1# Copyright 2016 The Chromium OS Authors. All rights reserved. 2# Use of this source code is governed by a BSD-style license that can be 3# found in the LICENSE file. 4 5"""Services relating to generating a suite timeline and report.""" 6 7from __future__ import print_function 8 9import common 10import json 11 12from autotest_lib.client.common_lib import time_utils 13from autotest_lib.server import frontend 14from autotest_lib.server.lib import status_history 15from chromite.lib import cros_logging as logging 16 17 18HostJobHistory = status_history.HostJobHistory 19 20# TODO: Handle other statuses like infra failures. 21TKO_STATUS_MAP = { 22 'ERROR': 'fail', 23 'FAIL': 'fail', 24 'GOOD': 'pass', 25 'PASS': 'pass', 26 'ABORT': 'aborted', 27 'Failed': 'fail', 28 'Completed': 'pass', 29 'Aborted': 'aborted', 30} 31 32 33def parse_tko_status_string(status_string): 34 """Parse a status string from TKO or the HQE databases. 35 36 @param status_string: A status string from TKO or HQE databases. 37 38 @return A status string suitable for inclusion within Cloud Datastore. 39 """ 40 return TKO_STATUS_MAP.get(status_string, 'unknown:' + status_string) 41 42 43def make_entry(entry_id, name, status, start_time, 44 finish_time=None, parent=None): 45 """Generate an event log entry to be stored in Cloud Datastore. 46 47 @param entry_id: A (Kind, id) tuple representing the key. 48 @param name: A string identifying the event 49 @param status: A string identifying the status of the event. 50 @param start_time: A unix timestamp of the start of the event. 51 @param finish_time: A unix timestamp of the finish of the event. 52 @param parent: A (Kind, id) tuple representing the parent key. 53 54 @return A dictionary representing the entry suitable for dumping via JSON. 55 """ 56 entry = { 57 'id': entry_id, 58 'name': name, 59 'status': status, 60 'start_time': start_time, 61 } 62 if finish_time is not None: 63 entry['finish_time'] = finish_time 64 if parent is not None: 65 entry['parent'] = parent 66 return entry 67 68 69def find_start_finish_times(statuses): 70 """Determines the start and finish times for a list of statuses. 71 72 @param statuses: A list of job test statuses. 73 74 @return (start_tme, finish_time) tuple of seconds past epoch. If either 75 cannot be determined, None for that time. 76 """ 77 starts = {int(time_utils.to_epoch_time(s.test_started_time)) 78 for s in statuses if s.test_started_time != 'None'} 79 finishes = {int(time_utils.to_epoch_time(s.test_finished_time)) 80 for s in statuses if s.test_finished_time != 'None'} 81 start_time = min(starts) if starts else None 82 finish_time = max(finishes) if finishes else None 83 return start_time, finish_time 84 85 86def make_job_entry(tko, job, parent=None, suite_job=False, job_entries=None): 87 """Generate a Suite or HWTest event log entry. 88 89 @param tko: TKO database handle. 90 @param job: A frontend.Job to generate an entry for. 91 @param parent: A (Kind, id) tuple representing the parent key. 92 @param suite_job: A boolean indicating wheret this represents a suite job. 93 @param job_entries: A dictionary mapping job id to earlier job entries. 94 95 @return A dictionary representing the entry suitable for dumping via JSON. 96 """ 97 statuses = tko.get_job_test_statuses_from_db(job.id) 98 status = 'pass' 99 dut = None 100 for s in statuses: 101 parsed_status = parse_tko_status_string(s.status) 102 # TODO: Improve this generation of status. 103 if parsed_status != 'pass': 104 status = parsed_status 105 if s.hostname: 106 dut = s.hostname 107 if s.test_started_time == 'None' or s.test_finished_time == 'None': 108 logging.warn('TKO entry for %d missing time: %s' % (job.id, str(s))) 109 start_time, finish_time = find_start_finish_times(statuses) 110 entry = make_entry(('Suite' if suite_job else 'HWTest', int(job.id)), 111 job.name.split('/')[-1], status, start_time, 112 finish_time=finish_time, parent=parent) 113 114 entry['job_id'] = int(job.id) 115 if dut: 116 entry['dut'] = dut 117 if job.shard: 118 entry['shard'] = job.shard 119 # Determine the try of this job by looking back through what the 120 # original job id is. 121 if 'retry_original_job_id' in job.keyvals: 122 original_job_id = int(job.keyvals['retry_original_job_id']) 123 original_job = job_entries.get(original_job_id, None) 124 if original_job: 125 entry['try'] = original_job['try'] + 1 126 else: 127 entry['try'] = 0 128 else: 129 entry['try'] = 1 130 entry['gs_url'] = status_history.get_job_gs_url(job) 131 return entry 132 133 134def make_hqe_entry(hostname, hqe, hqe_statuses, parent=None): 135 """Generate a HQE event log entry. 136 137 @param hostname: A string of the hostname. 138 @param hqe: A host history to generate an event for. 139 @param hqe_statuses: A dictionary mapping HQE ids to job status. 140 @param parent: A (Kind, id) tuple representing the parent key. 141 142 @return A dictionary representing the entry suitable for dumping via JSON. 143 """ 144 entry = make_entry( 145 ('HQE', int(hqe.id)), hostname, 146 hqe_statuses.get(hqe.id, parse_tko_status_string(hqe.job_status)), 147 hqe.start_time, finish_time=hqe.end_time, parent=parent) 148 149 entry['task_name'] = hqe.name.split('/')[-1] 150 entry['in_suite'] = hqe.id in hqe_statuses 151 entry['job_url'] = hqe.job_url 152 entry['gs_url'] = hqe.gs_url 153 if hqe.job_id is not None: 154 entry['job_id'] = hqe.job_id 155 entry['is_special'] = hqe.is_special 156 return entry 157 158 159def generate_suite_report(suite_job_id, afe=None, tko=None): 160 """Generate a list of events corresonding to a single suite job. 161 162 @param suite_job_id: The AFE id of the suite job. 163 @param afe: AFE database handle. 164 @param tko: TKO database handle. 165 166 @return A list of entries suitable for dumping via JSON. 167 """ 168 if afe is None: 169 afe = frontend.AFE() 170 if tko is None: 171 tko = frontend.TKO() 172 173 # Retrieve the main suite job. 174 suite_job = afe.get_jobs(id=suite_job_id)[0] 175 176 suite_entry = make_job_entry(tko, suite_job, suite_job=True) 177 entries = [suite_entry] 178 179 # Retrieve the child jobs and cache all their statuses 180 logging.debug('Fetching child jobs...') 181 child_jobs = afe.get_jobs(parent_job_id=suite_job_id) 182 logging.debug('... fetched %s child jobs.' % len(child_jobs)) 183 job_statuses = {} 184 job_entries = {} 185 for j in child_jobs: 186 job_entry = make_job_entry(tko, j, suite_entry['id'], 187 job_entries=job_entries) 188 entries.append(job_entry) 189 job_statuses[j.id] = job_entry['status'] 190 job_entries[j.id] = job_entry 191 192 # Retrieve the HQEs from all the child jobs, record statuses from 193 # job statuses. 194 child_job_ids = {j.id for j in child_jobs} 195 logging.debug('Fetching HQEs...') 196 hqes = afe.get_host_queue_entries(job_id__in=list(child_job_ids)) 197 logging.debug('... fetched %s HQEs.' % len(hqes)) 198 hqe_statuses = {h.id: job_statuses.get(h.job.id, None) for h in hqes} 199 200 # Generate list of hosts. 201 hostnames = {h.host.hostname for h in hqes if h.host} 202 logging.debug('%s distinct hosts participated in the suite.' % 203 len(hostnames)) 204 205 # Retrieve histories for the time of the suite for all associated hosts. 206 # TODO: Include all hosts in the pool. 207 if suite_entry['start_time'] and suite_entry['finish_time']: 208 histories = [HostJobHistory.get_host_history(afe, hostname, 209 suite_entry['start_time'], 210 suite_entry['finish_time']) 211 for hostname in sorted(hostnames)] 212 213 for history in histories: 214 entries.extend(make_hqe_entry(history.hostname, h, hqe_statuses, 215 suite_entry['id']) for h in history) 216 217 return entries 218 219def dump_entries_as_json(entries, output_file): 220 """Dump event log entries as json to a file. 221 222 @param entries: A list of event log entries to dump. 223 @param output_file: The file to write to. 224 """ 225 # Write the entries out as JSON. 226 logging.debug('Dumping %d entries' % len(entries)) 227 for e in entries: 228 json.dump(e, output_file, sort_keys=True) 229 output_file.write('\n') 230