BatchComputeService.h
1 
11 #ifndef WRENCH_BATCH_SERVICE_H
12 #define WRENCH_BATCH_SERVICE_H
13 
14 #include "wrench/services/compute/ComputeService.h"
15 #include "wrench/services/compute/standard_job_executor/StandardJobExecutor.h"
16 #include "wrench/services/compute/batch/BatchJob.h"
17 #include "wrench/services/compute/batch/BatschedNetworkListener.h"
18 #include "wrench/services/compute/batch/BatchComputeServiceProperty.h"
19 #include "wrench/services/compute/batch/BatchComputeServiceMessagePayload.h"
20 #include "wrench/services/helpers/Alarm.h"
21 #include "wrench/workflow/job/StandardJob.h"
22 #include "wrench/workflow/job/WorkflowJob.h"
23 #include "wrench/services/compute/batch/batch_schedulers/BatchScheduler.h"
24 
25 #include <deque>
26 #include <queue>
27 #include <set>
28 #include <tuple>
29 
30 namespace wrench {
31 
32  class WorkloadTraceFileReplayer; // forward
33 
50 
54  private:
55 
56  std::map<std::string, std::string> default_property_values = {
65 #ifdef ENABLE_BATSCHED
67 // {BatchComputeServiceProperty::BATCH_SCHEDULING_ALGORITHM, "easy_bf"},
68 // {BatchComputeServiceProperty::BATCH_SCHEDULING_ALGORITHM, "easy_bf_fast"},
69 
71 #else
73 #endif
83  };
84 
85  std::map<std::string, double> default_messagepayload_values = {
102  };
103 
104  public:
105  BatchComputeService(const std::string &hostname,
106  std::vector<std::string> compute_hosts,
107  std::string scratch_space_mount_point,
108  std::map<std::string, std::string> property_list = {},
109  std::map<std::string, double> messagepayload_list = {}
110  );
111 
112  /***********************/
114  /***********************/
115  std::map<std::string,double> getStartTimeEstimates(std::set<std::tuple<std::string,unsigned long,unsigned long, double>> resources);
116 
117  std::vector<std::tuple<std::string, std::string, int, int, int, double, double>> getQueue();
118 
119  /***********************/
121  /***********************/
122 
123  /***********************/
125  /***********************/
126  ~BatchComputeService() override;
127  /***********************/
129  /***********************/
130 
131  private:
132 
133  friend class WorkloadTraceFileReplayer;
134  friend class FCFSBatchScheduler;
135  friend class CONSERVATIVEBFBatchScheduler;
137  friend class BatschedBatchScheduler;
138 
139  BatchComputeService(const std::string hostname,
140  std::vector<std::string> compute_hosts,
141  unsigned long cores_per_host,
142  double ram_per_host,
143  std::string scratch_space_mount_point,
144  std::map<std::string, std::string> property_list,
145  std::map<std::string, double> messagepayload_list,
146  std::string suffix
147  );
148 
149  // helper function
150  static unsigned long parseUnsignedLongServiceSpecificArgument(std::string key, const std::map<std::string, std::string> &args);
151 
152  // helper function
153  void submitWorkflowJob(std::shared_ptr<WorkflowJob> job, const std::map<std::string, std::string> &batch_job_args);
154 
155  //submits a standard job
156  void submitStandardJob(std::shared_ptr<StandardJob> job, const std::map<std::string, std::string> &batch_job_args) override;
157 
158  //submits a standard job
159  void submitPilotJob(std::shared_ptr<PilotJob> job, const std::map<std::string, std::string> &batch_job_args) override;
160 
161  // helper function
162  void terminateWorkflowJob(std::shared_ptr<WorkflowJob> job);
163 
164  // terminate a standard job
165  void terminateStandardJob(std::shared_ptr<StandardJob> job) override;
166 
167  // terminate a pilot job
168  void terminatePilotJob(std::shared_ptr<PilotJob> job) override;
169 
170  std::vector<std::tuple<std::string, double, double, double, double, unsigned int, std::string>> workload_trace;
171  std::shared_ptr<WorkloadTraceFileReplayer> workload_trace_replayer;
172 
173  bool clean_exit = false;
174 
175  //Used to add appropriate overhead to grid universe jobs.
176  bool grid_execution = false;
177  bool grid_execution_start = false;
178  bool grid_execution_end = false;
179 
180  //Configuration to create randomness in measurement period initially
181  unsigned long random_interval = 10;
182 
183  //create alarms for standard jobs
184  std::map<std::string,std::shared_ptr<Alarm>> standard_job_alarms;
185 
186  //alarms for pilot jobs (only one pilot job alarm)
187  std::map<std::string,std::shared_ptr<Alarm>> pilot_job_alarms;
188 
189  /* Resources information in batch */
190  unsigned long total_num_of_nodes;
191  unsigned long num_cores_per_node;
192  std::map<std::string, unsigned long> nodes_to_cores_map;
193  std::vector<double> timeslots;
194  std::map<std::string, unsigned long> available_nodes_to_cores;
195  std::map<unsigned long, std::string> host_id_to_names;
196  std::vector<std::string> compute_hosts;
197  /* End Resources information in batch */
198 
199  // Vector of standard job executors
200  std::set<std::shared_ptr<StandardJobExecutor>> running_standard_job_executors;
201 
202  // Vector of standard job executors (which is cleared periodically)
203  std::set<std::shared_ptr<StandardJobExecutor>> finished_standard_job_executors;
204 
205  // Master List of batch jobs
206  std::set<std::shared_ptr<BatchJob>> all_jobs;
207 
208  //A set of running batch jobs
209  std::set<std::shared_ptr<BatchJob>> running_jobs;
210 
211  // The batch queue
212  std::deque<std::shared_ptr<BatchJob>> batch_queue;
213 
214  // A set of "waiting" batch jobs, i.e., jobs that are waiting to be sent to
215  // the scheduler (useful for batsched only)
216  std::set<std::shared_ptr<BatchJob>> waiting_jobs;
217 
218  // Scheduler
219  std::unique_ptr<BatchScheduler> scheduler;
220 
221 
222 #ifdef ENABLE_BATSCHED
223 
224  std::set<std::string> scheduling_algorithms = {"conservative_bf", "crasher", "easy_bf", "easy_bf_fast",
225  "easy_bf_plot_liquid_load_horizon",
226  "energy_bf", "energy_bf_dicho", "energy_bf_idle_sleeper",
227  "energy_bf_monitoring",
228  "energy_bf_monitoring_inertial", "energy_bf_subpart_sleeper",
229  "energy_watcher", "fcfs_fast", "fast_conservative_bf",
230  "filler", "killer", "killer2", "random", "rejecter",
231  "sequencer", "sleeper", "submitter", "waiting_time_estimator"
232  };
233 
234  std::set<std::string> queue_ordering_options = {"fcfs", "lcfs", "desc_bounded_slowdown", "desc_slowdown",
235  "asc_size", "desc_size", "asc_walltime", "desc_walltime"
236 
237  };
238 #else
239  std::set<std::string> scheduling_algorithms = {"fcfs", "conservative_bf", "conservative_bf_core_level"
240  };
241 
242  //Batch queue ordering options
243  std::set<std::string> queue_ordering_options = {
244  };
245 
246 #endif
247 
248  unsigned long generateUniqueJobID();
249 
250  void removeJobFromRunningList(std::shared_ptr<BatchJob> job);
251 
252  void removeJobFromBatchQueue(std::shared_ptr<BatchJob> job);
253 
254  void removeBatchJobFromJobsList(std::shared_ptr<BatchJob> job);
255 
256  int main() override;
257 
258  bool processNextMessage();
259 
260  void startBackgroundWorkloadProcess();
261 
262  void processGetResourceInformation(const std::string &answer_mailbox);
263 
264  void processStandardJobCompletion(std::shared_ptr<StandardJobExecutor> executor, std::shared_ptr<StandardJob> job);
265 
266  void processStandardJobFailure(std::shared_ptr<StandardJobExecutor> executor,
267  std::shared_ptr<StandardJob> job,
268  std::shared_ptr<FailureCause> cause);
269 
270  void terminateRunningStandardJob(std::shared_ptr<StandardJob> job);
271 
272 
273  //Terminate the batch service (this is usually for pilot jobs when they act as a batch service)
274  void cleanup(bool has_returned_from_main, int return_value) override;
275 
276  // Terminate currently running pilot jobs
277  void terminateRunningPilotJobs();
278 
279  //Fail the standard jobs
280  void failCurrentStandardJobs();
281 
282  //Process the pilot job completion
283  void processPilotJobCompletion(std::shared_ptr<PilotJob> job);
284 
285  //Process standard job timeout
286  void processStandardJobTimeout(std::shared_ptr<StandardJob> job);
287 
288  //process standard job termination request
289  void processStandardJobTerminationRequest(std::shared_ptr<StandardJob> job, std::string answer_mailbox);
290 
291  //process pilot job termination request
292  void processPilotJobTerminationRequest(std::shared_ptr<PilotJob> job, std::string answer_mailbox);
293 
294  // process a batch job tiemout event
295  void processAlarmJobTimeout(std::shared_ptr<BatchJob>job);
296 
297  //Process pilot job timeout
298  void processPilotJobTimeout(std::shared_ptr<PilotJob> job);
299 
300  //free up resources
301  void freeUpResources(std::map<std::string, std::tuple<unsigned long, double>> resources);
302 
303  //send call back to the pilot job submitters
304  void sendPilotJobExpirationNotification(std::shared_ptr<PilotJob> job);
305 
306  //send call back to the standard job submitters
307  void sendStandardJobFailureNotification(std::shared_ptr<StandardJob> job, std::string job_id, std::shared_ptr<FailureCause> cause);
308 
309  // process a job submission
310  void processJobSubmission(std::shared_ptr<BatchJob>job, std::string answer_mailbox);
311 
312  //start a job
313  void startJob(std::map<std::string, std::tuple<unsigned long, double>>, std::shared_ptr<WorkflowJob> ,
314  std::shared_ptr<BatchJob>, unsigned long, unsigned long, unsigned long);
315 
316 
317  void processExecuteJobFromBatSched(std::string bat_sched_reply);
318 
319  };
320 }
321 
322 
323 #endif //WRENCH_BATCH_SERVICE_H
static const std::string SIMULATE_COMPUTATION_AS_SLEEP
Simulate computation as just a sleep instead of an actual compute thread. This is for scalability rea...
Definition: BatchComputeServiceProperty.h:142
static const std::string PILOT_JOB_EXPIRED_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to state that a pilot job has expired.
Definition: ComputeServiceMessagePayload.h:44
static const std::string TERMINATE_PILOT_JOB_ANSWER_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to acknowledge a pilot job termination.
Definition: ComputeServiceMessagePayload.h:54
A class that implements a FCFS batch scheduler.
Definition: FCFSBatchScheduler.h:25
static const std::string SUPPORTS_GRID_UNIVERSE
Indicates batch service being used for grid universe through HTCondor.
Definition: BatchComputeServiceProperty.h:168
static const std::string SUBMIT_STANDARD_JOB_REQUEST_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to submit a standard job.
Definition: ComputeServiceMessagePayload.h:26
static const std::string DAEMON_STOPPED_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to confirm it has terminated.
Definition: ServiceMessagePayload.h:33
static const std::string TERMINATE_STANDARD_JOB_ANSWER_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to acknowledge a standard job terminati...
Definition: ComputeServiceMessagePayload.h:36
static const std::string IGNORE_INVALID_JOBS_IN_WORKLOAD_TRACE_FILE
Whether, when simulating a workload trace file, to abort when there is an invalid job specification (...
Definition: BatchComputeServiceProperty.h:104
static const std::string USE_REAL_RUNTIMES_AS_REQUESTED_RUNTIMES_IN_WORKLOAD_TRACE_FILE
Whether, when simulating a workload trace file, to use the actual runtimes as requested runtimes (i....
Definition: BatchComputeServiceProperty.h:95
static const std::string RESOURCE_DESCRIPTION_REQUEST_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to ask it for information on its resour...
Definition: ComputeServiceMessagePayload.h:56
static const std::string BATSCHED_CONTIGUOUS_ALLOCATION
Controls Batsched node allocation policy.
Definition: BatchComputeServiceProperty.h:160
static const std::string HOST_SELECTION_ALGORITHM
The host selection algorithm. Can be:
Definition: BatchComputeServiceProperty.h:62
The compute service base class.
Definition: ComputeService.h:33
static const std::string SUBMIT_TIME_OF_FIRST_JOB_IN_WORKLOAD_TRACE_FILE
A specification of the submit time of the first job in a provided trace file.
Definition: BatchComputeServiceProperty.h:111
A class that defines a conservative backfilling batch scheduler.
Definition: CONSERVATIVEBFBatchSchedulerCoreLevel.h:26
static const std::string GRID_PRE_EXECUTION_DELAY
Overhead delay in seconds between condor and slurm for the start of execution.
Definition: BatchComputeServiceProperty.h:174
static const std::string SIMULATED_WORKLOAD_TRACE_FILE
Path to a workload trace file to be replayed. The trace file can be be in the SWF format (see http://...
Definition: BatchComputeServiceProperty.h:85
static const std::string TERMINATE_PILOT_JOB_REQUEST_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to terminate a pilot job.
Definition: ComputeServiceMessagePayload.h:52
A batch-scheduled compute service that manages a set of compute hosts and controls access to their re...
Definition: BatchComputeService.h:49
static const std::string STANDARD_JOB_DONE_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to state that it has completed a standa...
Definition: ComputeServiceMessagePayload.h:30
static const std::string STANDARD_JOB_FAILED_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to state that a running standard job ha...
Definition: ComputeServiceMessagePayload.h:32
Definition: Alarm.cpp:20
static const std::string TERMINATE_STANDARD_JOB_REQUEST_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to terminate a standard job.
Definition: ComputeServiceMessagePayload.h:34
static const std::string SUBMIT_PILOT_JOB_ANSWER_MESSAGE_PAYLOAD
The number of bytes in the control message sent from the daemon to acknowledge a pilot job submission...
Definition: ComputeServiceMessagePayload.h:40
std::vector< std::tuple< std::string, std::string, int, int, int, double, double > > getQueue()
Gets the state of the batch queue.
Definition: BatchComputeService.cpp:247
std::map< std::string, double > getStartTimeEstimates(std::set< std::tuple< std::string, unsigned long, unsigned long, double >> resources)
Retrieve start time estimates for a set of job configurations.
Definition: BatchComputeService.cpp:222
A class that defines a batsched batch scheduler.
Definition: BatschedBatchScheduler.h:24
static const std::string TASK_STARTUP_OVERHEAD
The overhead to start a task execution, in seconds.
Definition: BatchComputeServiceProperty.h:27
static const std::string BATSCHED_LOGGING_MUTED
Controls Batsched logging.
Definition: BatchComputeServiceProperty.h:150
static const std::string TASK_SELECTION_ALGORITHM
The algorithm to pick which ready computational task (within a standard job executed by the batch ser...
Definition: BatchComputeServiceProperty.h:72
A service that goes through a job submission trace (as loaded by a TraceFileLoader),...
Definition: WorkloadTraceFileReplayer.h:28
std::map< std::string, double > messagepayload_list
The service's messagepayload list.
Definition: Service.h:112
static const std::string RESOURCE_DESCRIPTION_ANSWER_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to state information on its resources.
Definition: ComputeServiceMessagePayload.h:58
static const std::string OUTPUT_CSV_JOB_LOG
Path to a to-be-generated Batsim-style CSV trace file (e.g. for b3atch schedule visualization purpose...
Definition: BatchComputeServiceProperty.h:122
static const std::string STOP_DAEMON_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to terminate it.
Definition: ServiceMessagePayload.h:31
static const std::string PILOT_JOB_STARTED_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to state that a pilot job has started.
Definition: ComputeServiceMessagePayload.h:42
std::string hostname
The name of the host on which the daemon is running.
Definition: S4U_Daemon.h:51
static const std::string SUBMIT_STANDARD_JOB_ANSWER_MESSAGE_PAYLOAD
The number of bytes in the control message sent by the daemon to acknowledge a standard job submissio...
Definition: ComputeServiceMessagePayload.h:28
static const std::string BATCH_SCHEDULING_ALGORITHM
The batch scheduling algorithm. Can be:
Definition: BatchComputeServiceProperty.h:43
static const std::string SUBMIT_PILOT_JOB_REQUEST_MESSAGE_PAYLOAD
The number of bytes in the control message sent to the daemon to submit a pilot job.
Definition: ComputeServiceMessagePayload.h:38
static const std::string GRID_POST_EXECUTION_DELAY
Overhead delay in seconds between condor and slurm for the completion of execution.
Definition: BatchComputeServiceProperty.h:180
static const std::string SUPPORTS_PILOT_JOBS
Whether the compute service supports pilot jobs (true or false)
Definition: ComputeServiceProperty.h:26
static const std::string BATCH_RJMS_PADDING_DELAY
Integral number of seconds that the Batch Scheduler adds to the runtime of each incoming job....
Definition: BatchComputeServiceProperty.h:132
A class that defines a conservative backfilling batch scheduler.
Definition: CONSERVATIVEBFBatchScheduler.h:26
BatchComputeService(const std::string &hostname, std::vector< std::string > compute_hosts, std::string scratch_space_mount_point, std::map< std::string, std::string > property_list={}, std::map< std::string, double > messagepayload_list={})
Constructor.
Definition: BatchComputeService.cpp:61
static const std::string BATCH_QUEUE_ORDERING_ALGORITHM
The batch queue ordering algorithm. Can be:
Definition: BatchComputeServiceProperty.h:52
std::map< std::string, std::string > property_list
The service's property list.
Definition: Service.h:109
static const std::string SUPPORTS_STANDARD_JOBS
Whether the compute service supports standard jobs (true or false)
Definition: ComputeServiceProperty.h:24