dataDistribution.py 5.32 KB
Newer Older
1
2
import random

3
import xtreemfs_client.osd as osd
4
5

"""
6
7
8
9
class to keep track of the osd (object storage device) locations of different folders, i.e.,
their physical location.

this class also allows to calculate a 'good' osd for new data, based on the
10
11
12
13
14
15
distribution known beforehand.
"""


class DataDistribution(object):
    def __init__(self):
16
        self.OSDs = {}
17

18
19
20
    def add_new_osd(self, osd_uuid):
        if osd_uuid in self.OSDs:
            print("key: " + osd_uuid + " is already present!")
21
            return
22
23
        new_osd = osd.OSD(osd_uuid)
        self.OSDs[osd_uuid] = new_osd
24

25
26
27
    def add_osd(self, new_osd):
        if new_osd.uuid in self.OSDs:
            print("key: " + new_osd.uuid + " is already present!")
28
            return
29
        self.OSDs[new_osd.uuid] = new_osd
30

31
32
33
34
35
    def add_osd_list(self, osd_list):
        for osdUuid in osd_list:
            if osdUuid not in self.OSDs:
                new_osd = osd.OSD(osdUuid)
                self.OSDs[osdUuid] = new_osd
36
37
38

    def get_osd_list(self):
        osd_list = []
39
        for osd_name in self.OSDs.keys():
40
41
42
            osd_list.append(osd_name)
        return osd_list

43
44
45
46
    def get_containing_osd(self, folder_id):
        for checked_osd in self.OSDs.values():
            if checked_osd.contains_folder(folder_id):
                return checked_osd
47
48
        return None

49
50
    def get_average_folder_size(self):
        total_size = 0
51
        total_number_of_folders = 0
52
53
54
        for one_osd in self.OSDs.values():
            total_size += one_osd.totalFolderSize
            total_number_of_folders += len(one_osd.folders)
55
56
        if total_number_of_folders == 0:
            return 0
57
58
59
60
61
62
63
64
65
        return total_size / total_number_of_folders

    def assign_new_osd(self, folder_id, new_osd):
        old_osd = self.get_containing_osd(folder_id)
        if old_osd is None:
            self.OSDs[new_osd].add_folder(folder_id, self.get_average_folder_size())
        else:
            self.OSDs[new_osd].add_folder(folder_id, self.OSDs[old_osd.uuid].folders[folder_id])
            self.OSDs[old_osd.uuid].remove_folder(folder_id)
66
67
68
69

    '''
    adds a list of folders to the data distribution.
    returns a list of assignments from folders to OSDs, for which (folders) there was previously no assignment.
70
71

    if the optional arguments are given, OSDs are assigned data proportionally to their ratio_parameter.
72
    '''
73

74
    def add_folders(self, folders, osd_information=None, ratio_parameter='', random_osd_assignment=False):
75
76
        new_folders = []
        for folder in folders:
77
78
79
            containing_osd = self.get_containing_osd(folder.id)
            if containing_osd is not None:
                containing_osd.add_folder(folder.id, folder.size)
80
81
82
83
84
            else:
                new_folders.append(folder)

        osds_for_new_folders = []

85
86
87
88
89
90
91
92
93
94
95
        # totally random OSD assignment, even ignoring OSD capacities
        # (might lead to I/O errors when too many groups are assigned to an OSD)
        if random_osd_assignment:
            for folder in new_folders:
                osds_for_new_folders.append((folder.id,
                                             random.choice(self.OSDs.values())))

        # balanced deterministic OSD assignment
        # (following largest processing time first or post-greedy approach)
        list.sort(new_folders, key=lambda x: x.size, reverse=True)

96
97
98
99
100
101
102
103
104
105
106
107
        osd_ratios = {}
        if osd_information is not None and ratio_parameter != '':
            total_osd_size = 0
            for osd_size in osd_information.values():
                total_osd_size += osd_size[ratio_parameter]
            for osd_uuid, osd_size in osd_information.items():
                osd_ratios[osd_uuid] = float(osd_size[ratio_parameter]) / float(total_osd_size)

        else:
            for osd_uuid in self.OSDs.keys():
                osd_ratios[osd_uuid] = float(1)

108
        for folder in new_folders:
109
110
111
112
113
114
115
            least_used_osd = None
            for one_osd in self.OSDs.values():
                if (least_used_osd is None) or \
                                        one_osd.totalFolderSize / osd_ratios[one_osd.uuid] \
                                <= least_used_osd.totalFolderSize / osd_ratios[least_used_osd.uuid]:
                    least_used_osd = one_osd
            least_used_osd.add_folder(folder.id, folder.size)
116
117
            osds_for_new_folders.append((folder.id,
                                         least_used_osd.uuid))
118

119
120
121
122
123
        return osds_for_new_folders

    '''
    updates the size of a given folder
    '''
124

125
    def update_folder(self, folder, size):
126
127
128
        for one_osd in self.OSDs.values():
            if folder in one_osd.folders.keys():
                one_osd.update_folder(folder, size)
129
130
131
132

    '''
    generates a string describing this data distribution
    '''
133

134
135
    def description(self):
        string = ""
136
137
        for one_osd in self.OSDs.values():
            string += str(one_osd)
138
            string += "\n"
139
            string += "folders : " + str(one_osd.folders)
140
            string += "\n"
141
        string += "average folder size: " + str(self.get_average_folder_size())
142
143
144
        return string

    def __str__(self):
145
146
147
148
149
        string_representation = "DataDistribution has " + str(len(self.OSDs)) \
                                + " osds: \n"
        for key, value in self.OSDs.items():
            string_representation += str(value) + " \n"
        return string_representation