ViewVC Help
View File | Revision Log | Show Annotations | Revision Graph | Root Listing
root/i-scream/projects/cms/source/host/ihost-perl/ihost.pl
Revision: 1.27
Committed: Wed Mar 14 18:00:16 2001 UTC (23 years, 6 months ago) by pjm2
Content type: text/plain
Branch: MAIN
Changes since 1.26: +8 -1 lines
Log Message:
Added support for reconfiguring upon any type of heartbeat error.

File Contents

# User Rev Content
1 pjm2 1.6 #!/usr/bin/perl -w
2 pjm2 1.1
3     # -----------------------------------------------------------
4     # Perl i-scream Host.
5 pjm2 1.12 # http://www.i-scream.org.uk
6 pjm2 1.1 #
7     # An all-in-one script to act as an i-scream host on
8     # a typical Unix/Linux box. You may adapt the data-gathering
9     # methods as you see fit.
10     # - pjm2@ukc.ac.uk
11     #
12 tdb 1.26 # $Author: tdb1 $
13 pjm2 1.27 # $Id: ihost.pl,v 1.26 2001/03/12 23:51:15 tdb1 Exp $
14 pjm2 1.1 #------------------------------------------------------------
15    
16     $| = 1;
17    
18     use strict;
19     use IO::Socket;
20     use Sys::Hostname;
21    
22     use vars qw (
23     $filter_manager_addr
24     $filter_manager_port
25     $seq_no
26     $udp_update_time
27     $tcp_update_time
28     $last_udp_time
29     $last_tcp_time
30     $last_modified
31     $udp_port
32     $tcp_port
33     $filter_addr
34     $file_list
35 pjm2 1.17 $fqdn
36 tdb 1.25 $pidfile
37 pjm2 1.1 );
38    
39     if (@ARGV != 2) {
40     die "Usage: ihost.pl [i-scream filter manager] [TCP port]\n";
41     }
42    
43     $filter_manager_addr = $ARGV[0];
44     $filter_manager_port = $ARGV[1];
45    
46     $seq_no = 1;
47    
48 tdb 1.25 # write our PID to a file
49 tdb 1.26 $pidfile = "/var/tmp/ihost.pid";
50 tdb 1.25 &write_pid();
51    
52 pjm2 1.1 &tcp_configure();
53     &send_udp_packet();
54    
55     $last_udp_time = time;
56     $last_tcp_time = time;
57     while (1) {
58 pjm2 1.9 my($time) = time;
59     if ($time >= $last_udp_time + $udp_update_time) {
60 pjm2 1.1 &send_udp_packet();
61 pjm2 1.9 $last_udp_time = $time;
62 pjm2 1.1 }
63 pjm2 1.9 if ($time >= $last_tcp_time + $tcp_update_time) {
64 pjm2 1.1 &send_tcp_heartbeat();
65 pjm2 1.9 $last_tcp_time = $time;
66 pjm2 1.1 }
67 pjm2 1.9 my($next_udp) = $udp_update_time - $time + $last_udp_time;
68     my($next_tcp) = $tcp_update_time - $time + $last_tcp_time;
69     my($delay);
70     if ($next_udp < $next_tcp) {
71     $delay = $next_udp
72     }
73     else {
74     $delay = $next_tcp;
75     }
76     `sleep $delay`;
77 pjm2 1.1 }
78    
79 tdb 1.25 # we'll probably never get here...
80     `rm -f $pidfile`;
81 pjm2 1.1 exit(0);
82    
83 pjm2 1.12
84     #-----------------------------------------------------------------------
85     # tcp_configure
86     # Establishes a TCP connection to the specified i-scream filter manager.
87     # The host then requests details from the server, such as the intervals
88     # at which to send UDP packets.
89     #-----------------------------------------------------------------------
90 pjm2 1.1 sub tcp_configure() {
91    
92     my($sock) = new IO::Socket::INET(
93     PeerAddr => $filter_manager_addr,
94     PeerPort => $filter_manager_port,
95     Proto => 'tcp'
96 pjm2 1.21 ) or die "Cannot connect!";
97 pjm2 1.11 if (!defined $sock) {
98     print "IHOST ERROR: Could not connect to $filter_manager_addr:$filter_manager_port.\n";
99     print "Please check that there is an i-scream server at this address.\n";
100     print "Program ended.\n";
101     exit(1);
102     }
103 pjm2 1.1
104 pjm2 1.12 # Now run through the configuration process...
105 pjm2 1.1 my($response);
106    
107     print $sock "STARTCONFIG\n";
108     $response = <$sock>;
109     if (!chop $response eq "OK") {
110     print "The i-scream server rejected the STARTCONFIG command. Terminated.";
111     exit(1);
112     }
113    
114     print "Config started okay.\n";
115    
116     print $sock "LASTMODIFIED\n";
117     $response = <$sock>;
118     chop $response;
119     $last_modified = $response;
120    
121 pjm2 1.7 print "Config last modified: ". (scalar localtime $last_modified/1000) . "\n";
122 pjm2 1.1
123     print $sock "FILELIST\n";
124     $response = <$sock>;
125     chop $response;
126     $file_list = $response;
127    
128     print "File list obtained: $file_list\n";
129 pjm2 1.17
130     print $sock "FQDN\n";
131     $response = <$sock>;
132     chop $response;
133     $fqdn = $response;
134    
135     print "FQDN returned: $fqdn\n";
136 pjm2 1.1
137     print $sock "UDPUpdateTime\n";
138     $response = <$sock>;
139     chop $response;
140     $udp_update_time = $response;
141    
142     print $sock "TCPUpdateTime\n";
143     $response = <$sock>;
144     chop $response;
145     $tcp_update_time = $response;
146    
147 pjm2 1.7 print "UDP packet period: $udp_update_time seconds.\nTCP heartbeat period: $tcp_update_time seconds.\n";
148 pjm2 1.1
149     print $sock "ENDCONFIG\n";
150     $response = <$sock>;
151     chomp $response;
152     if (!$response eq "OK") {
153     print "ENDCONFIG command to server failed. Terminated.\n";
154     exit(1);
155     }
156    
157     print "Config ended.\n";
158    
159     print $sock "FILTER\n";
160     $response = <$sock>;
161 pjm2 1.24 unless (defined($response)) {
162     print "Failed: Could not get a filter address from the filter manager.\n";
163     exit(1);
164     }
165 pjm2 1.1 chop $response;
166     $response =~ /(.*);(.*);(.*)/;
167     ($filter_addr, $udp_port, $tcp_port) = ($1, $2, $3);
168 pjm2 1.24 unless (defined($filter_addr) && defined($udp_port) && defined($tcp_port)) {
169     print "Failed: Filter address response from server did not make sense: $response\n";
170     exit(1);
171     }
172 pjm2 1.1
173     print "Got filter data ($filter_addr, $udp_port, $tcp_port)\n";
174    
175     print $sock "END\n";
176     $response = <$sock>;
177     chop $response;
178     if ($response eq "OK") {
179     print "Host successfully configured via TCP.\n"
180     }
181     else {
182     print "The server failed the host configuration on the END command.";
183     exit(1);
184     }
185    
186     close($sock);
187    
188     print "Configuration finished sucessfully!\n";
189 pjm2 1.12
190     return;
191 pjm2 1.1 }
192    
193 pjm2 1.12
194    
195    
196     #-----------------------------------------------------------------------
197     # send_udp_packet
198     # Sends a UDP packet to an i-scream filter.
199     # The packet contains XML markup describing some of the machine's state.
200     # Receipt of UDP packets is not guaranteed.
201     #-----------------------------------------------------------------------
202 pjm2 1.1 sub send_udp_packet() {
203    
204     my(@statgrab) = `./statgrab.pl`;
205     my(%packet);
206 pjm2 1.13 for (my($i) = 0; $i <= $#statgrab; $i++) {
207 pjm2 1.1 $statgrab[$i] =~ /^([^\s]*) (.*)$/;
208     $packet{$1} = $2;
209     }
210    
211     my($date) = time;
212    
213     my($disk_info) = "<disk>";
214     my($i) = 0;
215     while (defined $packet{"packet.disk.p$i.attributes.mount"}) {
216 pjm2 1.15 $disk_info .= "<p$i";
217     $disk_info .= " name=\"" . $packet{"packet.disk.p$i.attributes.name"} . "\"";
218     $disk_info .= " kbytes=\"" . $packet{"packet.disk.p$i.attributes.kbytes"} . "\"";
219     $disk_info .= " used=\"" . $packet{"packet.disk.p$i.attributes.used"} . "\"";
220     $disk_info .= " avail=\"" . $packet{"packet.disk.p$i.attributes.avail"} . "\"";
221     $disk_info .= " mount=\"" . $packet{"packet.disk.p$i.attributes.mount"} . "\"";
222     $disk_info .= "></p$i>";
223 pjm2 1.1 ++$i;
224     }
225     $disk_info .= "</disk>";
226 pjm2 1.13
227 tdb 1.19 my($ip) = inet_ntoa(scalar(gethostbyname(hostname())) || 'localhost');
228 pjm2 1.1
229     # Build the XML packet this way, as we can clearly
230     # see the structure and contents... I like this ;-)
231 pjm2 1.12 # [Note that the server rejects UDP packets that are
232     # larger than 8196 bytes]
233 pjm2 1.1 my($xml) = <<EOF;
234    
235 tdb 1.19 <packet seq_no="$seq_no" machine_name="$fqdn" date="$date" type="data" ip="$ip">
236 pjm2 1.1 <load>
237     <load1>$packet{"packet.load.load1"}</load1>
238     <load5>$packet{"packet.load.load5"}</load5>
239     <load15>$packet{"packet.load.load15"}</load15>
240     </load>
241     <os>
242     <name>$packet{"packet.os.name"}</name>
243     <release>$packet{"packet.os.release"}</release>
244     <platform>$packet{"packet.os.platform"}</platform>
245     <sysname>$packet{"packet.os.sysname"}</sysname>
246     <version>$packet{"packet.os.version"}</version>
247 tdb 1.10 <uptime>$packet{"packet.os.uptime"}</uptime>
248 pjm2 1.1 </os>
249     <users>
250     <count>$packet{"packet.users.count"}</count>
251     <list>$packet{"packet.users.list"}</list>
252     </users>
253     <processes>
254     <total>$packet{"packet.processes.total"}</total>
255     <sleeping>$packet{"packet.processes.sleeping"}</sleeping>
256     <zombie>$packet{"packet.processes.zombie"}</zombie>
257     <stopped>$packet{"packet.processes.stopped"}</stopped>
258     <cpu>$packet{"packet.processes.cpu"}</cpu>
259     </processes>
260     <cpu>
261     <idle>$packet{"packet.cpu.idle"}</idle>
262     <user>$packet{"packet.cpu.user"}</user>
263     <kernel>$packet{"packet.cpu.kernel"}</kernel>
264     <iowait>$packet{"packet.cpu.iowait"}</iowait>
265     <swap>$packet{"packet.cpu.swap"}</swap>
266     </cpu>
267     <memory>
268 pjm2 1.13 <total>$packet{"packet.memory.total"}</total>
269 pjm2 1.1 <free>$packet{"packet.memory.free"}</free>
270     </memory>
271     <swap>
272 pjm2 1.13 <total>$packet{"packet.swap.total"}</total>
273     <free>$packet{"packet.swap.free"}</free>
274 pjm2 1.1 </swap>
275     $disk_info
276     </packet>
277    
278     EOF
279    
280 pjm2 1.12 # Make the packet smaller by stripping out newlines and leading spaces.
281 pjm2 1.1 $xml =~ s/\n\s*//g;
282 tdb 1.19
283 pjm2 1.1 my($sock) = new IO::Socket::INET (
284     PeerPort => $udp_port,
285     PeerAddr => $filter_addr,
286     Proto => 'udp'
287 pjm2 1.21 ) or die "Could not send UDP: $!\n";
288 pjm2 1.1
289     print $sock $xml or die "Could not send UDP packet: $!\n";
290     close($sock);
291     $seq_no++;
292     print "-";
293 pjm2 1.12
294     return;
295 pjm2 1.1 }
296    
297 pjm2 1.12
298    
299    
300     #-----------------------------------------------------------------------
301     # send_tcp_heartbeat
302     # Establishes a TCP connection to an i-scream filter.
303     # The heartbeat is used as a guaranteed "I'm alive" delivery mechanism.
304     #-----------------------------------------------------------------------
305 pjm2 1.1 sub send_tcp_heartbeat() {
306    
307     my($sock) = new IO::Socket::INET(
308     PeerAddr => $filter_addr,
309     PeerPort => $tcp_port,
310     Proto => 'tcp'
311 pjm2 1.21 ) or return;
312 pjm2 1.11 if (!defined $sock) {
313     print "IHOST WARNING: Failed to deliver a heartbeat to the i-scream filter.\n";
314 pjm2 1.27 &tcp_configure();
315 pjm2 1.11 return;
316     }
317 pjm2 1.1
318     # Now run through the configuration process.
319     my($response);
320    
321     print $sock "HEARTBEAT\n";
322     $response = <$sock>;
323 pjm2 1.23 return if (!defined $response);
324     chop $response;
325 pjm2 1.1 if (!$response eq "OK") {
326     close($sock);
327     print "Server gave wrong response to HEARTBEAT: $response\n";
328 pjm2 1.27 &tcp_configure();
329 pjm2 1.1 return;
330     }
331    
332     print $sock "CONFIG\n";
333     $response = <$sock>;
334 pjm2 1.23 return if (!defined $response);
335     chop $response;
336 pjm2 1.1 if (!$response eq "OK") {
337     close($sock);
338     print "Server gave wrong response to CONFIG: $response\n";
339 pjm2 1.27 &tcp_configure();
340 pjm2 1.1 return;
341     }
342    
343     print $sock "$file_list\n";
344     $response = <$sock>;
345 pjm2 1.23 return if (!defined $response);
346     chop $response;
347 pjm2 1.1 if (!$response eq "OK") {
348     close($sock);
349     print "Server gave wrong response to file list: $response\n";
350 pjm2 1.27 &tcp_configure();
351 pjm2 1.1 return;
352     }
353    
354     print $sock "$last_modified\n";
355     $response = <$sock>;
356 pjm2 1.23 return if (!defined $response);
357     chop $response;
358 pjm2 1.1 if ($response eq "ERROR") {
359     close($sock);
360 pjm2 1.27 print "Server configuration changed. Reconfiguring with filter manager.\n";
361 pjm2 1.1 &tcp_configure();
362     return;
363     }
364     if (!$response eq "OK") {
365     close($sock);
366     print "Server gave wrong response to HEARTBEAT: $response\n";
367 pjm2 1.27 &tcp_configure();
368 pjm2 1.1 return;
369     }
370    
371     print $sock "ENDHEARTBEAT\n";
372 pjm2 1.23 $response = <$sock>;
373     return if (!defined $response);
374 pjm2 1.1 chop $response;
375     if (!$response eq "OK") {
376     close($sock);
377     print "Server gave wrong response to ENDHEARTBEAT: $response\n";
378 pjm2 1.27 &tcp_configure();
379 pjm2 1.1 return;
380     }
381    
382     close($sock);
383     print "^";
384 tdb 1.25
385     return;
386     }
387    
388     #-----------------------------------------------------------------------
389     # write_pid
390     # Writes the PID (process ID) of this instance to $pidfile.
391     # This is then used by a seperate script to check (and restart) ihost.
392     #-----------------------------------------------------------------------
393     sub write_pid() {
394     open PID, ">$pidfile";
395     print PID $$;
396     close PID;
397 pjm2 1.12
398     return;
399 pjm2 1.1 }