> |
*/ |
187 |
> |
protected Queue getQueue() { |
188 |
> |
return MonitorManager.getInstance().getHeartbeatQueue(); |
189 |
> |
} |
190 |
> |
|
191 |
|
//---ATTRIBUTES--- |
192 |
|
|
193 |
|
/** |
205 |
|
* A reference to the configuration proxy in use |
206 |
|
*/ |
207 |
|
private ConfigurationProxy _cp = ConfigurationProxy.getInstance(); |
208 |
< |
|
208 |
> |
|
209 |
> |
/** |
210 |
> |
* A HashMap of hosts, with associated HeartbeatHolder's. |
211 |
> |
*/ |
212 |
|
private HashMap _hosts = new HashMap(); |
213 |
+ |
|
214 |
+ |
/** |
215 |
+ |
* A reference to the system logger. |
216 |
+ |
*/ |
217 |
+ |
private Logger _logger = ReferenceManager.getInstance().getLogger(); |
218 |
|
|
219 |
|
//---STATIC ATTRIBUTES--- |
220 |
|
|
221 |
|
//---INNER CLASSES--- |
222 |
< |
|
222 |
> |
|
223 |
> |
/** |
224 |
> |
* This inner class simply holding some information |
225 |
> |
* about a specific host. |
226 |
> |
*/ |
227 |
|
private class HeartbeatHolder { |
228 |
|
|
229 |
< |
public HeartbeatHolder(HashMap registerHash) { |
230 |
< |
_registerHash = registerHash; |
229 |
> |
/** |
230 |
> |
* Construct a new HeartbeatHolder. |
231 |
> |
*/ |
232 |
> |
public HeartbeatHolder(Register register) { |
233 |
> |
_register = register; |
234 |
|
} |
235 |
|
|
236 |
+ |
/** |
237 |
+ |
* Set the time of the last heartbeat |
238 |
+ |
*/ |
239 |
|
public void setLastHeartbeat(long lastHeartbeat) { |
240 |
|
_lastHeartbeat = lastHeartbeat; |
241 |
|
} |
242 |
|
|
243 |
+ |
/** |
244 |
+ |
* Get the time of the last heartbeat |
245 |
+ |
*/ |
246 |
|
public long getLastHeartbeat() { |
247 |
|
return _lastHeartbeat; |
248 |
|
} |
249 |
|
|
250 |
< |
public HashMap getRegisterHash() { |
251 |
< |
return _registerHash; |
250 |
> |
/** |
251 |
> |
* Get the Register |
252 |
> |
*/ |
253 |
> |
public Register getRegister() { |
254 |
> |
return _register; |
255 |
|
} |
256 |
|
|
257 |
+ |
/** |
258 |
+ |
* last heartbeat time |
259 |
+ |
*/ |
260 |
|
private long _lastHeartbeat; |
261 |
< |
private HashMap _registerHash; |
262 |
< |
} |
263 |
< |
|
261 |
> |
|
262 |
> |
/** |
263 |
> |
* register ref |
264 |
> |
*/ |
265 |
> |
private Register _register; |
266 |
> |
} |
267 |
> |
|
268 |
> |
/** |
269 |
> |
* This worker thread just checks all the hosts and then |
270 |
> |
* waits a period of time before doing it again. It sends |
271 |
> |
* Alerts as required. |
272 |
> |
*/ |
273 |
> |
private class HeartbeatWorker extends Thread { |
274 |
> |
|
275 |
> |
/** |
276 |
> |
* The main run method of this worker thread. It simply |
277 |
> |
* checks through all the hosts it has stored, running |
278 |
> |
* the analyseHB method on each. It then removes any |
279 |
> |
* that have passed a FINAL, and waits a (configured) |
280 |
> |
* length of time before doing it again. |
281 |
> |
*/ |
282 |
> |
public void run() { |
283 |
> |
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
284 |
> |
while(true) { |
285 |
> |
// this cycle period of this monitor's checks |
286 |
> |
int checkPeriod = 0; |
287 |
> |
try { |
288 |
> |
checkPeriod = Integer.parseInt(cp.getProperty(_name, "Monitor.Heartbeat.checkPeriod")); |
289 |
> |
} catch (PropertyNotFoundException e) { |
290 |
> |
checkPeriod = DEFAULT_CHECK_PERIOD; |
291 |
> |
_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Monitor.Heartbeat.checkPeriod value unavailable using default of " + checkPeriod + " seconds"); |
292 |
> |
} catch (NumberFormatException e) { |
293 |
> |
checkPeriod = DEFAULT_CHECK_PERIOD; |
294 |
> |
_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous Monitor.Heartbeat.checkPeriod value in configuration using default of " + checkPeriod + " seconds"); |
295 |
> |
} |
296 |
> |
|
297 |
> |
synchronized(Heartbeat__Monitor.this) { |
298 |
> |
// perform the checks (use HB hash, although they *should* be the same) |
299 |
> |
Iterator i = _hosts.keySet().iterator(); |
300 |
> |
while(i.hasNext()) { |
301 |
> |
// get host |
302 |
> |
String source = (String) i.next(); |
303 |
> |
// check it |
304 |
> |
boolean remove = analyseHB(source); |
305 |
> |
// remove it if it's passed a FINAL |
306 |
> |
if(remove) { |
307 |
> |
i.remove(); |
308 |
> |
} |
309 |
> |
} |
310 |
> |
} |
311 |
> |
|
312 |
> |
// wait a while |
313 |
> |
try {Thread.sleep(checkPeriod * 1000);} catch (InterruptedException e) {} |
314 |
> |
} |
315 |
> |
} |
316 |
> |
|
317 |
> |
/** |
318 |
> |
* Analyses a given host's state, and if need be generates |
319 |
> |
* a relevant Alert. Note that it also checks if the last |
320 |
> |
* alert sent is FINAL, in which case it returns true to |
321 |
> |
* indicate removal of this host. |
322 |
> |
* |
323 |
> |
* @param source the host to check |
324 |
> |
* @return whether this host can be deleted |
325 |
> |
*/ |
326 |
> |
private boolean analyseHB(String source) { |
327 |
> |
ConfigurationProxy cp = ConfigurationProxy.getInstance(); |
328 |
> |
HeartbeatHolder hbHolder = (HeartbeatHolder) _hosts.get(source); |
329 |
> |
Register reg = hbHolder.getRegister(); |
330 |
> |
|
331 |
> |
// get host's HB interval (seconds) |
332 |
> |
// this should always exist, thus we set to 0 |
333 |
> |
int hostHBinterval = 0; |
334 |
> |
try { |
335 |
> |
hostHBinterval = Integer.parseInt(cp.getProperty("Host."+source, "Host.TCPUpdateTime")); |
336 |
> |
} catch (PropertyNotFoundException e) { |
337 |
> |
hostHBinterval = 0; |
338 |
> |
_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "TCPUpdateTime value unavailable using default of " + hostHBinterval + " seconds"); |
339 |
> |
} catch (NumberFormatException e) { |
340 |
> |
hostHBinterval = 0; |
341 |
> |
_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous TCPUpdateTime value in configuration using default of " + hostHBinterval + " seconds"); |
342 |
> |
} |
343 |
> |
|
344 |
> |
// get host's last HB time (seconds) |
345 |
> |
long lastHeartbeat = hbHolder.getLastHeartbeat(); |
346 |
> |
// time since last heartbeat (seconds) |
347 |
> |
long timeSinceLastHB = (System.currentTimeMillis()/1000) - lastHeartbeat; |
348 |
> |
// time since (or until if negative) the expected heartbeat |
349 |
> |
long timeSinceExpectedHB = timeSinceLastHB - (long) hostHBinterval; |
350 |
> |
|
351 |
> |
// best do a check in case the expected heartbeat is in the future |
352 |
> |
if(timeSinceExpectedHB < 0) { |
353 |
> |
timeSinceExpectedHB = 0; |
354 |
> |
} |
355 |
> |
|
356 |
> |
// find out the threshold level we're at |
357 |
> |
int newThreshold = checkAttributeThreshold(timeSinceExpectedHB, reg); |
358 |
> |
|
359 |
> |
// process the alert |
360 |
> |
Heartbeat__Monitor.this.processAlert(newThreshold, "Heartbeat", reg, source, String.valueOf(timeSinceExpectedHB)); |
361 |
> |
|
362 |
> |
if(reg.getLastAlertLevel() == Alert.alertFINAL) { |
363 |
> |
return true; |
364 |
> |
} |
365 |
> |
return false; |
366 |
> |
} |
367 |
> |
} |
368 |
|
} |