source: trunk/cluster_status.php@ 38

Last change on this file since 38 was 38, checked in by gegorbet, 6 years ago

add taito-local cluster status logic
File size: 12.9 KB
Line 
1<?php
2
3$us3bin = exec( "ls -d ~us3/lims/bin" );
4include "$us3bin/listen-config.php";
5
6if ( ! preg_match( "/_local/", $class_dir ) )
7{
8 $xml = get_data();
9
10 if ( $xml != "" )
11 parse( $xml );
12}
13
14$data = array();
15
16local_status();
17
18foreach ( $data as $item )
19{
20 if ( ! preg_match( "/error/", $item[ 'running' ] ) )
21 update( $item[ 'cluster' ], $item[ 'queued' ], $item[ 'status' ], $item[ 'running' ] );
22}
23
24//exit();
25exit(0);
26
27// Get the cluster status
28
29function get_data()
30{
31 global $self;
32 $url = "http://community.ucs.indiana.edu:19444/orps-service/XML/gateway/ultrascan";
33
34 try
35 {
36 $post = new HttpRequest( $url, HttpRequest::METH_GET );
37 $http = $post->send();
38 $xml = $post->getResponseBody();
39 }
40 catch ( HttpException $e )
41 {
42// write_log( "$self: Cluster Status not available" );
43 return "";
44 }
45
46 return $xml;
47}
48
49// Parse the xml
50
51function parse( $xml )
52{
53 global $data;
54
55 $data = array();
56
57 $x = new XML_Array( $xml );
58 $d = $x->ReturnArray();
59
60 if ( ! isset( $d[ 'summaries' ] ) ) exit(); // Bad input
61
62 foreach ( $d[ 'summaries' ] as $item )
63 {
64 $a = Array();
65
66
67 $a[ 'queued' ] = $item[ 'waitingJobs' ];
68 $a[ 'running' ] = $item[ 'runningJobs' ];
69
70 if ( $a[ 'queued' ] == "" || $a[ 'queued' ] < 0 ) $a[ 'queued' ] = 0;
71 if ( $a[ 'running' ] == "" || $a[ 'running' ] < 0 ) $a[ 'running' ] = 0;
72
73 $clusterParts = explode( ".", $item[ 'resourceId' ] );
74 $cluster = preg_replace( '/\d+$/', "", $clusterParts[ 0 ] );
75
76 if ( $cluster == 'uthscsa-bcf' ) $cluster = 'bcf';
77 if ( $cluster == 'uthscsa-alamo' ) $cluster = 'alamo';
78
79 $a[ 'cluster' ] = $cluster;
80
81 switch ( $item[ 'resourceStatus' ] )
82 {
83 case 'UP' :
84 $status = "up";
85 break;
86
87 case 'DOWN' :
88 $status = "down";
89 break;
90
91 case 'WARN' :
92 $status = "warn";
93 break;
94
95 case 'FAILED' :
96 default :
97 $status = "unknown";
98 break;
99 }
100
101 $a[ 'status' ] = $status;
102
103 $data[] = $a;
104 }
105}
106
107// Put it in the DB
108
109function update( $cluster, $queued, $status, $running )
110{
111 global $dbhost;
112 global $guser;
113 global $gpasswd;
114 global $gDB;
115 global $self;
116//echo " $cluster $queued, $status, $running\n";
117
118 $gfac_link = mysqli_connect( $dbhost, $guser, $gpasswd, $gDB );
119
120 if ( ! $gfac_link )
121 {
122 write_log( "$self: Could not connect to DB $gDB" );
123 echo "Could not connect to DB $gDB.\n";
124 exit();
125 }
126
127 $query = "SELECT * FROM cluster_status WHERE cluster='$cluster'";
128 $result = mysqli_query( $gfac_link, $query );
129
130 if ( ! $result )
131 {
132 write_log( "$self: Query failed $query - " . mysqli_error( $gfac_link ) );
133 echo "$self: Query failed $query - " . mysqli_error( $gfac_link ) . "\n";
134 exit();
135 }
136
137 $rows = mysqli_num_rows( $result );
138
139 if ( $rows == 0 ) // INSERT
140 {
141 $query = "INSERT INTO cluster_status SET " .
142 "cluster='$cluster', " .
143 "queued=$queued, " .
144 "running=$running, " .
145 "status='$status'";
146 }
147 else // UPDATE
148 {
149 $query = "UPDATE cluster_status SET " .
150 "queued=$queued, " .
151 "running=$running, " .
152 "status='$status' " .
153 "WHERE cluster='$cluster'";
154 }
155
156 $result = mysqli_query( $gfac_link, $query );
157
158 if ( ! $result )
159 {
160 write_log( "$self: Query failed $query - " . mysqli_error( $gfac_link ) );
161 echo "$self: Query failed $query - " . mysqli_error( $gfac_link ) . "\n";
162 }
163}
164
165// Get local cluster status
166
167function local_status()
168{
169 global $self;
170 global $data;
171 global $dbhost;
172 global $org_domain;
173 global $class_dir;
174
175 if ( preg_match( "/_local/", $class_dir ) )
176 {
177 if ( preg_match( "/attlocal/", $org_domain ) )
178 $clusters = array( "us3iab-devel", "alamo-local", "taito-local" );
179 else
180 $clusters = array( "us3iab-node0", "taito-local");
181 }
182 else
183 {
184 $clusters = array( "lonestar5", "comet",
185 "stampede2", "jetstream" );
186 }
187
188 foreach ( $clusters as $clname )
189 {
190 $a = Array();
191//echo "$self: clname=$clname\n";
192
193 switch( $clname )
194 {
195 case 'us3iab-node0':
196 case 'us3iab-node1':
197 case 'us3iab-devel':
198 {
199 $qstat = `/usr/bin/qstat -B 2>&1|tail -1`;
200
201 $sparts = preg_split( '/\s+/', $qstat );
202 $que = $sparts[ 3 ];
203 $run = $sparts[ 4 ];
204 $sta = $sparts[ 10 ];
205 if ( $sta == "Active" )
206 $sta = "up";
207 else
208 $sta = "down";
209 break;
210 }
211 case 'alamo-local':
212 case 'alamo':
213 {
214 $host = "us3@alamo.uthscsa.edu";
215 $qstat = `ssh $host '/usr/bin/qstat -B 2>&1|tail -1'`;
216 $sparts = preg_split( '/\s+/', $qstat );
217 $que = $sparts[ 3 ];
218 $run = $sparts[ 4 ];
219 $sta = $sparts[ 10 ];
220 if ( $sta == "Active" )
221 $sta = "up";
222 else
223 $sta = "down";
224 break;
225 }
226 case 'jacinto':
227 {
228 $host = "us3@jacinto.uthscsa.edu";
229 $qstat = `ssh $host '/opt/torque/bin/qstat -B 2>&1|tail -1'`;
230 $sparts = preg_split( '/\s+/', $qstat );
231 $que = $sparts[ 3 ];
232 $run = $sparts[ 4 ];
233 $sta = $sparts[ 9 ];
234 if ( $sta == "Active" )
235 $sta = "up";
236 else
237 $sta = "down";
238 break;
239 }
240 case 'stampede2':
241 {
242 $host = "us3@stampede2.tacc.utexas.edu";
243 $qstat = `ssh $host '/usr/local/bin/showq 2>/dev/null|grep "Total Jobs"'`;
244 $sparts = preg_split( '/\s+/', $qstat );
245 $tot = $sparts[ 2 ];
246 $run = $sparts[ 5 ];
247 $que = $sparts[ 8 ];
248 $sta = "up";
249 if ( $tot == '' || $tot == '0' )
250 $sta = "down";
251 break;
252 }
253 case 'lonestar5':
254 {
255 $host = "us3@ls5.tacc.utexas.edu";
256 $qstat = `ssh $host '/usr/local/bin/showq 2>/dev/null|grep "Total Jobs"'`;
257 $sparts = preg_split( '/\s+/', $qstat );
258 $tot = $sparts[ 2 ];
259 $run = '0';
260 $que = '0';
261 $sta = "up";
262 if ( $tot == '' || $tot == '0' )
263 {
264 $sta = "down";
265 }
266 else
267 {
268 $run = $sparts[ 5 ];
269 $que = $sparts[ 8 ];
270// $que = $sparts[ 11 ];
271 }
272 break;
273 }
274 case 'taito-local':
275 {
276 $host = "rb_2001068_taito01@taito.csc.fi";
277 //$qstat = `ssh $host '/usr/bin/sinfo -s -p compute -o "%a %F" |tail -1'`;
278 $qstat = `ssh -i /home/us3/.ssh/id_rsa_taito_robot $host '/homeappl/home/rb_2001068_taito01/scripts/cstat 2>&1'`;
279 $sparts = preg_split( '/\s+/', $qstat );
280 $tot = $sparts[ 1 ];
281 $run = '0';
282 $que = '0';
283 $sta = "up";
284 if ( $tot == '' || $tot == '0' )
285 {
286 $sta = "down";
287 }
288 else
289 {
290 $run = $sparts[ 3 ];
291 $que = $sparts[ 5 ];
292 }
293 break;
294 }
295 case 'comet':
296 {
297 $host = "us3@comet.sdsc.edu";
298 //$qstat = `ssh $host '/usr/bin/sinfo -s -p compute -o "%a %F" |tail -1'`;
299 $qstat = `ssh $host '/home/us3/scripts/cstat 2>&1'`;
300 $sparts = preg_split( '/\s+/', $qstat );
301 $tot = $sparts[ 1 ];
302 $run = '0';
303 $que = '0';
304 $sta = "up";
305 if ( $tot == '' || $tot == '0' )
306 {
307 $sta = "down";
308 }
309 else
310 {
311 $run = $sparts[ 3 ];
312 $que = $sparts[ 5 ];
313 }
314 break;
315 }
316 case 'jureca':
317 {
318 $host = "swus1@jureca.fz-juelich.de";
319 $qstat = `ssh $host '~swus1/scripts/qstat-jureca 2>&1'`;
320 $sparts = preg_split( '/\s+/', $qstat );
321 $sta = $sparts[ 0 ];
322 $run = $sparts[ 1 ];
323 $que = $sparts[ 2 ];
324 break;
325 }
326 case 'jetstream-local':
327 case 'jetstream':
328 {
329 $host = "us3@js-169-137.jetstream-cloud.org";
330// $qstat = `ssh $host '/usr/bin/sinfo -s -p batch -o "%a %F" |tail -1'`;
331 $qstat = `ssh $host '/home/us3/bin/clusstat |tail -n 1'`;
332 $sparts = preg_split( '/\s+/', $qstat );
333 $sta = $sparts[ 0 ];
334 $knts = $sparts[ 1 ];
335 $sparts = preg_split( '/\//', $knts );
336 $run = $sparts[ 0 ];
337 $que = $sparts[ 2 ];
338 $tot = $sparts[ 3 ];
339 if ( $sta == "" )
340 $sta = "down";
341 break;
342 }
343 }
344
345 if ( $sta == "" )
346 $sta = "down";
347
348 if ( $sta == "down" )
349 {
350 $que = "0";
351 $run = "0";
352 }
353
354 // Insure queued,running counts are numeric
355 $que_s = $que;
356 $run_s = $run;
357 $que_e = preg_replace( '/=/', "", $que_s );
358 $run_e = preg_replace( '/=/', "", $run_s );
359 $que = intval( $que_e );
360 $run = intval( $run_e );
361if($que!=$que_s || $run!=$run_s)
362 echo "$self: *** que s,e,i $que_s $que_e $que run s,e,i $run_s $run_e $run\n";
363
364 // Save cluster status values
365 $a[ 'cluster' ] = $clname;
366 $a[ 'queued' ] = $que;
367 $a[ 'running' ] = $run;
368 $a[ 'status' ] = $sta;
369echo "$self: $clname $que $run $sta\n";
370
371 $data[] = $a;
372
373 if ( $clname == 'alamo' ||
374 $clname == 'jacinto' ||
375 $clname == 'jetstream' )
376 {
377 $a[ 'cluster' ] = $clname . "-local";
378 $data[] = $a;
379 }
380 }
381}
382
383class XML_Array
384{
385 var $_data = Array();
386 var $_name = Array();
387 var $_rep = Array();
388 var $_parser = 0;
389 var $_level = 0;
390 var $_index = 0;
391
392 function XML_Array( &$data )
393 {
394 $this->_parser = xml_parser_create();
395
396 xml_set_object ( $this->_parser, $this );
397 xml_parser_set_option ( $this->_parser, XML_OPTION_CASE_FOLDING, false );
398 xml_set_element_handler ( $this->_parser, "_startElement", "_endElement" );
399 xml_set_character_data_handler( $this->_parser, "_cdata" );
400
401 $this->_data = array();
402 $this->_level = 0;
403
404 if ( ! xml_parse( $this->_parser, $data, true ) )
405 return false;
406
407 xml_parser_free( $this->_parser );
408 }
409
410 function & ReturnArray()
411 {
412 return $this->_data[ 0 ];
413 }
414
415 function _startElement( $parser, $name, $attrs )
416 {
417 if ( $name == "resourceHealth" )
418 {
419## $name .= $this->_index;
420 $this->_index++;
421 }
422
423 if ( ! isset( $this->_rep[ $name ] ) ) $this->_rep[ $name ] = 0;
424
425 $this->_addElement( $name, $this->_data[ $this->_level ], $attrs );
426 $this->_name[ $this->_level ] = $name;
427 $this->_level++;
428 }
429
430 function _endElement( $parser, $name )
431 {
432 if ( isset( $this->_data[ $this->_level ] ) )
433 {
434 $this->_addElement( $this->_name[ $this->_level - 1 ],
435 $this->_data[ $this->_level - 1 ],
436 $this->_data[ $this->_level ]
437 );
438 }
439
440 unset( $this->_data[ $this->_level ] );
441 $this->_level--;
442 $this->_rep[ $name ]++;
443 }
444
445 function _cdata( $parser, $data )
446 {
447 if ( $this->_name[ $this->_level - 1 ] )
448 {
449 $this->_addElement( $this->_name[ $this->_level - 1 ],
450 $this->_data[ $this->_level - 1 ],
451 str_replace( array( "&gt;", "&lt;","&quot;", "&amp;" ),
452 array( ">" , "<" , '"' , "&" ),
453 $data
454 )
455 );
456 }
457 }
458
459 function _addElement( &$name, &$start, $add = array() )
460 {
461 if ( ( sizeof( $add ) == 0 && is_array( $add ) ) || ! $add )
462 {
463 if ( ! isset( $start[ $name ] ) ) $start[ $name ] = '';
464 $add = '';
465 }
466
467 $update = &$start[ $name ];
468
469 if ( is_array( $add) &&
470 is_array( $update ) ) $update += $add;
471 elseif ( is_array( $update ) ) return;
472 elseif ( is_array( $add ) ) $update = $add;
473 elseif ( $add ) $update .= $add;
474 }
475}
476?>
Note: See TracBrowser for help on using the repository browser.