ViewVC Help
View File | Revision Log | Show Annotations | Revision Graph | Root Listing
root/i-scream/projects/cms/source/host/ihost-perl/ihost.pl
Revision: 1.28
Committed: Wed Mar 14 18:48:33 2001 UTC (23 years, 8 months ago) by pjm2
Content type: text/plain
Branch: MAIN
Changes since 1.27: +129 -97 lines
Log Message:
Added retry waits to the configuration process.

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 pjm2 1.28 # $Author: pjm2 $
13     # $Id: ihost.pl,v 1.27 2001/03/14 18:00:16 pjm2 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 1.28 $retry_wait
38 pjm2 1.1 );
39    
40     if (@ARGV != 2) {
41     die "Usage: ihost.pl [i-scream filter manager] [TCP port]\n";
42     }
43    
44     $filter_manager_addr = $ARGV[0];
45     $filter_manager_port = $ARGV[1];
46    
47     $seq_no = 1;
48 pjm2 1.28 $retry_wait = 6;
49 pjm2 1.1
50 tdb 1.25 # write our PID to a file
51 tdb 1.26 $pidfile = "/var/tmp/ihost.pid";
52 tdb 1.25 &write_pid();
53    
54 pjm2 1.1 &tcp_configure();
55     &send_udp_packet();
56    
57     $last_udp_time = time;
58     $last_tcp_time = time;
59     while (1) {
60 pjm2 1.9 my($time) = time;
61     if ($time >= $last_udp_time + $udp_update_time) {
62 pjm2 1.1 &send_udp_packet();
63 pjm2 1.9 $last_udp_time = $time;
64 pjm2 1.1 }
65 pjm2 1.9 if ($time >= $last_tcp_time + $tcp_update_time) {
66 pjm2 1.1 &send_tcp_heartbeat();
67 pjm2 1.9 $last_tcp_time = $time;
68 pjm2 1.1 }
69 pjm2 1.9 my($next_udp) = $udp_update_time - $time + $last_udp_time;
70     my($next_tcp) = $tcp_update_time - $time + $last_tcp_time;
71     my($delay);
72     if ($next_udp < $next_tcp) {
73     $delay = $next_udp
74     }
75     else {
76     $delay = $next_tcp;
77     }
78     `sleep $delay`;
79 pjm2 1.1 }
80    
81 tdb 1.25 # we'll probably never get here...
82     `rm -f $pidfile`;
83 pjm2 1.1 exit(0);
84    
85 pjm2 1.12
86     #-----------------------------------------------------------------------
87 pjm2 1.28 # wait_then_retry
88     # Waits for the period of time specified in $retry_wait, then attempts
89     # to reconfigure with the server.
90     #-----------------------------------------------------------------------
91     sub wait_then_retry() {
92     print "Will retry configuration with filter manager in $retry_wait seconds.\n";
93     `sleep $retry_wait`;
94     }
95    
96    
97     #-----------------------------------------------------------------------
98 pjm2 1.12 # tcp_configure
99     # Establishes a TCP connection to the specified i-scream filter manager.
100     # The host then requests details from the server, such as the intervals
101     # at which to send UDP packets.
102     #-----------------------------------------------------------------------
103 pjm2 1.1 sub tcp_configure() {
104    
105 pjm2 1.28 while (1 == 1) {
106     my($sock) = new IO::Socket::INET(
107     PeerAddr => $filter_manager_addr,
108     PeerPort => $filter_manager_port,
109     Proto => 'tcp'
110     ) or die "Cannot connect!";
111     if (!defined $sock) {
112     print "IHOST ERROR: Could not connect to $filter_manager_addr:$filter_manager_port.\n";
113     print "Please check that there is an i-scream server at this address.\n";
114     close($sock);
115     wait_then_retry();
116     next;
117     }
118    
119     # Now run through the configuration process...
120     my($response);
121    
122     print $sock "STARTCONFIG\n";
123     $response = <$sock>;
124     if (!chop $response eq "OK") {
125     print "The i-scream server rejected the STARTCONFIG command. Terminated.";
126     close($sock);
127     wait_then_retry();
128     next;
129     }
130    
131     print "Config started okay.\n";
132    
133     print $sock "LASTMODIFIED\n";
134     $response = <$sock>;
135     chop $response;
136     $last_modified = $response;
137    
138     print "Config last modified: ". (scalar localtime $last_modified/1000) . "\n";
139    
140     print $sock "FILELIST\n";
141     $response = <$sock>;
142     chop $response;
143     $file_list = $response;
144    
145     print "File list obtained: $file_list\n";
146    
147     print $sock "FQDN\n";
148     $response = <$sock>;
149     chop $response;
150     $fqdn = $response;
151    
152     print "FQDN returned: $fqdn\n";
153    
154     print $sock "UDPUpdateTime\n";
155     $response = <$sock>;
156     chop $response;
157     $udp_update_time = $response;
158    
159     print $sock "TCPUpdateTime\n";
160     $response = <$sock>;
161     chop $response;
162     $tcp_update_time = $response;
163    
164     print "UDP packet period: $udp_update_time seconds.\nTCP heartbeat period: $tcp_update_time seconds.\n";
165    
166     print $sock "ENDCONFIG\n";
167     $response = <$sock>;
168     chomp $response;
169     if (!$response eq "OK") {
170     print "ENDCONFIG command to server failed. Terminated.\n";
171     close($sock);
172     wait_then_retry();
173     next;
174     }
175    
176     print "Config ended.\n";
177    
178     print $sock "FILTER\n";
179     $response = <$sock>;
180     unless (defined($response)) {
181     print "Failed: Could not get a filter address from the filter manager.\n";
182     close($sock);
183     wait_then_retry();
184     next;
185     }
186     chop $response;
187     $response =~ /(.*);(.*);(.*)/;
188     if ($response eq "ERROR") {
189     print "There are no active configured filters for your host.\n";
190     close($sock);
191     wait_then_retry();
192     next;
193     }
194     ($filter_addr, $udp_port, $tcp_port) = ($1, $2, $3);
195     unless (defined($filter_addr) && defined($udp_port) && defined($tcp_port)) {
196     print "Failed: Filter address response from server did not make sense: $response\n";
197     close($sock);
198     wait_then_retry();
199     next;
200     }
201    
202     print "Got filter data ($filter_addr, $udp_port, $tcp_port)\n";
203    
204     print $sock "END\n";
205     $response = <$sock>;
206     chop $response;
207     if ($response eq "OK") {
208     print "Host successfully configured via TCP.\n"
209     }
210     else {
211     print "The server failed the host configuration on the END command.";
212     close($sock);
213     wait_then_retry();
214     next;
215     }
216 pjm2 1.1
217 pjm2 1.28 close($sock);
218 pjm2 1.1
219 pjm2 1.28 print "Configuration finished sucessfully!\n";
220     last;
221 pjm2 1.1 }
222 pjm2 1.12 return;
223 pjm2 1.1 }
224    
225 pjm2 1.12
226    
227    
228     #-----------------------------------------------------------------------
229     # send_udp_packet
230     # Sends a UDP packet to an i-scream filter.
231     # The packet contains XML markup describing some of the machine's state.
232     # Receipt of UDP packets is not guaranteed.
233     #-----------------------------------------------------------------------
234 pjm2 1.1 sub send_udp_packet() {
235    
236     my(@statgrab) = `./statgrab.pl`;
237     my(%packet);
238 pjm2 1.13 for (my($i) = 0; $i <= $#statgrab; $i++) {
239 pjm2 1.1 $statgrab[$i] =~ /^([^\s]*) (.*)$/;
240     $packet{$1} = $2;
241     }
242    
243     my($date) = time;
244    
245     my($disk_info) = "<disk>";
246     my($i) = 0;
247     while (defined $packet{"packet.disk.p$i.attributes.mount"}) {
248 pjm2 1.15 $disk_info .= "<p$i";
249     $disk_info .= " name=\"" . $packet{"packet.disk.p$i.attributes.name"} . "\"";
250     $disk_info .= " kbytes=\"" . $packet{"packet.disk.p$i.attributes.kbytes"} . "\"";
251     $disk_info .= " used=\"" . $packet{"packet.disk.p$i.attributes.used"} . "\"";
252     $disk_info .= " avail=\"" . $packet{"packet.disk.p$i.attributes.avail"} . "\"";
253     $disk_info .= " mount=\"" . $packet{"packet.disk.p$i.attributes.mount"} . "\"";
254     $disk_info .= "></p$i>";
255 pjm2 1.1 ++$i;
256     }
257     $disk_info .= "</disk>";
258 pjm2 1.13
259 tdb 1.19 my($ip) = inet_ntoa(scalar(gethostbyname(hostname())) || 'localhost');
260 pjm2 1.1
261     # Build the XML packet this way, as we can clearly
262     # see the structure and contents... I like this ;-)
263 pjm2 1.12 # [Note that the server rejects UDP packets that are
264     # larger than 8196 bytes]
265 pjm2 1.1 my($xml) = <<EOF;
266    
267 tdb 1.19 <packet seq_no="$seq_no" machine_name="$fqdn" date="$date" type="data" ip="$ip">
268 pjm2 1.1 <load>
269     <load1>$packet{"packet.load.load1"}</load1>
270     <load5>$packet{"packet.load.load5"}</load5>
271     <load15>$packet{"packet.load.load15"}</load15>
272     </load>
273     <os>
274     <name>$packet{"packet.os.name"}</name>
275     <release>$packet{"packet.os.release"}</release>
276     <platform>$packet{"packet.os.platform"}</platform>
277     <sysname>$packet{"packet.os.sysname"}</sysname>
278     <version>$packet{"packet.os.version"}</version>
279 tdb 1.10 <uptime>$packet{"packet.os.uptime"}</uptime>
280 pjm2 1.1 </os>
281     <users>
282     <count>$packet{"packet.users.count"}</count>
283     <list>$packet{"packet.users.list"}</list>
284     </users>
285     <processes>
286     <total>$packet{"packet.processes.total"}</total>
287     <sleeping>$packet{"packet.processes.sleeping"}</sleeping>
288     <zombie>$packet{"packet.processes.zombie"}</zombie>
289     <stopped>$packet{"packet.processes.stopped"}</stopped>
290     <cpu>$packet{"packet.processes.cpu"}</cpu>
291     </processes>
292     <cpu>
293     <idle>$packet{"packet.cpu.idle"}</idle>
294     <user>$packet{"packet.cpu.user"}</user>
295     <kernel>$packet{"packet.cpu.kernel"}</kernel>
296     <iowait>$packet{"packet.cpu.iowait"}</iowait>
297     <swap>$packet{"packet.cpu.swap"}</swap>
298     </cpu>
299     <memory>
300 pjm2 1.13 <total>$packet{"packet.memory.total"}</total>
301 pjm2 1.1 <free>$packet{"packet.memory.free"}</free>
302     </memory>
303     <swap>
304 pjm2 1.13 <total>$packet{"packet.swap.total"}</total>
305     <free>$packet{"packet.swap.free"}</free>
306 pjm2 1.1 </swap>
307     $disk_info
308     </packet>
309    
310     EOF
311    
312 pjm2 1.12 # Make the packet smaller by stripping out newlines and leading spaces.
313 pjm2 1.1 $xml =~ s/\n\s*//g;
314 tdb 1.19
315 pjm2 1.1 my($sock) = new IO::Socket::INET (
316     PeerPort => $udp_port,
317     PeerAddr => $filter_addr,
318     Proto => 'udp'
319 pjm2 1.21 ) or die "Could not send UDP: $!\n";
320 pjm2 1.1
321     print $sock $xml or die "Could not send UDP packet: $!\n";
322     close($sock);
323     $seq_no++;
324     print "-";
325 pjm2 1.12
326     return;
327 pjm2 1.1 }
328    
329 pjm2 1.12
330    
331    
332     #-----------------------------------------------------------------------
333     # send_tcp_heartbeat
334     # Establishes a TCP connection to an i-scream filter.
335     # The heartbeat is used as a guaranteed "I'm alive" delivery mechanism.
336     #-----------------------------------------------------------------------
337 pjm2 1.1 sub send_tcp_heartbeat() {
338    
339     my($sock) = new IO::Socket::INET(
340     PeerAddr => $filter_addr,
341     PeerPort => $tcp_port,
342     Proto => 'tcp'
343 pjm2 1.21 ) or return;
344 pjm2 1.11 if (!defined $sock) {
345     print "IHOST WARNING: Failed to deliver a heartbeat to the i-scream filter.\n";
346 pjm2 1.27 &tcp_configure();
347 pjm2 1.11 return;
348     }
349 pjm2 1.1
350     # Now run through the configuration process.
351     my($response);
352    
353     print $sock "HEARTBEAT\n";
354     $response = <$sock>;
355 pjm2 1.23 return if (!defined $response);
356     chop $response;
357 pjm2 1.1 if (!$response eq "OK") {
358     close($sock);
359     print "Server gave wrong response to HEARTBEAT: $response\n";
360 pjm2 1.27 &tcp_configure();
361 pjm2 1.1 return;
362     }
363    
364     print $sock "CONFIG\n";
365     $response = <$sock>;
366 pjm2 1.23 return if (!defined $response);
367     chop $response;
368 pjm2 1.1 if (!$response eq "OK") {
369     close($sock);
370     print "Server gave wrong response to CONFIG: $response\n";
371 pjm2 1.27 &tcp_configure();
372 pjm2 1.1 return;
373     }
374    
375     print $sock "$file_list\n";
376     $response = <$sock>;
377 pjm2 1.23 return if (!defined $response);
378     chop $response;
379 pjm2 1.1 if (!$response eq "OK") {
380     close($sock);
381     print "Server gave wrong response to file list: $response\n";
382 pjm2 1.27 &tcp_configure();
383 pjm2 1.1 return;
384     }
385    
386     print $sock "$last_modified\n";
387     $response = <$sock>;
388 pjm2 1.23 return if (!defined $response);
389     chop $response;
390 pjm2 1.1 if ($response eq "ERROR") {
391     close($sock);
392 pjm2 1.27 print "Server configuration changed. Reconfiguring with filter manager.\n";
393 pjm2 1.1 &tcp_configure();
394     return;
395     }
396     if (!$response eq "OK") {
397     close($sock);
398     print "Server gave wrong response to HEARTBEAT: $response\n";
399 pjm2 1.27 &tcp_configure();
400 pjm2 1.1 return;
401     }
402    
403     print $sock "ENDHEARTBEAT\n";
404 pjm2 1.23 $response = <$sock>;
405     return if (!defined $response);
406 pjm2 1.1 chop $response;
407     if (!$response eq "OK") {
408     close($sock);
409     print "Server gave wrong response to ENDHEARTBEAT: $response\n";
410 pjm2 1.27 &tcp_configure();
411 pjm2 1.1 return;
412     }
413    
414     close($sock);
415     print "^";
416 tdb 1.25
417     return;
418     }
419    
420     #-----------------------------------------------------------------------
421     # write_pid
422     # Writes the PID (process ID) of this instance to $pidfile.
423     # This is then used by a seperate script to check (and restart) ihost.
424     #-----------------------------------------------------------------------
425     sub write_pid() {
426     open PID, ">$pidfile";
427     print PID $$;
428     close PID;
429 pjm2 1.12
430     return;
431 pjm2 1.1 }