BareMetalComputeService.h

 
#ifndef WRENCH_BAREMETALCOMPUTESERVICE_H
#define WRENCH_BAREMETALCOMPUTESERVICE_H
 
 
#include <queue>
 
#include "wrench/services/compute/ComputeService.h"
#include "wrench/services/compute/standard_job_executor/StandardJobExecutor.h"
#include "BareMetalComputeServiceProperty.h"
#include "BareMetalComputeServiceMessagePayload.h"
#include "wrench/services/compute/workunit_executor/Workunit.h"
#include "wrench/services/helpers/HostStateChangeDetector.h"
 
 
 
namespace wrench {
 
    class Simulation;
 
    class StorageService;
 
    class FailureCause;
 
    class Alarm;
 
 
    class BareMetalComputeService : public ComputeService {
 
        friend class CloudComputeService;
        friend class BatchComputeService;
 
    private:
 
        std::map<std::string, std::string> default_property_values = {
                {BareMetalComputeServiceProperty::SUPPORTS_STANDARD_JOBS,                         "true"},
                {BareMetalComputeServiceProperty::SUPPORTS_PILOT_JOBS,                            "false"},
                {BareMetalComputeServiceProperty::TASK_STARTUP_OVERHEAD,                          "0.0"},
                {BareMetalComputeServiceProperty::TERMINATE_WHENEVER_ALL_RESOURCES_ARE_DOWN,      "false"},
        };
 
        std::map<std::string, double> default_messagepayload_values = {
                {BareMetalComputeServiceMessagePayload::STOP_DAEMON_MESSAGE_PAYLOAD,                    1024},
                {BareMetalComputeServiceMessagePayload::DAEMON_STOPPED_MESSAGE_PAYLOAD,                 1024},
                {BareMetalComputeServiceMessagePayload::SUBMIT_STANDARD_JOB_REQUEST_MESSAGE_PAYLOAD,    1024},
                {BareMetalComputeServiceMessagePayload::SUBMIT_STANDARD_JOB_ANSWER_MESSAGE_PAYLOAD,     1024},
                {BareMetalComputeServiceMessagePayload::JOB_TYPE_NOT_SUPPORTED_MESSAGE_PAYLOAD,         1024},
                {BareMetalComputeServiceMessagePayload::NOT_ENOUGH_CORES_MESSAGE_PAYLOAD,               1024},
                {BareMetalComputeServiceMessagePayload::STANDARD_JOB_DONE_MESSAGE_PAYLOAD,              1024},
                {BareMetalComputeServiceMessagePayload::STANDARD_JOB_FAILED_MESSAGE_PAYLOAD,            1024},
                {BareMetalComputeServiceMessagePayload::TERMINATE_STANDARD_JOB_REQUEST_MESSAGE_PAYLOAD, 1024},
                {BareMetalComputeServiceMessagePayload::TERMINATE_STANDARD_JOB_ANSWER_MESSAGE_PAYLOAD,  1024},
                {BareMetalComputeServiceMessagePayload::SUBMIT_PILOT_JOB_REQUEST_MESSAGE_PAYLOAD,       1024},
                {BareMetalComputeServiceMessagePayload::SUBMIT_PILOT_JOB_ANSWER_MESSAGE_PAYLOAD,        1024},
                {BareMetalComputeServiceMessagePayload::PILOT_JOB_STARTED_MESSAGE_PAYLOAD,              1024},
                {BareMetalComputeServiceMessagePayload::PILOT_JOB_EXPIRED_MESSAGE_PAYLOAD,              1024},
                {BareMetalComputeServiceMessagePayload::PILOT_JOB_FAILED_MESSAGE_PAYLOAD,               1024},
                {BareMetalComputeServiceMessagePayload::TERMINATE_PILOT_JOB_REQUEST_MESSAGE_PAYLOAD,    1024},
                {BareMetalComputeServiceMessagePayload::TERMINATE_PILOT_JOB_ANSWER_MESSAGE_PAYLOAD,     1024},
                {BareMetalComputeServiceMessagePayload::RESOURCE_DESCRIPTION_REQUEST_MESSAGE_PAYLOAD,     1024},
                {BareMetalComputeServiceMessagePayload::RESOURCE_DESCRIPTION_ANSWER_MESSAGE_PAYLOAD,      1024},
        };
 
    public:
 
        // Public Constructor
        BareMetalComputeService(const std::string &hostname,
                                const std::map<std::string, std::tuple<unsigned long, double>> compute_resources,
                                std::string scratch_space_mount_point,
                                std::map<std::string, std::string> property_list = {},
                                std::map<std::string, double> messagepayload_list = {}
        );
 
        // Public Constructor
        BareMetalComputeService(const std::string &hostname,
                                const 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 = {}
        );
 
 
        /***********************/
        /***********************/
 
        void submitStandardJob(StandardJob *job, std::map<std::string, std::string> &service_specific_args) override;
 
        void submitPilotJob(PilotJob *job, std::map<std::string, std::string> &service_specific_args) override;
 
        void terminateStandardJob(StandardJob *job) override;
 
        void terminatePilotJob(PilotJob *job) override;
 
        ~BareMetalComputeService();
 
        /***********************/
        /***********************/
 
    private:
 
        friend class Simulation;
 
        void validateProperties();
 
        // Low-level Constructor
        BareMetalComputeService(const std::string &hostname,
                                std::map<std::string, std::tuple<unsigned long, double>> compute_resources,
                                std::map<std::string, std::string> property_list,
                                std::map<std::string, double> messagepayload_list,
                                double ttl,
                                PilotJob *pj, std::string suffix,
                                std::shared_ptr<StorageService> scratch_space); // reference to upper level scratch space
 
        // Private Constructor
        BareMetalComputeService(const std::string &hostname,
                                std::map<std::string, std::tuple<unsigned long, double>> compute_resources,
                                std::map<std::string, std::string> property_list,
                                std::map<std::string, double> messagepayload_list,
                                std::shared_ptr<StorageService> scratch_space);
 
        // Low-level constructor helper method
        void initiateInstance(const std::string &hostname,
                              std::map<std::string, std::tuple<unsigned long, double>> compute_resources,
                              std::map<std::string, std::string> property_list,
                              std::map<std::string, double> messagepayload_list,
                              double ttl,
                              PilotJob *pj);
 
 
        std::map<std::string, std::tuple<unsigned long, double>> compute_resources;
 
        // Core availabilities (for each hosts, how many cores and how many bytes of RAM are currently available on it)
        std::map<std::string, double> ram_availabilities;
        std::map<std::string, unsigned long> running_thread_counts;
 
        unsigned long total_num_cores;
 
        double ttl;
        bool has_ttl;
        double death_date;
        std::shared_ptr<Alarm> death_alarm = nullptr;
 
        PilotJob *containing_pilot_job; // In case this service is in fact a pilot job
 
        std::map<StandardJob *, std::set<WorkflowFile*>> files_in_scratch;
 
        // Set of running jobs
        std::set<StandardJob *> running_jobs;
 
        // Job task execution specs
        std::map<StandardJob *, std::map<WorkflowTask *, std::tuple<std::string, unsigned long>>> job_run_specs;
 
        // Map of all Workunits
        std::map<StandardJob *, std::set<std::shared_ptr<Workunit>>> all_workunits;
 
        std::deque<std::shared_ptr<Workunit>> ready_workunits;
//        std::map<StandardJob *, std::set<Workunit *>> running_workunits;
        std::map<StandardJob *, std::set<std::shared_ptr<Workunit>>> completed_workunits;
 
        // Set of running WorkunitExecutors
        std::map<StandardJob *, std::set<std::shared_ptr<WorkunitExecutor>>> workunit_executors;
 
        // Add the scratch files of one standardjob to the list of all the scratch files of all the standard jobs inside the pilot job
        void storeFilesStoredInScratch(std::set<WorkflowFile*> scratch_files);
 
        // Cleanup the scratch if I am a pilot job
        void cleanUpScratch();
 
        int main() override;
 
        // Helper functions to make main() a bit more palatable
 
        void terminate(bool notify_pilot_job_submitters);
 
        void failCurrentStandardJobs();
 
        void processWorkunitExecutorCompletion(std::shared_ptr<WorkunitExecutor> workunit_executor, std::shared_ptr<Workunit> workunit);
 
        void processWorkunitExecutorFailure(std::shared_ptr<WorkunitExecutor> workunit_executor, std::shared_ptr<Workunit> workunit, std::shared_ptr<FailureCause> cause);
 
        void processWorkunitExecutorCrash(std::shared_ptr<WorkunitExecutor> workunit_executor);
 
        void forgetWorkunitExecutor(std::shared_ptr<WorkunitExecutor> workunit_executor);
 
        void processStandardJobTerminationRequest(StandardJob *job, const std::string &answer_mailbox);
 
        bool processNextMessage();
 
        void dispatchReadyWorkunits();
 
//        void someHostIsBackOn(simgrid::s4u::Host const &h);
//        bool host_back_on = false;
 
 
        enum JobTerminationCause {
                    TERMINATED,
 
                    COMPUTE_SERVICE_KILLED
        };
 
        void terminateRunningStandardJob(StandardJob *job, JobTerminationCause termination_cause);
 
        void failRunningStandardJob(StandardJob *job, std::shared_ptr<FailureCause> cause);
 
        void processGetResourceInformation(const std::string &answer_mailbox);
 
        void processSubmitPilotJob(const std::string &answer_mailbox, PilotJob *job, std::map<std::string, std::string> service_specific_args);
 
        void processSubmitStandardJob(const std::string &answer_mailbox, StandardJob *job,
                                      std::map<std::string, std::string> &service_specific_arguments);
 
        std::tuple<std::string, unsigned long> pickAllocation(WorkflowTask *task,
                                                              std::string required_host, unsigned long required_num_cores, double required_ram,
                                                              std::set<std::string> &hosts_to_avoid);
 
        bool jobCanRun(StandardJob *job, std::map<std::string, std::string> &service_specific_arguments);
 
        bool isThereAtLeastOneHostWithResources(unsigned long num_cores, double ram);
 
        void cleanup(bool has_terminated_cleanly, int return_value) override;
 
        bool areAllComputeResourcesDownWithNoWUERunning();
 
 
        int exit_code = 0;
 
        std::shared_ptr<HostStateChangeDetector> host_state_change_monitor;
 
    };
};
 
 
#endif //WRENCH_BAREMETALCOMPUTESERVICE_H