| 1 | /*
 | 
|---|
| 2 |  * Project: MoleCuilder
 | 
|---|
| 3 |  * Description: creates and alters molecular systems
 | 
|---|
| 4 |  * Copyright (C)  2011 University of Bonn. All rights reserved.
 | 
|---|
| 5 |  * Please see the LICENSE file or "Copyright notice" in builder.cpp for details.
 | 
|---|
| 6 |  */
 | 
|---|
| 7 | 
 | 
|---|
| 8 | /*
 | 
|---|
| 9 |  * FragmentQueue.cpp
 | 
|---|
| 10 |  *
 | 
|---|
| 11 |  *  Created on: Oct 19, 2011
 | 
|---|
| 12 |  *      Author: heber
 | 
|---|
| 13 |  */
 | 
|---|
| 14 | 
 | 
|---|
| 15 | // include config.h
 | 
|---|
| 16 | #ifdef HAVE_CONFIG_H
 | 
|---|
| 17 | #include <config.h>
 | 
|---|
| 18 | #endif
 | 
|---|
| 19 | 
 | 
|---|
| 20 | #include "CodePatterns/MemDebug.hpp"
 | 
|---|
| 21 | 
 | 
|---|
| 22 | #include "FragmentQueue.hpp"
 | 
|---|
| 23 | 
 | 
|---|
| 24 | #include "CodePatterns/Assert.hpp"
 | 
|---|
| 25 | #include "CodePatterns/Observer/Channels.hpp"
 | 
|---|
| 26 | #include "CodePatterns/Log.hpp"
 | 
|---|
| 27 | 
 | 
|---|
| 28 | FragmentResult::ptr FragmentQueue::NoResult( new FragmentResult(-1) );
 | 
|---|
| 29 | FragmentResult::ptr FragmentQueue::NoResultQueued( new FragmentResult(-2) );
 | 
|---|
| 30 | FragmentResult::ptr FragmentQueue::ResultDelivered( new FragmentResult(-3) );
 | 
|---|
| 31 | size_t FragmentQueue::Max_Attempts = (size_t)3;
 | 
|---|
| 32 | 
 | 
|---|
| 33 | /** Constructor for class FragmentQueue.
 | 
|---|
| 34 |  *
 | 
|---|
| 35 |  */
 | 
|---|
| 36 | FragmentQueue::FragmentQueue() :
 | 
|---|
| 37 |     Observable("FragmentQueue")
 | 
|---|
| 38 | {
 | 
|---|
| 39 |   // observable stuff
 | 
|---|
| 40 |   Channels *OurChannel = new Channels;
 | 
|---|
| 41 |   NotificationChannels.insert( std::make_pair(this, OurChannel) );
 | 
|---|
| 42 |   // add instance for each notification type
 | 
|---|
| 43 |   for (size_t type = 0; type < NotificationType_MAX; ++type)
 | 
|---|
| 44 |     OurChannel->addChannel(type);
 | 
|---|
| 45 | }
 | 
|---|
| 46 | 
 | 
|---|
| 47 | /** Destructor for class FragmentQueue.
 | 
|---|
| 48 |  *
 | 
|---|
| 49 |  */
 | 
|---|
| 50 | FragmentQueue::~FragmentQueue()
 | 
|---|
| 51 | {
 | 
|---|
| 52 |   jobs.clear();
 | 
|---|
| 53 |   results.clear();
 | 
|---|
| 54 | }
 | 
|---|
| 55 | 
 | 
|---|
| 56 | /** Checks whether there are jobs in the queue at all.
 | 
|---|
| 57 |  * \return true - jobs present, false - queue is empty
 | 
|---|
| 58 |  */
 | 
|---|
| 59 | bool FragmentQueue::isJobPresent() const
 | 
|---|
| 60 | {
 | 
|---|
| 61 |   return !jobs.empty();
 | 
|---|
| 62 | }
 | 
|---|
| 63 | 
 | 
|---|
| 64 | /** Pushes a FragmentJob into the internal queue for delivery to server.
 | 
|---|
| 65 |  *
 | 
|---|
| 66 |  * \note we throw assertion when jobid has already been used.
 | 
|---|
| 67 |  *
 | 
|---|
| 68 |  * \param job job to enter into queue
 | 
|---|
| 69 |  */
 | 
|---|
| 70 | void FragmentQueue::pushJob(FragmentJob::ptr job)
 | 
|---|
| 71 | {
 | 
|---|
| 72 |   if ((job->getId() != JobId::IllegalJob) && (!results.count(job->getId()))) {
 | 
|---|
| 73 |     OBSERVE;
 | 
|---|
| 74 |     NOTIFY(JobAdded);
 | 
|---|
| 75 |     results.insert( std::make_pair(job->getId(), NoResult));
 | 
|---|
| 76 |     jobs.push_back(job);
 | 
|---|
| 77 |   } else {
 | 
|---|
| 78 |     ELOG(1, "job to push has IllegalJob id or id "+toString(job->getId())+" has already been used.");
 | 
|---|
| 79 |     ASSERT(false,
 | 
|---|
| 80 |         "job to push has IllegalJob id or id "+toString(job->getId())+" has already been used.");
 | 
|---|
| 81 |   }
 | 
|---|
| 82 | }
 | 
|---|
| 83 | 
 | 
|---|
| 84 | /** Pushes a bunch of FragmentJob's into the internal queue for delivery to server.
 | 
|---|
| 85 |  *
 | 
|---|
| 86 |  * \note we throw assertion when jobid has already been used.
 | 
|---|
| 87 |  *
 | 
|---|
| 88 |  * \sa pushJob()
 | 
|---|
| 89 |  *
 | 
|---|
| 90 |  * \param _jobs jobs to enter into queue
 | 
|---|
| 91 |  */
 | 
|---|
| 92 | void FragmentQueue::pushJobs(std::vector<FragmentJob::ptr> &_jobs)
 | 
|---|
| 93 | {
 | 
|---|
| 94 |   for (std::vector<FragmentJob::ptr>::iterator iter = _jobs.begin();
 | 
|---|
| 95 |       iter != _jobs.end(); ++iter)
 | 
|---|
| 96 |     pushJob(*iter);
 | 
|---|
| 97 | }
 | 
|---|
| 98 | 
 | 
|---|
| 99 | /** Pops top-most FragmentJob from internal queue.
 | 
|---|
| 100 |  *
 | 
|---|
| 101 |  * From here on, we expect a result in FragmentQueue::results.
 | 
|---|
| 102 |  *
 | 
|---|
| 103 |  * \return job topmost job from queue
 | 
|---|
| 104 |  */
 | 
|---|
| 105 | FragmentJob::ptr FragmentQueue::popJob()
 | 
|---|
| 106 | {
 | 
|---|
| 107 |   ASSERT(jobs.size(),
 | 
|---|
| 108 |       "FragmentQueue::popJob() - there are no jobs on the queue.");
 | 
|---|
| 109 |   FragmentJob::ptr job = jobs.front();
 | 
|---|
| 110 | #ifndef NDEBUG
 | 
|---|
| 111 |   std::pair< BackupMap::iterator, bool> inserter =
 | 
|---|
| 112 | #endif
 | 
|---|
| 113 |   backup.insert( std::make_pair( job->getId(), job ));
 | 
|---|
| 114 |   ASSERT (inserter.second,
 | 
|---|
| 115 |       "FragmentQueue::popJob() - job "+toString(job->getId())+
 | 
|---|
| 116 |       " is already in the backup.");
 | 
|---|
| 117 |   ResultMap::iterator iter = results.find(job->getId());
 | 
|---|
| 118 |   ASSERT(iter != results.end(),
 | 
|---|
| 119 |       "FragmentQueue::popJob() - for job "+toString(job->getId())+" no result place has been stored.");
 | 
|---|
| 120 |   iter->second = NoResultQueued;
 | 
|---|
| 121 |   jobs.pop_front();
 | 
|---|
| 122 |   return job;
 | 
|---|
| 123 | }
 | 
|---|
| 124 | 
 | 
|---|
| 125 | /** Internal function to check whether result is not one of static entities.
 | 
|---|
| 126 |  *
 | 
|---|
| 127 |  * @param result result to check against
 | 
|---|
| 128 |  * @return true - result is a present, valid result, false - result is one of the statics
 | 
|---|
| 129 |  */
 | 
|---|
| 130 | bool FragmentQueue::isPresentResult(const FragmentResult::ptr result) const
 | 
|---|
| 131 | {
 | 
|---|
| 132 |   return (*result != *NoResult)
 | 
|---|
| 133 |       && (*result != *NoResultQueued)
 | 
|---|
| 134 |       && (*result != *ResultDelivered);
 | 
|---|
| 135 | }
 | 
|---|
| 136 | 
 | 
|---|
| 137 | /** Queries whether a job has already been finished and the result is present.
 | 
|---|
| 138 |  *
 | 
|---|
| 139 |  * \param jobid id of job to query
 | 
|---|
| 140 |  * \return true - result is present, false - result is not present
 | 
|---|
| 141 |  */
 | 
|---|
| 142 | bool FragmentQueue::isResultPresent(JobId_t jobid) const
 | 
|---|
| 143 | {
 | 
|---|
| 144 |   ResultMap::const_iterator iter = results.find(jobid);
 | 
|---|
| 145 |   return ((iter != results.end())
 | 
|---|
| 146 |       && isPresentResult(iter->second));
 | 
|---|
| 147 | }
 | 
|---|
| 148 | 
 | 
|---|
| 149 | /** Counts the number of jobs for which we have a calculated result present.
 | 
|---|
| 150 |  *
 | 
|---|
| 151 |  * \return number of calculated results
 | 
|---|
| 152 |  */
 | 
|---|
| 153 | size_t FragmentQueue::getDoneJobs() const
 | 
|---|
| 154 | {
 | 
|---|
| 155 |   size_t doneJobs = 0;
 | 
|---|
| 156 |   for (ResultMap::const_iterator iter = results.begin();
 | 
|---|
| 157 |       iter != results.end(); ++iter)
 | 
|---|
| 158 |     if (isPresentResult(iter->second))
 | 
|---|
| 159 |       ++doneJobs;
 | 
|---|
| 160 |   return doneJobs;
 | 
|---|
| 161 | }
 | 
|---|
| 162 | 
 | 
|---|
| 163 | /** Counts the number of jobs for which still have to be calculated.
 | 
|---|
| 164 |  *
 | 
|---|
| 165 |  * \return number of jobs to be calculated
 | 
|---|
| 166 |  */
 | 
|---|
| 167 | size_t FragmentQueue::getPresentJobs() const
 | 
|---|
| 168 | {
 | 
|---|
| 169 |   const size_t presentJobs = jobs.size();
 | 
|---|
| 170 |   return presentJobs;
 | 
|---|
| 171 | }
 | 
|---|
| 172 | 
 | 
|---|
| 173 | /** Delivers result for a finished job.
 | 
|---|
| 174 |  *
 | 
|---|
| 175 |  * \note we throw assertion if not present
 | 
|---|
| 176 |  *
 | 
|---|
| 177 |  * \param jobid id of job
 | 
|---|
| 178 |  * \return result for job of given \a jobid
 | 
|---|
| 179 |  */
 | 
|---|
| 180 | FragmentResult::ptr FragmentQueue::getResult(JobId_t jobid)
 | 
|---|
| 181 | {
 | 
|---|
| 182 |   ResultMap::iterator iter = results.find(jobid);
 | 
|---|
| 183 |   ASSERT(iter != results.end(),
 | 
|---|
| 184 |       "FragmentQueue::pushResult() - job "+toString(jobid)+" is not known to us.");
 | 
|---|
| 185 |   ASSERT(*iter->second != *NoResult,
 | 
|---|
| 186 |       "FragmentQueue::pushResult() - job "+toString(jobid)+" has not been request for calculation yet.");
 | 
|---|
| 187 |   ASSERT(*iter->second != *NoResultQueued,
 | 
|---|
| 188 |       "FragmentQueue::pushResult() - job "+toString(jobid)+"'s calculation is underway but not result has arrived yet.");
 | 
|---|
| 189 |   ASSERT(*iter->second != *ResultDelivered,
 | 
|---|
| 190 |       "FragmentQueue::pushResult() - job "+toString(jobid)+"'s result has already been delivered.");
 | 
|---|
| 191 |   /// store result
 | 
|---|
| 192 |   FragmentResult::ptr _result = iter->second;
 | 
|---|
| 193 |   /// mark as delivered in map
 | 
|---|
| 194 |   iter->second = ResultDelivered;
 | 
|---|
| 195 |   /// and return result
 | 
|---|
| 196 |   return _result;
 | 
|---|
| 197 | }
 | 
|---|
| 198 | 
 | 
|---|
| 199 | std::vector<FragmentResult::ptr> FragmentQueue::getAllResults()
 | 
|---|
| 200 | {
 | 
|---|
| 201 |   std::vector<FragmentResult::ptr> returnresults;
 | 
|---|
| 202 |   for (ResultMap::iterator iter = results.begin();
 | 
|---|
| 203 |       iter != results.end(); ++iter) {
 | 
|---|
| 204 |     if (isPresentResult(iter->second)) {
 | 
|---|
| 205 |       returnresults.push_back(getResult(iter->first));
 | 
|---|
| 206 |       iter = results.begin();
 | 
|---|
| 207 |     }
 | 
|---|
| 208 |   }
 | 
|---|
| 209 | 
 | 
|---|
| 210 |   return returnresults;
 | 
|---|
| 211 | }
 | 
|---|
| 212 | 
 | 
|---|
| 213 | /** Pushes a result for a finished job.
 | 
|---|
| 214 |  *
 | 
|---|
| 215 |  * \note we throw assertion if job already has result or is not known.
 | 
|---|
| 216 |  *
 | 
|---|
| 217 |  * \param result result of job to store
 | 
|---|
| 218 |  */
 | 
|---|
| 219 | void FragmentQueue::pushResult(FragmentResult::ptr &_result)
 | 
|---|
| 220 | {
 | 
|---|
| 221 |   const JobId_t id = _result->getId();
 | 
|---|
| 222 |   /// check for presence
 | 
|---|
| 223 |   ResultMap::iterator iter = results.find(id);
 | 
|---|
| 224 |   ASSERT(iter != results.end(),
 | 
|---|
| 225 |       "FragmentQueue::pushResult() - job "+toString(id)+" is not known to us.");
 | 
|---|
| 226 |   ASSERT(*iter->second == *NoResultQueued,
 | 
|---|
| 227 |       "FragmentQueue::pushResult() - is not waiting for the result of job "+toString(id)+".");
 | 
|---|
| 228 |   // check whether this is a resubmitted job
 | 
|---|
| 229 |   AttemptsMap::iterator attemptiter = attempts.find(id);
 | 
|---|
| 230 |   // check whether succeeded or (finally) failed
 | 
|---|
| 231 |   if ((_result->exitflag == 0) 
 | 
|---|
| 232 |        || ((attemptiter != attempts.end()) && (attemptiter->second >= Max_Attempts))
 | 
|---|
| 233 |        || (Max_Attempts == 1)) {
 | 
|---|
| 234 |     // give notice if it is resubmitted job
 | 
|---|
| 235 |     if (attemptiter != attempts.end()) {
 | 
|---|
| 236 |       if (attemptiter->second >= Max_Attempts)
 | 
|---|
| 237 |         ELOG(1, "Job #" << id << " failed on " << Max_Attempts << "th attempt for the last time.");
 | 
|---|
| 238 |       else
 | 
|---|
| 239 |         LOG(1, "INFO: Job #" << id << " succeeded on " << attemptiter->second << "th attempt.");
 | 
|---|
| 240 |     }
 | 
|---|
| 241 |     // remove in attempts
 | 
|---|
| 242 |     if (attemptiter != attempts.end())
 | 
|---|
| 243 |       attempts.erase(attemptiter);
 | 
|---|
| 244 |     // remove in backup map
 | 
|---|
| 245 |     BackupMap::iterator backupiter = backup.find(id);
 | 
|---|
| 246 |     ASSERT( backupiter != backup.end(),
 | 
|---|
| 247 |         "FragmentQueue::pushResult() - cannot find job "+toString(id)
 | 
|---|
| 248 |         +" in backup.");
 | 
|---|
| 249 |     backup.erase(backupiter);
 | 
|---|
| 250 |     /// and overwrite NoResult in found entry
 | 
|---|
| 251 |     iter->second = _result;
 | 
|---|
| 252 |   } else {
 | 
|---|
| 253 |     LOG(1, "Job " << id << " failed, resubmitting.");
 | 
|---|
| 254 |     // increase attempts
 | 
|---|
| 255 |     if (attemptiter != attempts.end())
 | 
|---|
| 256 |       ++(attemptiter->second);
 | 
|---|
| 257 |     else
 | 
|---|
| 258 |       attempts.insert( std::make_pair(id, (size_t)1) );
 | 
|---|
| 259 |     // resubmit job
 | 
|---|
| 260 |     resubmitJob(id);
 | 
|---|
| 261 |   }
 | 
|---|
| 262 | }
 | 
|---|
| 263 | 
 | 
|---|
| 264 | /** Resubmit a job which a worker failed to calculate.
 | 
|---|
| 265 |  *
 | 
|---|
| 266 |  * @param jobid id of the failed job
 | 
|---|
| 267 |  */
 | 
|---|
| 268 | void FragmentQueue::resubmitJob(const JobId_t jobid)
 | 
|---|
| 269 | {
 | 
|---|
| 270 |   BackupMap::iterator iter = backup.find(jobid);
 | 
|---|
| 271 |   ASSERT( iter != backup.end(),
 | 
|---|
| 272 |       "FragmentQueue::resubmitJob() - job id "+toString(jobid)
 | 
|---|
| 273 |       +" not stored in backup.");
 | 
|---|
| 274 |   if (iter != backup.end()) {
 | 
|---|
| 275 |     // remove present result
 | 
|---|
| 276 |     ResultMap::iterator resiter = results.find(jobid);
 | 
|---|
| 277 |     ASSERT( resiter != results.end(),
 | 
|---|
| 278 |         "FragmentQueue::resubmitJob() - job "+toString(jobid)
 | 
|---|
| 279 |         +" to resubmit has no result present.");
 | 
|---|
| 280 |     results.erase(resiter);
 | 
|---|
| 281 |     pushJob(iter->second);
 | 
|---|
| 282 |     backup.erase(iter);
 | 
|---|
| 283 |   }
 | 
|---|
| 284 | }
 | 
|---|
| 285 | 
 | 
|---|