File:
1.09.4a/server/bin/tester.php (
View as Code)
1: 2: /* -------------------------------------------------------------
3: This file is part of FreeNATS
4:
5: FreeNATS is (C) Copyright 2008 PurplePixie Systems
6:
7: FreeNATS is free software: you can redistribute it and/or modify
8: it under the terms of the GNU General Public License as published by
9: the Free Software Foundation, either version 3 of the License, or
10: (at your option) any later version.
11:
12: FreeNATS is distributed in the hope that it will be useful,
13: but WITHOUT ANY WARRANTY; without even the implied warranty of
14: MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15: GNU General Public License for more details.
16:
17: You should have received a copy of the GNU General Public License
18: along with FreeNATS. If not, see www.gnu.org/licenses
19:
20: For more information see www.purplepixie.org/freenats
21: -------------------------------------------------------------- */
22:
23: if ((isset($argc))&&(isset($argv))) // specific node or all nodes
24: {
25: if ($argc>1)
26: {
27: $nfilter=$argv[1];
28: }
29: else $nfilter="";
30: }
31:
32: require("include.php");
33:
34: $dbt="";
35:
36: function db($txt,$nl=true) // debug text
37: {
38: global $dbt;
39: echo $txt;
40: $dbt.=$txt;
41: if ($nl)
42: {
43: echo "\n";
44: $dbt.="
\n";
45: }
46: }
47:
48: $NATS->Start();
49: if ($nfilter!="") $st=": Node ".$nfilter;
50: else $st="";
51: db("NATS Tester Script Starting".$st);
52:
53: $highalertlevel=-1;
54: $talertc=0;
55:
56: // check if already running
57: $still_running=false;
58: $cq="SELECT trid,startx FROM fntestrun WHERE fnode=\"".ss($nfilter)."\" AND finishx=0 LIMIT 0,1";
59: $cr=$NATS->DB->Query($cq);
60:
61: if ($runrow=$NATS->DB->Fetch_Array($cr))
62: { // yes there is a testrun session for this node(s)
63: $timelimit=$NATS->Cfg->Get("test.session.limit",60*60);
64: if ( (!is_numeric($timelimit)) || ($timelimit<0) ) $timelimit=60*60; // bogus config value
65:
66: // n.b. a timelimit of 0 means the session will never expire so...
67: if ( ($timelimit>0) && ((time()-$runrow['startx'])>$timelimit) )
68: {
69: // valid time limit and the difference is more than the limit so close it
70: $uq="UPDATE fntestrun SET finishx=1 WHERE trid=".$runrow['trid'];
71: $NATS->DB->Query($uq);
72: if ($NATS->DB->Affected_Rows()>0)
73: {
74: $NATS->Event("Tester Already Running: Cleared",3,"Tester","Stale");
75: db("Tester Already Running: Cleared");
76: }
77: else $NATS->Event("Tester Already Running: Failed to Clear",1,"Tester","Stale"); // weirdness... run anyway
78: }
79: else $still_running=true; // either never timesout or newer than timelimit ago
80: }
81: $NATS->DB->Free($cr);
82:
83: // and if it is then don't continue
84: if ($still_running)
85: {
86: $NATS->Event("Tester Already Running: Aborted",1,"Tester","Error");
87: db("Tester Already Running: Aborted");
88: $NATS->Stop();
89: exit();
90: }
91:
92:
93: $gq="INSERT INTO fntestrun(startx,fnode) VALUES(".time().",\"".ss($nfilter)."\")";
94: $NATS->DB->Query($gq);
95: $trid=$NATS->DB->Insert_Id();
96: db("Test ID: ".$trid." (Started at ".nicedt(time()).")");
97: $NATS->Event("Tester ".$trid." Started",5,"Tester","Start");
98:
99: db(" ");
100:
101: // Find node to test - must be enabled, have id if set, and be due to be tested (nextrunx)
102:
103: $q="SELECT * FROM fnnode WHERE nodeenabled=1";
104: if ($nfilter!="") $q.=" AND nodeid=\"".ss($nfilter)."\"";
105: $q.=" AND nextrunx<=".time();
106:
107: $r=$NATS->DB->Query($q);
108:
109:
110: while ($row=$NATS->DB->Fetch_Array($r))
111: {
112: $dotests=true;
113: $alertlevel=0;
114: $alerts=array();
115: $alertc=0;
116: db("NodeID: ".$row['nodeid']);
117: $NATS->Event("Tester ".$trid." Node ".$row['nodeid'],10,"Tester","Node");
118:
119: // Scheduling Test In Here - sets dotests to false and alertlevel to -1 untested
120: if ($row['scheduleid']!=0) // has a schedule
121: {
122: db(" Has Schedule: Yes - Checking");
123: $run=run_x_in_schedule(time(),$row['scheduleid']);
124: if (!$run)
125: {
126: db(" In Schedule: No - Skipping Tests");
127: $NATS->Event("Tester ".$trid." Skipped by Schedule",5,"Tester","Node");
128: $dotests=false;
129: $alertlevel=-1;
130: }
131: else db(" In Schedule: Yes");
132: }
133:
134: $eventdata=array( "nodeid" => $row['nodeid'], "in_schedule" => $dotests );
135: $NATS->EventHandler("node_test_start",$eventdata);
136:
137:
138: $ptr=0;
139: $pal=0;
140:
141:
142: // Update lastrun and nextrun regardless of dotests
143: $q="UPDATE fnnode SET lastrunx=".time().",nextrunx=".next_run_x($row['testinterval'])." WHERE nodeid=\"".ss($row['nodeid'])."\"";
144: $NATS->DB->Query($q);
145:
146: $pingpassed=false; // this will only be set to true if a test is done and passes - for the "child" nodes
147: if ($row['pingtest']&&$dotests)
148: {
149: db(" Ping Test: Yes");
150: $NATS->Event("Tester ".$trid." Pinging Node ".$row['nodeid'],10,"Tester","Ping");
151: $ptr=PingTest($row['hostname']);
152: $NATS->Event("Tester ".$trid." Ping Node ".$row['nodeid']." Returned ".$ptr,10,"Tester","Ping");
153: db(" Ping Returned: ".$ptr);
154: if ( ($ptr<=0) && ($NATS->Cfg->Get("test.icmp.attempts","2")>1) )
155: {
156: $att=$NATS->Cfg->Get("test.icmp.attempts","2");
157: for ($a=2; $a<=$att; $a++) // starting on second attempt
158: {
159: // try again...
160: test_sleep();
161: db(" Trying Ping Again - X".$a);
162: $NATS->Event("Tester ".$trid." Ping X".$a." Node ".$row['nodeid'],10,"Tester","Ping");
163: $ptr=PingTest($row['hostname']);
164: $NATS->Event("Tester ".$trid." Ping Node ".$row['nodeid']." Returned ".$ptr,10,"Tester","Ping");
165: db(" Ping Returned: ".$ptr);
166: if ($ptr>0) $a=$att+1; // break out of the loop
167: }
168: }
169:
170: if ($ptr<=0)
171: {
172: $alertlevel=2;
173: db(" Ping Test: Failed");
174: $alerts[$alertc++]="ping failed";
175: $pal=2;
176: }
177: else
178: {
179: db(" Ping Test: Passed");
180: $pingpassed=true;
181: }
182:
183: // pingtest output bodge
184: // is there a test entry for ICMP
185: $fq="SELECT localtestid FROM fnlocaltest WHERE nodeid=\"".$row['nodeid']."\" AND testtype=\"ICMP\"";
186: $fr=$NATS->DB->Query($fq);
187: $ltid_icmp="";
188: if ($irow=$NATS->DB->Fetch_Array($fr))
189: { // exists
190: $uq="UPDATE fnlocaltest SET alertlevel=".$pal.",lastrunx=".time().",lastvalue=".$ptr.",testrecord=1,testinterval=0 WHERE localtestid=".$irow['localtestid'];
191: $ltid_icmp=$irow['localtestid'];
192: //echo $uq;
193: $NATS->DB->Query($uq);
194: }
195: else
196: { // doesn't exist
197: $uq="INSERT INTO fnlocaltest(nodeid,testrecord,testinterval,testtype,alertlevel,lastrunx,lastvalue) VALUES(\"".$row['nodeid']."\",1,0,\"ICMP\",".$pal.",".time().",".$ptr.")";
198: //echo $uq;
199: $NATS->DB->Query($uq);
200: $ltid_icmp=$NATS->DB->Insert_Id();
201: }
202: $NATS->DB->Free($fr);
203:
204: // record the ICMP bodge-test here
205: $rq="INSERT INTO fnrecord(testid,recordx,testvalue,alertlevel,nodeid) VALUES(\"L".$ltid_icmp."\",".time().",".$ptr.",".$pal.",\"".$row['nodeid']."\")";
206: $NATS->DB->Query($rq);
207: //echo $rq." ".$NATS->DB->Affected_Rows()."\n";
208:
209: }
210: else
211: { // further ICMP bodge - update to -1 or do nothing if the test doesn't exist
212: $uq="UPDATE fnlocaltest SET alertlevel=-1,lastrunx=".time()." WHERE nodeid=\"".$row['nodeid']."\" AND testtype=\"ICMP\"";
213: $NATS->DB->Query($uq);
214: }
215:
216: if ($dotests&&($row['pingfatal'])&&($ptr<=0))
217: {
218: db(" Ping Fatal: Yes - Not Continuing");
219: $NATS->Event("Tester ".$trid." Ping Fatal for Node ".$row['nodeid'],10,"Tester","Ping");
220: $dotests=false;
221: }
222:
223: // do the tests - only actually exec if dotests true
224:
225: $first_test=true;
226:
227: db("Doing Local Tests");
228: $NATS->Event("Tester ".$trid." Testing Node ".$row['nodeid'],10,"Tester","Test");
229: $q="SELECT * FROM fnlocaltest WHERE nodeid=\"".$row['nodeid']."\" AND testtype!=\"ICMP\" AND testenabled=1 ORDER BY localtestid ASC";
230: $res=$NATS->DB->Query($q);
231: while ($lrow=$NATS->DB->Fetch_Array($res))
232: {
233: if ($lrow['nextrunx']<=time()) $testdue=true;
234: else $testdue=false;
235:
236: if ($first_test)
237: {
238: $first_test=false;
239: if ($row['pingtest']==1) test_sleep(); // sleep if has done a ping
240: }
241: else test_sleep();
242:
243: if ($testdue)
244: {
245:
246: $eventdata=array("nodeid"=>$row['nodeid'],"testid"=>"L".$lrow['testparam'],"testtype"=>$lrow['testtype']);
247: $NATS->EventHandler("localtest_start",$eventdata);
248:
249: db(" Test: ".$lrow['testtype']." (".$lrow['testparam'].")");
250:
251: // Build parameter array
252: $params=array();
253: $params[0]=$lrow['testparam']; // pass standard param in as 0
254: for ($a=1; $a<10; $a++)
255: {
256: $parstr="testparam".$a;
257: $params[$a]=$lrow[$parstr];
258: }
259:
260: if ($dotests)
261: {
262: $NATS->Event("Tester ".$trid." Node ".$row['nodeid']." Doing ".$lrow['testtype']."(".$lrow['testparam'].")",10,"Tester","Test");
263: $result=DoTest($lrow['testtype'],$lrow['testparam'],$row['hostname'],$lrow['timeout'],$params,$row['nodeid']);
264: $NATS->Event("Tester ".$trid." Node ".$row['nodeid']." Result ".$result." from ".$lrow['testtype']."(".$lrow['testparam'].")",10,"Tester","Test");
265: db(" Result: ".$result);
266: }
267: else $result=0;
268:
269: if ($dotests)
270: {
271: // evaluation
272: if ($lrow['simpleeval']==1) $lvl=SimpleEval($lrow['testtype'],$result);
273: else $lvl=nats_eval("L".$lrow['localtestid'],$result);
274: db(" Eval: ".$lvl);
275:
276: // put in the custom retries based on attempts here - we KNOW dotests is on so don't need to worry about untested status
277: $att=$lrow['attempts'];
278: if ( ($lvl!=0) && (is_numeric($att)) && ($att>1) )
279: {
280: for ($a=2; $a<=$att; $a++)
281: {
282: test_sleep();
283: db(" Test: ".$lrow['testtype']." (".$lrow['testparam'].") X".$a);
284: $NATS->Event("Tester ".$trid." Node ".$row['nodeid']." X".$a." Doing ".$lrow['testtype']."(".$lrow['testparam'].")",10,"Tester","Test");
285: $result=DoTest($lrow['testtype'],$lrow['testparam'],$row['hostname'],$lrow['timeout'],$params,$row['nodeid']);
286: db(" Result: ".$result);
287: if ($lrow['simpleeval']==1) $lvl=SimpleEval($lrow['testtype'],$result);
288: else $lvl=nats_eval("L".$lrow['localtestid'],$result);
289: db(" Eval: ".$lvl);
290: if ($lvl==0) $a=$att+1; // test passed
291: }
292: }
293:
294: // $lvl is now the last lvl regardless of where it came from
295:
296: if ($lvl>$alertlevel) $alertlevel=$lvl;
297: if ($lvl>0)
298: {
299: if ($lrow['testname']=="") $s=$lrow['testtype']."/".substr($lrow['testparam'],0,5)." ";
300: else $s=$lrow['testname']." ";
301: /*
302: if ($lvl>1) $s.=$NATS->Cfg->Get("site.text.failed","failed");
303: else $s.=$NATS->Cfg->Get("site.text.warning","warning");
304: */
305: $s.=oText($lvl);
306: // site.alert.showvalue -- includes value in alert messages if numeric
307: // site.alert.showtext -- includes value in alert messages if textual
308: if (is_numeric($result))
309: {
310: if ($NATS->Cfg->Get("site.alert.showvalue",0)==1) $s.=" (".$result.")";
311: }
312: else // non-numeric
313: {
314: if ($NATS->Cfg->Get("site.alert.showtext",0)==1) $s.=" (".$result.")";
315: }
316: $alerts[$alertc++]=$s;
317: }
318: } else $lvl=-1;
319:
320: // record it
321: if ($lrow['testrecord']==1)
322: {
323: $tid="L".$lrow['localtestid'];
324: $iq="INSERT INTO fnrecord(testid,nodeid,alertlevel,testvalue,recordx) VALUES(";
325: $iq.="\"".$tid."\",\"".$row['nodeid']."\",".$lvl.",".$result.",".time().")";
326: $NATS->DB->Query($iq);
327: db(" Recording Test");
328: }
329: if ((!isset($result))||(!is_numeric($result))) $result=0; // safety net
330:
331: // update localtest record
332: $uq="UPDATE fnlocaltest SET lastrunx=".time().",nextrunx=".next_run_x($lrow['testinterval']).",alertlevel=".$lvl.",lastvalue=".$result." WHERE localtestid=".$lrow['localtestid'];
333: $NATS->DB->Query($uq);
334:
335: $eventdata=array("nodeid"=>$row['nodeid'],"testid"=>"L".$lrow['testparam'],"testtype"=>$lrow['testtype'],"alertlevel"=>$lvl);
336: $NATS->EventHandler("localtest_finish",$eventdata);
337: }
338:
339: else // test not due so take pre-existing level for it
340: {
341: $lvl=$lrow['alertlevel'];
342: if (($lvl>0)&&($lvl>$alertlevel)) $alertlevel=$lvl;
343: }
344:
345:
346: }
347:
348: // Node-side testy magic
349: db("Nodeside Testing");
350: $freshdata=false;
351: if ( $dotests && ($row['nsenabled']==1) && ($row['nspullenabled']==1) ) // should be doing a pull
352: {
353: $pullalert=$row['nspullalert']; // what happened the last time we tried
354:
355: if ($row['nsnextx']<=time()) // the time is right
356: {
357: db(" Pulling Data");
358: $pull_result=$NATS->Nodeside_Pull($row['nodeid']);
359:
360: if ($pull_result===false) // Pull Failed
361: {
362: db(" Pull Failed");
363: $pullalert=1; // alert
364: $alerts[$alertc++]="pull failed";
365: $alertlevel=2;
366: }
367: else // Pull Worked
368: {
369: $freshdata=true;
370: $pullalert=0; // ok
371: db(" Pull Succeeded");
372: }
373:
374:
375: db(" Updating Pull nslast/nextx and nspullalert");
376: $uq="UPDATE fnnode SET nsnextx=".next_run_x($row['nsinterval']).",nspullalert=".$pullalert.",nslastx=".time()." WHERE nodeid=\"".$row['nodeid']."\"";
377: $NATS->DB->Query($uq);
378: if ($NATS->DB->Affected_Rows()<=0) db(" - Failed");
379: }
380: /*
381: // Process for alerts in here - whether pulled or not!
382: $tq="SELECT testtype,testname,alertlevel FROM fnnstest WHERE nodeid=\"".$row['nodeid']."\" AND testenabled=1 AND testalerts=1 AND alertlevel>0";
383: $tr=$NATS->DB->Query($tq);
384: while ($trow=$NATS->DB->Fetch_Array($tr))
385: {
386: if ($trow['testname']=="") $tname=$trow['testtype'];
387: else $tname=$trow['testname'];
388: if ($freshdata) $alerts[$alertc++]=$tname." ".oText($trow['alertlevel']); // only record text to log if fresh
389: if ($trow['alertlevel']>$alertlevel) $alertlevel=$trow['alertlevel'];
390: }
391: */
392:
393: // and finally again use pullalert - this is either the new value if a pull was attempted or just remains the same as the old one
394: // if pull not scheduled yet
395: if ($pullalert>0) $alertlevel=2; // so mark a failure
396:
397: }
398:
399: if ( ($dotests && ($row['nsenabled']==1) && ($row['nspullenabled']==1)) || // pull and tests are on
400: (($row['nsenabled']==1)&&($row['nspushenabled']==1)) ) // or pushed
401: {
402: if ($row['nsfreshpush']==1)
403: {
404: $freshdata=true;
405: $uq="UPDATE fnnode SET nsfreshpush=0 WHERE nodeid=\"".$row['nodeid']."\"";
406: $NATS->DB->Query($uq);
407: }
408: // Process for alerts in here - whether pulled or not!
409: $tq="SELECT testtype,testname,alertlevel,lastvalue FROM fnnstest WHERE nodeid=\"".$row['nodeid']."\" AND testenabled=1 AND testalerts=1 AND alertlevel>0";
410: $tr=$NATS->DB->Query($tq);
411: while ($trow=$NATS->DB->Fetch_Array($tr))
412: {
413: if ($trow['testname']=="") $tname=$trow['testtype'];
414: else $tname=$trow['testname'];
415: if ($freshdata) // only record text to log if fresh
416: {
417: $s=$tname." ".oText($trow['alertlevel']);
418: $result=$trow['lastvalue'];
419: if (is_numeric($result))
420: {
421: if ($NATS->Cfg->Get("site.alert.showvalue",0)==1) $s.=" (".$result.")";
422: }
423: else // non-numeric
424: {
425: if ($NATS->Cfg->Get("site.alert.showtext",0)==1) $s.=" (".$result.")";
426: }
427: $alerts[$alertc++]=$s;
428: }
429: if ($trow['alertlevel']>$alertlevel) $alertlevel=$trow['alertlevel'];
430: }
431: }
432:
433: $NATS->Event("Tester ".$trid." Finished Node ".$row['nodeid'],10,"Tester","Node");
434:
435: $eventdata=array( "nodeid" => $row['nodeid'], "alertlevel" => $alertlevel );
436: $NATS->EventHandler("node_test_finish",$eventdata);
437:
438: db("Highest Alert Level: ".$alertlevel);
439: db("Alert Count : ".$alertc);
440: $als="";
441: foreach($alerts as $al) $als.=$al.", ";
442: db("Alerts: ".$als);
443:
444: $NATS->SetAlerts($row['nodeid'],$alertlevel,$alerts);
445:
446: // This is where child/slave nodes spawn
447:
448: // $pingpassed bool holds if pingtest has passed
449: // $alertlevel holds the overall status
450: $chq="SELECT nodeid,masterjustping FROM fnnode WHERE masterid=\"".$row['nodeid']."\"";
451: //echo $chq;
452:
453: $chr=$NATS->DB->Query($chq);
454: $spawnlist=array();
455:
456: while ($child=$NATS->DB->Fetch_Array($chr))
457: {
458: if (($child['masterjustping']==1)&&($pingpassed)) $spawnlist[]=$child['nodeid'];
459: else if ($alertlevel==0) $spawnlist[]=$child['nodeid'];
460: // logic: if the child requires a ping and ping has passed ok then spawn it
461: // otherwise (pass on any alert) if everything has passed spawn it
462: }
463: $NATS->DB->Free($chr);
464:
465: if (count($spawnlist)>0)
466: {
467: $cmd="php ./test-threaded.php";
468: foreach($spawnlist as $child)
469: $cmd.=" ".$child;
470: $cmd.=" > /tmp/ns.master.".$row['nodeid']." &";
471: db("Children Spawning: ".$cmd);
472: exec($cmd);
473: }
474:
475:
476: // End of the node... carry forward highest level
477:
478: db(" ");
479:
480: if ($alertlevel>$highalertlevel) $highalertlevel=$alertlevel;
481: $talertc+=$alertc;
482:
483: }
484:
485:
486:
487: db("Finished Tests... Finishing Off");
488: db("Summary: Tester ".$trid." Highest Level ".$highalertlevel.", Alerts ".$talertc);
489: if ($highalertlevel>-1)
490: {
491: $uq="UPDATE fntestrun SET finishx=".time().",routput=\"".ss($dbt)."\" WHERE trid=".$trid;
492: $NATS->DB->Query($uq);
493: }
494: else
495: {
496: $uq="DELETE FROM fntestrun WHERE trid=".$trid;
497: $NATS->DB->Query($uq);
498: }
499:
500:
501: $NATS->Event("Tester ".$trid." Highest Level ".$highalertlevel.", Alerts ".$talertc,7,"Tester","Stat");
502: $NATS->Event("Tester ".$trid." Finished",5,"Tester","Stop");
503:
504: // in here for now...
505: $NATS->ActionFlush();
506:
507: $NATS->Stop();
508: db("NATS Stopped... Finished");
509: ?>
510:
511: