[FEAT]: resume waiting/running, dedup on tuner side (TPE-only)

nni/algorithms/hpo/tpe_tuner.py

@@ -215,7 +215,21 @@ def import_data(self, data):  # for resuming experiment
                 loss = -loss
             for key, value in param.items():
                 self._history[key].append(Record(value, loss))
-        _logger.info(f'Replayed {len(data)} trials')
+                self.dedup.add_history(param)
+        _logger.info(f'Replayed {len(data)} FINISHED trials')
+
+    def import_customized_data(self, data): # for dedup customized / resumed
+        if isinstance(data, str):
+            data = nni.load(data)
+
+        for trial in data:
+            # {'parameter_id': 0, 'parameter_source': 'resumed', 'parameters': {'batch_size': 128, ...}
+            if isinstance(trial, str):
+                trial = nni.load(trial)
+            param = format_parameters(trial['parameters'], self.space)
+            self._running_params[trial['parameter_id']] = param
+            self.dedup.add_history(param)
+        _logger.info(f'Replayed {len(data)} RUNING/WAITING trials')
 
 def suggest(args, rng, space, history):
     params = {}

nni/common/hpo_utils/dedup.py

@@ -79,6 +79,12 @@ def _not_dup(self, formatted_parameters: FormattedParameters) -> bool:
             self._history.add(params_str)
             return True
 
+    def add_history(self, formatted_parameters: FormattedParameters) -> None:
+        params = deformat_parameters(formatted_parameters, self._space)
+        params_str = typing.cast(str, nni.dump(params, sort_keys=True))
+        if params_str not in self._history:
+            self._history.add(params_str)
+
 def _spec_never_dup(spec: ParameterSpec) -> bool:
     if spec.is_nested():
         return False  # "not chosen" duplicates with "not chosen"

nni/runtime/msg_dispatcher.py

@@ -121,8 +121,14 @@ def handle_import_data(self, data):
 
     def handle_add_customized_trial(self, data):
         # data: parameters
-        id_ = _create_parameter_id()
-        _customized_parameter_ids.add(id_)
+        if not isinstance(data, list):
+            data = [data]
+
+        for _ in data:
+            id_ = _create_parameter_id()
+            _customized_parameter_ids.add(id_)
+
+        self.tuner.import_customized_data(data)
 
     def handle_report_metric_data(self, data):
         """
@@ -187,7 +193,8 @@ def _handle_final_metric_data(self, data):
             self.tuner.receive_trial_result(id_, _trial_params[id_], value, customized=customized,
                                             trial_job_id=data.get('trial_job_id'))
         else:
-            _logger.warning('Find unknown job parameter id %s, maybe something goes wrong.', _trial_params[id_])
+            _logger.warning('Find unknown job parameter id %s, maybe something goes wrong.', id_)
+            _logger.warning('_trial_params %s', _trial_params)
 
     def _handle_intermediate_metric_data(self, data):
         """Call assessor to process intermediate results

nni/tuner.py

@@ -219,6 +219,14 @@ def import_data(self, data: list[TrialRecord]) -> None:
         # data: a list of dictionarys, each of which has at least two keys, 'parameter' and 'value'
         pass
 
+    def import_customized_data(self, data: list[TrialRecord]) -> None:
+        """
+        Internal API under revising, not recommended for end users.
+        """
+        # Import resume data for avoiding duplications
+        # data: a list of dictionarys, each of which has at least two keys, 'parameter_id' and 'parameters'
+        pass
+
     def _on_exit(self) -> None:
         pass
 

test/ut/sdk/test_builtin_tuners.py

@@ -272,7 +272,7 @@ def import_data_test(self, tuner_factory, stype="choice_str", support_middle=Tru
             search_space = {
                 "choice_str": {
                     "_type": "choice",
-                    "_value": ["cat", "dog", "elephant", "cow", "sheep", "panda"]
+                    "_value": ["cat", "dog", "elephant", "cow", "sheep", "panda", "tiger"]
                 }
             }
         elif stype == "choice_num":

ts/nni_manager/common/datastore.ts

@@ -4,7 +4,7 @@
 import { ExperimentProfile, TrialJobStatistics } from './manager';
 import { TrialJobDetail, TrialJobStatus } from './trainingService';
 
-type TrialJobEvent = TrialJobStatus | 'USER_TO_CANCEL' | 'ADD_CUSTOMIZED' | 'ADD_HYPERPARAMETER' | 'IMPORT_DATA';
+type TrialJobEvent = TrialJobStatus | 'USER_TO_CANCEL' | 'ADD_CUSTOMIZED' | 'ADD_HYPERPARAMETER' | 'IMPORT_DATA' |'ADD_RESUMED';
 type MetricType = 'PERIODICAL' | 'FINAL' | 'CUSTOM' | 'REQUEST_PARAMETER';
 
 interface ExperimentProfileRecord {

ts/nni_manager/common/trainingService.ts

@@ -34,6 +34,7 @@ interface TrialJobApplicationForm {
     readonly sequenceId: number;
     readonly hyperParameters: HyperParameters;
     readonly placementConstraint?: PlacementConstraint;
+    id?: string;
 }
 
 interface TrialCommandContent {

ts/nni_manager/core/nnimanager.ts

@@ -23,7 +23,7 @@ import {
 import { delay, getCheckpointDir, getExperimentRootDir, getLogDir, getMsgDispatcherCommand, mkDirP, getTunerProc, getLogLevel, isAlive, killPid } from '../common/utils';
 import {
     INITIALIZE, INITIALIZED, KILL_TRIAL_JOB, NEW_TRIAL_JOB, NO_MORE_TRIAL_JOBS, PING,
-    REPORT_METRIC_DATA, REQUEST_TRIAL_JOBS, SEND_TRIAL_JOB_PARAMETER, TERMINATE, TRIAL_END, UPDATE_SEARCH_SPACE, IMPORT_DATA
+    REPORT_METRIC_DATA, REQUEST_TRIAL_JOBS, SEND_TRIAL_JOB_PARAMETER, TERMINATE, TRIAL_END, UPDATE_SEARCH_SPACE, IMPORT_DATA, ADD_CUSTOMIZED_TRIAL_JOB
 } from './commands';
 import { createDispatcherInterface, IpcInterface } from './ipcInterface';
 
@@ -43,6 +43,7 @@ class NNIManager implements Manager {
     private waitingTrials: TrialJobApplicationForm[];
     private trialJobs: Map<string, TrialJobDetail>;
     private trialDataForTuner: string;
+    private trialDataForResume: string;
     private readonly: boolean;
     private config!: ExperimentConfig;
 
@@ -55,6 +56,7 @@ class NNIManager implements Manager {
         this.waitingTrials = [];
         this.trialJobs = new Map<string, TrialJobDetail>();
         this.trialDataForTuner = '';
+        this.trialDataForResume = '';
         this.readonly = false;
 
         this.log = getLogger('NNIManager');
@@ -118,6 +120,45 @@ class NNIManager implements Manager {
         return this.dataStore.exportTrialHpConfigs();
     }
 
+    public addRecoveredTrialJob(allTrialJobs: Array<TrialJobInfo>): void {
+        const jobs: Array<TrialJobInfo> = allTrialJobs.filter((job: TrialJobInfo) => job.status === 'WAITING' || job.status === 'RUNNING');
+        const trialData: any[] = [];
+        let maxSequeceId = 0;
+        for (const job of jobs) {
+            if (job.sequenceId === undefined || job.hyperParameters === undefined) {
+                this.log.warning('The trial to be recovered missing sequenceId and/or hyperParameters', job);
+                continue;
+            }
+            const params: string = job.hyperParameters[0];
+            const sequenceId: number = job.sequenceId;
+            maxSequeceId = Math.max(maxSequeceId, sequenceId);
+
+            const hyperParams = JSON.parse(params);
+            const packedParameter = {
+                parameter_id: hyperParams['parameter_id'], // eslint-disable-line @typescript-eslint/camelcase
+                parameter_source: 'resumed', // eslint-disable-line @typescript-eslint/camelcase
+                parameters: hyperParams['parameters'],
+                parameter_index: hyperParams['parameter_index'], // eslint-disable-line @typescript-eslint/camelcase
+            }
+            const form: TrialJobApplicationForm = {
+                id: job.trialJobId,
+                sequenceId: sequenceId,
+                hyperParameters: {
+                    value: JSON.stringify(packedParameter),
+                    index: 0
+                },
+            };
+
+            this.waitingTrials.push(form);
+            trialData.push(packedParameter);
+            this.dataStore.storeTrialJobEvent('ADD_RESUMED', job.trialJobId, '');
+        }
+        this.trialDataForResume = JSON.stringify(trialData);
+
+        // next sequenceId
+        this.experimentProfile.nextSequenceId = maxSequeceId + 1;
+    }
+
     public addCustomizedTrialJob(hyperParams: string): Promise<number> {
         if (this.readonly) {
             return Promise.reject(new Error('Error: can not add customized trial job in readonly mode!'));
@@ -220,11 +261,7 @@ class NNIManager implements Manager {
 
         // Resume currSubmittedTrialNum
         this.currSubmittedTrialNum = allTrialJobs.length;
-
-        // Check the final status for WAITING and RUNNING jobs
-        await Promise.all(allTrialJobs
-            .filter((job: TrialJobInfo) => job.status === 'WAITING' || job.status === 'RUNNING')
-            .map((job: TrialJobInfo) => this.dataStore.storeTrialJobEvent('FAILED', job.trialJobId)));
+        this.addRecoveredTrialJob(allTrialJobs);
 
         // Collect generated trials and imported trials
         const finishedTrialData: string = await this.exportData();
@@ -807,6 +844,12 @@ class NNIManager implements Manager {
                     }
                     this.dispatcher.sendCommand(IMPORT_DATA, this.trialDataForTuner);
                 }
+                if (this.trialDataForResume.length > 0 ) {
+                    if (this.dispatcher === undefined) {
+                        throw new Error('Dispatcher error: tuner has not been setup');
+                    }
+                    this.dispatcher.sendCommand(ADD_CUSTOMIZED_TRIAL_JOB, this.trialDataForResume);
+                }
                 this.requestTrialJobs(this.experimentProfile.params.trialConcurrency);
                 break;
             }

ts/nni_manager/test/mock/datastore.ts

@@ -229,6 +229,8 @@ class MockedDataStore implements DataStore {
                 return 'USER_CANCELED';
             case 'ADD_CUSTOMIZED':
                 return 'WAITING';
+            case 'ADD_RESUMED':
+                return 'WAITING';
         }
         return <TrialJobStatus>event;
     }

ts/nni_manager/training_service/kubernetes/adl/adlTrainingService.ts

@@ -116,7 +116,7 @@ class AdlTrainingService extends KubernetesTrainingService implements Kubernetes
             this.kubernetesRestServerPort = restServer.clusterRestServerPort;
         }
 
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
         const adlJobName: string = `nni-exp-${this.experimentId}-trial-${trialJobId}`.toLowerCase();
         const initStatus: TrialJobStatus = 'WAITING';
         const codeDir = this.adlTrialConfig.codeDir;

ts/nni_manager/training_service/kubernetes/frameworkcontroller/frameworkcontrollerTrainingService.ts

@@ -131,7 +131,7 @@ class FrameworkControllerTrainingService extends KubernetesTrainingService imple
             await this.copyExpCodeDirPromise;
         }
 
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
         // Set trial's NFS working folder
         const trialWorkingFolder: string = path.join(this.CONTAINER_MOUNT_PATH, 'nni', getExperimentId(), trialJobId);
         const trialLocalTempFolder: string = path.join(getExperimentRootDir(), 'trials', trialJobId);

ts/nni_manager/training_service/kubernetes/kubeflow/kubeflowTrainingService.ts

@@ -78,7 +78,7 @@ class KubeflowTrainingService extends KubernetesTrainingService implements Kuber
             await this.copyExpCodeDirPromise;
         }
 
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
         const trialWorkingFolder: string = path.join(this.CONTAINER_MOUNT_PATH, 'nni', getExperimentId(), trialJobId);
         const kubeflowJobName: string = `nni-exp-${this.experimentId}-trial-${trialJobId}`.toLowerCase();
         const trialLocalTempFolder: string = path.join(getExperimentRootDir(), 'trials', trialJobId);

ts/nni_manager/training_service/local/localTrainingService.ts

@@ -193,7 +193,7 @@ class LocalTrainingService implements TrainingService {
     }
 
     public submitTrialJob(form: TrialJobApplicationForm): Promise<TrialJobDetail> {
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
         const trialJobDetail: LocalTrialJobDetail = new LocalTrialJobDetail(
             trialJobId,
             'WAITING',

ts/nni_manager/training_service/pai/paiTrainingService.ts

@@ -248,7 +248,7 @@ class PAITrainingService implements TrainingService {
     public async submitTrialJob(form: TrialJobApplicationForm): Promise<TrialJobDetail> {
         this.log.info('submitTrialJob: form:',  form);
 
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
         //TODO: use HDFS working folder instead
         const trialWorkingFolder: string = path.join(this.expRootDir, 'trials', trialJobId);
         const paiJobName: string = `nni_exp_${this.experimentId}_trial_${trialJobId}`;

ts/nni_manager/training_service/remote_machine/remoteMachineTrainingService.ts

@@ -226,7 +226,7 @@ class RemoteMachineTrainingService implements TrainingService {
      */
     public async submitTrialJob(form: TrialJobApplicationForm): Promise<TrialJobDetail> {
         // Generate trial job id(random)
-        const trialJobId: string = uniqueString(5);
+        const trialJobId: string = form.id === undefined ? uniqueString(5) : form.id;
 
         const trialJobDetail: RemoteMachineTrialJobDetail = new RemoteMachineTrialJobDetail(
             trialJobId,

ts/nni_manager/training_service/reusable/trialDispatcher.ts

@@ -160,7 +160,7 @@ class TrialDispatcher implements TrainingService {
     }
 
     public async submitTrialJob(form: TrialJobApplicationForm): Promise<TrialDetail> {
-        const trialId: string = uniqueString(5);
+        const trialId: string = form.id === undefined ? uniqueString(5) : form.id;
 
         const trialJobDetail: TrialDetail = new TrialDetail(trialId, "WAITING", Date.now(), "", form);