client/monitors/Heartbeat__Monitor.java

//---PACKAGE DECLARATION---
package uk.org.iscream.client.monitors;

//---IMPORTS---
import java.util.HashMap;
import java.util.Iterator;
import uk.org.iscream.client.*;
import uk.org.iscream.core.*;
import uk.org.iscream.util.*;
import uk.org.iscream.componentmanager.*;

/**
 * This Monitor watches heartbeats.
 * It generates an alert when a heartbeat that is expected
 * does not arrive. Unlike all the other monitors, this one
 * is driven by an event *not* occuring, rather than an
 * event occuring. This means it must be actively checking
 * for missing heartbeat's, and thus has an extra inner class
 * thread.
 *
 * @author  $Author: tdb1 $
 * @version $Id: Heartbeat__Monitor.java,v 1.18 2001/03/23 02:32:49 tdb1 Exp $
 */
public class Heartbeat__Monitor extends MonitorSkeleton {

//---FINAL ATTRIBUTES---

    /**
     * The current CVS revision of this class
     */
    public final String REVISION = "$Revision: 1.18 $";
    
    /**
     * A description of this monitor
     */
    public final String DESC = "Monitors Heartbeats.";
    
    /**
     * The default (used if not configured) period at
     * which to check for old heartbeats. (in seconds)
     */
    public final int DEFAULT_CHECK_PERIOD = 60;
    
//---STATIC METHODS---

//---CONSTRUCTORS---
    
    /**
     * Constructs a new Heartbeat monitor, and starts off
     * the worker thread.
     */
    public Heartbeat__Monitor() {
        super();
        new HeartbeatWorker().start();
    }

//---PUBLIC METHODS---
    
    /**
     * Analyse a packet of data. In this case, this will just
     * register the fact that a heartbeat has arrived.
     *
     * @param packet The packet of data to analyse
     */
    public void analysePacket(XMLPacket packet) {
        String source = packet.getParam("packet.attributes.machine_name");
        if (!_hosts.containsKey(source)) {
            synchronized(this) {
                _hosts.put(source, new HeartbeatHolder(new Register(source, _name)));
            }
        }
        HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hosts.get(source);
        lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000);
    }
    
    /**
     * Overrides the {@link java.lang.Object#toString() Object.toString()}
     * method to provide clean logging (every class should have this).
     *
     * This uses the uk.org.iscream.util.NameFormat class
     * to format the toString()
     *
     * @return the name of this class and its CVS revision
     */
    public String toString() {
        return FormatName.getName(
            _name,
            getClass().getName(),
            REVISION);
    }

    /**
     * return the String representation of what the monitor does
     */
    public String getDescription(){
        return DESC;
    }

//---PRIVATE METHODS---
    
    /**
     * Checks whether the time since the last heartbeat
     * is beyond the threshold(s).
     *
     * @param timeSinceLastHB a long time since the last heartbeat arrived
     * @param reg the Register for this host
     * @return the level which has been breached, if any
     */
    private int checkAttributeThreshold(long timeSinceLastHB, Register reg) {
        for(int thresholdLevel = Alert.thresholdLevels.length - 1; thresholdLevel >= 0; thresholdLevel--) {
            if (reg.getThreshold(thresholdLevel) != -1.0) {
                if (((long) reg.getThreshold(thresholdLevel)) < timeSinceLastHB) {
                    return thresholdLevel;
                }
            }
        }
        return Alert.thresholdNORMAL;
    }

//---ACCESSOR/MUTATOR METHODS---
    
    /**
     * Returns a reference to the Queue we're getting data
     * from. This is specific to this monitor.
     *
     * @return a reference to a Queue to get data from
     */
    protected Queue getQueue() {
        return MonitorManager.getInstance().getHeartbeatQueue();
    }
    
//---ATTRIBUTES---

    /**
     * This is the friendly identifier of the
     * component this class is running in.
     * eg, a Filter may be called "filter1",
     * If this class does not have an owning
     * component,  a name from the configuration
     * can be placed here.  This name could also
     * be changed to null for utility classes.
     */
    private String _name = "Heartbeat";
    
    /**
     * A reference to the configuration proxy in use
     */
    private ConfigurationProxy _cp = ConfigurationProxy.getInstance();
    
    /**
     * A HashMap of hosts, with associated HeartbeatHolder's.
     */
    private HashMap _hosts = new HashMap();
    
    /**
     * A reference to the system logger.
     */
    private Logger _logger = ReferenceManager.getInstance().getLogger();

//---STATIC ATTRIBUTES---

//---INNER CLASSES---
    
    /**
     * This inner class simply holding some information
     * about a specific host.
     */
    private class HeartbeatHolder {
        
        /**
         * Construct a new HeartbeatHolder.
         */
        public HeartbeatHolder(Register register) {
            _register = register;
        }
        
        /**
         * Set the time of the last heartbeat
         */
        public void setLastHeartbeat(long lastHeartbeat) {
            _lastHeartbeat = lastHeartbeat;
        }
        
        /**
         * Get the time of the last heartbeat
         */
        public long getLastHeartbeat() {
            return _lastHeartbeat;
        }
        
        /**
         * Get the Register
         */
        public Register getRegister() {
            return _register;
        }
        
        /**
         * last heartbeat time
         */
        private long _lastHeartbeat;
        
        /**
         * register ref
         */
        private Register _register;
    }
    
    /**
     * This worker thread just checks all the hosts and then
     * waits a period of time before doing it again. It sends
     * Alerts as required.
     */
    private class HeartbeatWorker extends Thread {
        
        /**
         * The main run method of this worker thread. It simply
         * checks through all the hosts it has stored, running
         * the analyseHB method on each. It then removes any
         * that have passed a FINAL, and waits a (configured)
         * length of time before doing it again.
         */
        public void run() {
            ConfigurationProxy cp = ConfigurationProxy.getInstance();
            while(true) {
                // this cycle period of this monitor's checks
                int checkPeriod = 0;
                try {
                    checkPeriod = Integer.parseInt(cp.getProperty(_name, "Monitor.Heartbeat.checkPeriod"));
                } catch (PropertyNotFoundException e) {
                    checkPeriod = DEFAULT_CHECK_PERIOD;
                    _logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Monitor.Heartbeat.checkPeriod value unavailable using default of " + checkPeriod + " seconds");
                } catch (NumberFormatException e) {
                    checkPeriod = DEFAULT_CHECK_PERIOD;
                    _logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous Monitor.Heartbeat.checkPeriod value in configuration using default of " + checkPeriod + " seconds");
                }
                
                synchronized(Heartbeat__Monitor.this) {
                    // perform the checks (use HB hash, although they *should* be the same)
                    Iterator i = _hosts.keySet().iterator();
                    while(i.hasNext()) {
                        // get host
                        String source = (String) i.next();
                        // check it
                        boolean remove = analyseHB(source);
                        // remove it if it's passed a FINAL
                        if(remove) {
                            i.remove();
                        }
                    }
                }
                
                // wait a while
                try {Thread.sleep(checkPeriod * 1000);} catch (InterruptedException e) {}
            }
        }
        
        /**
         * Analyses a given host's state, and if need be generates
         * a relevant Alert. Note that it also checks if the last
         * alert sent is FINAL, in which case it returns true to
         * indicate removal of this host.
         *
         * @param source the host to check
         * @return whether this host can be deleted
         */
        private boolean analyseHB(String source) {
            ConfigurationProxy cp = ConfigurationProxy.getInstance();
            HeartbeatHolder hbHolder = (HeartbeatHolder) _hosts.get(source);
            Register reg = hbHolder.getRegister();
            
            // get host's HB interval (seconds)
            // this should always exist, thus we set to 0
            int hostHBinterval = 0;
            try {
                hostHBinterval = Integer.parseInt(cp.getProperty("Host."+source, "Host.TCPUpdateTime"));
            } catch (PropertyNotFoundException e) {
                hostHBinterval = 0;
                _logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "TCPUpdateTime value unavailable using default of " + hostHBinterval + " seconds");
            } catch (NumberFormatException e) {
                hostHBinterval = 0;
                _logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous TCPUpdateTime value in configuration using default of " + hostHBinterval + " seconds");
            }
            
            // get host's last HB time (seconds)
            long lastHeartbeat = hbHolder.getLastHeartbeat();
            // time since last heartbeat (seconds)
            long timeSinceLastHB = (System.currentTimeMillis()/1000) - lastHeartbeat;
            // time since (or until if negative) the expected heartbeat
            long timeSinceExpectedHB = timeSinceLastHB - (long) hostHBinterval;
            
            // best do a check in case the expected heartbeat is in the future
            if(timeSinceExpectedHB < 0) {
                timeSinceExpectedHB = 0;
            }
            
            // find out the threshold level we're at
            int newThreshold = checkAttributeThreshold(timeSinceExpectedHB, reg);
            
            // process the alert
            Heartbeat__Monitor.this.processAlert(newThreshold, "Heartbeat", reg, source, String.valueOf(timeSinceExpectedHB));
            
            if(reg.getLastAlertLevel() == Alert.alertFINAL) {
                return true;
            }
            return false;
        }
    }
}
Revision:	1.19
Committed:	Mon Mar 26 18:07:11 2001 UTC (23 years, 2 months ago) by tdb
Branch:	MAIN
CVS Tags:	PROJECT_COMPLETION
Changes since 1.18:	+8 -8 lines
Log Message:	The "checking" part had moved from the main class to an inner class, thus the synchronize(this) blocks were pointing at different "this"'s. This how now been fixed, we hope.
#	Content
1	//---PACKAGE DECLARATION---
2	package uk.org.iscream.client.monitors;
3
4	//---IMPORTS---
5	import java.util.HashMap;
6	import java.util.Iterator;
7	import uk.org.iscream.client.*;
8	import uk.org.iscream.core.*;
9	import uk.org.iscream.util.*;
10	import uk.org.iscream.componentmanager.*;
11
12	/**
13	* This Monitor watches heartbeats.
14	* It generates an alert when a heartbeat that is expected
15	* does not arrive. Unlike all the other monitors, this one
16	* is driven by an event not occuring, rather than an
17	* event occuring. This means it must be actively checking
18	* for missing heartbeat's, and thus has an extra inner class
19	* thread.
20	*
21	* @author $Author: tdb1 $
22	* @version $Id: Heartbeat__Monitor.java,v 1.18 2001/03/23 02:32:49 tdb1 Exp $
23	*/
24	public class Heartbeat__Monitor extends MonitorSkeleton {
25
26	//---FINAL ATTRIBUTES---
27
28	/**
29	* The current CVS revision of this class
30	*/
31	public final String REVISION = "$Revision: 1.18 $";
32
33	/**
34	* A description of this monitor
35	*/
36	public final String DESC = "Monitors Heartbeats.";
37
38	/**
39	* The default (used if not configured) period at
40	* which to check for old heartbeats. (in seconds)
41	*/
42	public final int DEFAULT_CHECK_PERIOD = 60;
43
44	//---STATIC METHODS---
45
46	//---CONSTRUCTORS---
47
48	/**
49	* Constructs a new Heartbeat monitor, and starts off
50	* the worker thread.
51	*/
52	public Heartbeat__Monitor() {
53	super();
54	new HeartbeatWorker().start();
55	}
56
57	//---PUBLIC METHODS---
58
59	/**
60	* Analyse a packet of data. In this case, this will just
61	* register the fact that a heartbeat has arrived.
62	*
63	* @param packet The packet of data to analyse
64	*/
65	public void analysePacket(XMLPacket packet) {
66	String source = packet.getParam("packet.attributes.machine_name");
67	if (!_hosts.containsKey(source)) {
68	synchronized(this) {
69	_hosts.put(source, new HeartbeatHolder(new Register(source, _name)));
70	}
71	}
72	HeartbeatHolder lastHeartbeat = (HeartbeatHolder) _hosts.get(source);
73	lastHeartbeat.setLastHeartbeat(System.currentTimeMillis()/1000);
74	}
75
76	/**
77	* Overrides the {@link java.lang.Object#toString() Object.toString()}
78	* method to provide clean logging (every class should have this).
79	*
80	* This uses the uk.org.iscream.util.NameFormat class
81	* to format the toString()
82	*
83	* @return the name of this class and its CVS revision
84	*/
85	public String toString() {
86	return FormatName.getName(
87	_name,
88	getClass().getName(),
89	REVISION);
90	}
91
92	/**
93	* return the String representation of what the monitor does
94	*/
95	public String getDescription(){
96	return DESC;
97	}
98
99	//---PRIVATE METHODS---
100
101	/**
102	* Checks whether the time since the last heartbeat
103	* is beyond the threshold(s).
104	*
105	* @param timeSinceLastHB a long time since the last heartbeat arrived
106	* @param reg the Register for this host
107	* @return the level which has been breached, if any
108	*/
109	private int checkAttributeThreshold(long timeSinceLastHB, Register reg) {
110	for(int thresholdLevel = Alert.thresholdLevels.length - 1; thresholdLevel >= 0; thresholdLevel--) {
111	if (reg.getThreshold(thresholdLevel) != -1.0) {
112	if (((long) reg.getThreshold(thresholdLevel)) < timeSinceLastHB) {
113	return thresholdLevel;
114	}
115	}
116	}
117	return Alert.thresholdNORMAL;
118	}
119
120	//---ACCESSOR/MUTATOR METHODS---
121
122	/**
123	* Returns a reference to the Queue we're getting data
124	* from. This is specific to this monitor.
125	*
126	* @return a reference to a Queue to get data from
127	*/
128	protected Queue getQueue() {
129	return MonitorManager.getInstance().getHeartbeatQueue();
130	}
131
132	//---ATTRIBUTES---
133
134	/**
135	* This is the friendly identifier of the
136	* component this class is running in.
137	* eg, a Filter may be called "filter1",
138	* If this class does not have an owning
139	* component, a name from the configuration
140	* can be placed here. This name could also
141	* be changed to null for utility classes.
142	*/
143	private String _name = "Heartbeat";
144
145	/**
146	* A reference to the configuration proxy in use
147	*/
148	private ConfigurationProxy _cp = ConfigurationProxy.getInstance();
149
150	/**
151	* A HashMap of hosts, with associated HeartbeatHolder's.
152	*/
153	private HashMap _hosts = new HashMap();
154
155	/**
156	* A reference to the system logger.
157	*/
158	private Logger _logger = ReferenceManager.getInstance().getLogger();
159
160	//---STATIC ATTRIBUTES---
161
162	//---INNER CLASSES---
163
164	/**
165	* This inner class simply holding some information
166	* about a specific host.
167	*/
168	private class HeartbeatHolder {
169
170	/**
171	* Construct a new HeartbeatHolder.
172	*/
173	public HeartbeatHolder(Register register) {
174	_register = register;
175	}
176
177	/**
178	* Set the time of the last heartbeat
179	*/
180	public void setLastHeartbeat(long lastHeartbeat) {
181	_lastHeartbeat = lastHeartbeat;
182	}
183
184	/**
185	* Get the time of the last heartbeat
186	*/
187	public long getLastHeartbeat() {
188	return _lastHeartbeat;
189	}
190
191	/**
192	* Get the Register
193	*/
194	public Register getRegister() {
195	return _register;
196	}
197
198	/**
199	* last heartbeat time
200	*/
201	private long _lastHeartbeat;
202
203	/**
204	* register ref
205	*/
206	private Register _register;
207	}
208
209	/**
210	* This worker thread just checks all the hosts and then
211	* waits a period of time before doing it again. It sends
212	* Alerts as required.
213	*/
214	private class HeartbeatWorker extends Thread {
215
216	/**
217	* The main run method of this worker thread. It simply
218	* checks through all the hosts it has stored, running
219	* the analyseHB method on each. It then removes any
220	* that have passed a FINAL, and waits a (configured)
221	* length of time before doing it again.
222	*/
223	public void run() {
224	ConfigurationProxy cp = ConfigurationProxy.getInstance();
225	while(true) {
226	// this cycle period of this monitor's checks
227	int checkPeriod = 0;
228	try {
229	checkPeriod = Integer.parseInt(cp.getProperty(_name, "Monitor.Heartbeat.checkPeriod"));
230	} catch (PropertyNotFoundException e) {
231	checkPeriod = DEFAULT_CHECK_PERIOD;
232	_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Monitor.Heartbeat.checkPeriod value unavailable using default of " + checkPeriod + " seconds");
233	} catch (NumberFormatException e) {
234	checkPeriod = DEFAULT_CHECK_PERIOD;
235	_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous Monitor.Heartbeat.checkPeriod value in configuration using default of " + checkPeriod + " seconds");
236	}
237
238	synchronized(Heartbeat__Monitor.this) {
239	// perform the checks (use HB hash, although they should be the same)
240	Iterator i = _hosts.keySet().iterator();
241	while(i.hasNext()) {
242	// get host
243	String source = (String) i.next();
244	// check it
245	boolean remove = analyseHB(source);
246	// remove it if it's passed a FINAL
247	if(remove) {
248	i.remove();
249	}
250	}
251	}
252
253	// wait a while
254	try {Thread.sleep(checkPeriod * 1000);} catch (InterruptedException e) {}
255	}
256	}
257
258	/**
259	* Analyses a given host's state, and if need be generates
260	* a relevant Alert. Note that it also checks if the last
261	* alert sent is FINAL, in which case it returns true to
262	* indicate removal of this host.
263	*
264	* @param source the host to check
265	* @return whether this host can be deleted
266	*/
267	private boolean analyseHB(String source) {
268	ConfigurationProxy cp = ConfigurationProxy.getInstance();
269	HeartbeatHolder hbHolder = (HeartbeatHolder) _hosts.get(source);
270	Register reg = hbHolder.getRegister();
271
272	// get host's HB interval (seconds)
273	// this should always exist, thus we set to 0
274	int hostHBinterval = 0;
275	try {
276	hostHBinterval = Integer.parseInt(cp.getProperty("Host."+source, "Host.TCPUpdateTime"));
277	} catch (PropertyNotFoundException e) {
278	hostHBinterval = 0;
279	_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "TCPUpdateTime value unavailable using default of " + hostHBinterval + " seconds");
280	} catch (NumberFormatException e) {
281	hostHBinterval = 0;
282	_logger.write(Heartbeat__Monitor.this.toString(), Logger.WARNING, "Erronous TCPUpdateTime value in configuration using default of " + hostHBinterval + " seconds");
283	}
284
285	// get host's last HB time (seconds)
286	long lastHeartbeat = hbHolder.getLastHeartbeat();
287	// time since last heartbeat (seconds)
288	long timeSinceLastHB = (System.currentTimeMillis()/1000) - lastHeartbeat;
289	// time since (or until if negative) the expected heartbeat
290	long timeSinceExpectedHB = timeSinceLastHB - (long) hostHBinterval;
291
292	// best do a check in case the expected heartbeat is in the future
293	if(timeSinceExpectedHB < 0) {
294	timeSinceExpectedHB = 0;
295	}
296
297	// find out the threshold level we're at
298	int newThreshold = checkAttributeThreshold(timeSinceExpectedHB, reg);
299
300	// process the alert
301	Heartbeat__Monitor.this.processAlert(newThreshold, "Heartbeat", reg, source, String.valueOf(timeSinceExpectedHB));
302
303	if(reg.getLastAlertLevel() == Alert.alertFINAL) {
304	return true;
305	}
306	return false;
307	}
308	}
309	}