1 | using System;
|
---|
2 | using System.Linq;
|
---|
3 | using HeuristicLab.Services.Hive.DataAccess;
|
---|
4 | using HeuristicLab.Services.Hive.DataAccess.Interfaces;
|
---|
5 |
|
---|
6 | namespace HeuristicLab.Services.Hive.Manager {
|
---|
7 | public class NewEventManager : IEventManager {
|
---|
8 | private const string SlaveTimeout = "Slave timed out.";
|
---|
9 | private IPersistenceManager PersistenceManager {
|
---|
10 | get { return ServiceLocator.Instance.PersistenceManager; }
|
---|
11 | }
|
---|
12 | private IAuthorizationManager AuthorizationManager {
|
---|
13 | get { return ServiceLocator.Instance.AuthorizationManager; }
|
---|
14 | }
|
---|
15 | private ILogger Log {
|
---|
16 | get { return LogFactory.GetLogger(this.GetType().Namespace); }
|
---|
17 | }
|
---|
18 |
|
---|
19 | public void Cleanup() {
|
---|
20 | using (var pm = PersistenceManager) {
|
---|
21 | // same transactions as the old EventManager
|
---|
22 | pm.UseTransaction(() => {
|
---|
23 | SetTimeoutSlavesOffline(pm);
|
---|
24 | SetTimeoutTasksWaiting(pm);
|
---|
25 | DeleteObsoleteSlaves(pm);
|
---|
26 | pm.SubmitChanges();
|
---|
27 | });
|
---|
28 |
|
---|
29 | pm.UseTransaction(() => {
|
---|
30 | FinishParentTasks(pm);
|
---|
31 | pm.SubmitChanges();
|
---|
32 | });
|
---|
33 | }
|
---|
34 | }
|
---|
35 |
|
---|
36 | /// <summary>
|
---|
37 | /// Searches for slaves which are timed out, puts them and their task offline
|
---|
38 | /// </summary>
|
---|
39 | private void SetTimeoutSlavesOffline(IPersistenceManager pm) {
|
---|
40 | var slaveDao = pm.SlaveDao;
|
---|
41 | var slaves = slaveDao.GetOnlineSlaves();
|
---|
42 | foreach (var slave in slaves) {
|
---|
43 | if (!slave.LastHeartbeat.HasValue ||
|
---|
44 | (DateTime.Now - slave.LastHeartbeat.Value) >
|
---|
45 | HeuristicLab.Services.Hive.Properties.Settings.Default.SlaveHeartbeatTimeout) {
|
---|
46 | slave.SlaveState = SlaveState.Offline;
|
---|
47 | }
|
---|
48 | }
|
---|
49 | }
|
---|
50 |
|
---|
51 | /// <summary>
|
---|
52 | /// Looks for parent tasks which have FinishWhenChildJobsFinished and set their state to finished
|
---|
53 | /// </summary>
|
---|
54 | private void FinishParentTasks(IPersistenceManager pm) {
|
---|
55 | var resourceDao = pm.ResourceDao;
|
---|
56 | var taskDao = pm.TaskDao;
|
---|
57 | var resourceIds = resourceDao.GetAll().Select(x => x.ResourceId).ToList();
|
---|
58 | var parentTasksToFinish = taskDao.GetParentTasks(resourceIds, 0, true);
|
---|
59 | foreach (var task in parentTasksToFinish) {
|
---|
60 | task.State = TaskState.Finished;
|
---|
61 | task.StateLogs.Add(new StateLog {
|
---|
62 | State = task.State,
|
---|
63 | SlaveId = null,
|
---|
64 | UserId = null,
|
---|
65 | Exception = string.Empty,
|
---|
66 | DateTime = DateTime.Now
|
---|
67 | });
|
---|
68 | }
|
---|
69 | }
|
---|
70 |
|
---|
71 | /// <summary>
|
---|
72 | /// Looks for task which have not sent heartbeats for some time and reschedules them for calculation
|
---|
73 | /// </summary>
|
---|
74 | private void SetTimeoutTasksWaiting(IPersistenceManager pm) {
|
---|
75 | var taskDao = pm.TaskDao;
|
---|
76 | var tasks = taskDao.GetAll().Where(x => (x.State == TaskState.Calculating && (DateTime.Now - x.LastHeartbeat) > HeuristicLab.Services.Hive.Properties.Settings.Default.CalculatingJobHeartbeatTimeout)
|
---|
77 | || (x.State == TaskState.Transferring && (DateTime.Now - x.LastHeartbeat) > HeuristicLab.Services.Hive.Properties.Settings.Default.TransferringJobHeartbeatTimeout));
|
---|
78 | foreach (var task in tasks) {
|
---|
79 | task.State = TaskState.Waiting;
|
---|
80 | task.StateLogs.Add(new StateLog {
|
---|
81 | State = task.State,
|
---|
82 | SlaveId = null,
|
---|
83 | UserId = null,
|
---|
84 | Exception = SlaveTimeout,
|
---|
85 | DateTime = DateTime.Now
|
---|
86 | });
|
---|
87 | task.Command = null;
|
---|
88 | }
|
---|
89 | }
|
---|
90 |
|
---|
91 | /// <summary>
|
---|
92 | /// Searches for slaves that are disposable and deletes them if they were offline for too long
|
---|
93 | /// </summary>
|
---|
94 | private void DeleteObsoleteSlaves(IPersistenceManager pm) {
|
---|
95 | var slaveDao = pm.SlaveDao;
|
---|
96 | var downtimeDao = pm.DowntimeDao;
|
---|
97 | var slaveIds = slaveDao.GetAll()
|
---|
98 | .Where(x => x.IsDisposable.GetValueOrDefault()
|
---|
99 | && x.SlaveState == SlaveState.Offline
|
---|
100 | && (DateTime.Now - x.LastHeartbeat) > Hive.Properties.Settings.Default.SweepInterval)
|
---|
101 | .Select(x => x.ResourceId)
|
---|
102 | .ToList();
|
---|
103 | foreach (var id in slaveIds) {
|
---|
104 | bool downtimesAvailable = downtimeDao.GetByResourceId(id).Any();
|
---|
105 | if (!downtimesAvailable) {
|
---|
106 | slaveDao.Delete(id);
|
---|
107 | }
|
---|
108 | }
|
---|
109 | }
|
---|
110 | }
|
---|
111 | }
|
---|