1 | #!/usr/bin/env python |
---|
2 | # |
---|
3 | # Author: Bas van der Vlies <basv@sara.nl> |
---|
4 | # Date : 17 Aug 2001 |
---|
5 | # Desc. : This script displays the status of the PBS batch. |
---|
6 | # It display the status of a node. The ideas are |
---|
7 | # based on the awk-script of Willem Vermin |
---|
8 | # |
---|
9 | # CVS info: |
---|
10 | # $Id: pbsmon.py 242 2010-07-08 12:57:08Z bas $ |
---|
11 | # |
---|
12 | import sys |
---|
13 | import string |
---|
14 | import re |
---|
15 | |
---|
16 | import pbs |
---|
17 | |
---|
18 | pbs_ND_free_and_job = 'not_all_procs_used' |
---|
19 | |
---|
20 | translate_state = { |
---|
21 | pbs.ND_free : '_', |
---|
22 | pbs.ND_down : 'X', |
---|
23 | pbs.ND_offline : '.', |
---|
24 | pbs.ND_reserve : 'R', |
---|
25 | pbs.ND_job_exclusive : 'J', |
---|
26 | pbs.ND_job_sharing : 'S', |
---|
27 | pbs.ND_busy : '*', |
---|
28 | pbs.ND_state_unknown : '?', |
---|
29 | pbs.ND_timeshared : 'T', |
---|
30 | pbs.ND_cluster : 'C', |
---|
31 | pbs_ND_free_and_job : 'j' |
---|
32 | } |
---|
33 | |
---|
34 | |
---|
35 | |
---|
36 | def display_cluster_status(nl, sl): |
---|
37 | |
---|
38 | # Thanks to Daniel Olson, we have now code that can handle |
---|
39 | # 2 and 3 digit hostname numbers |
---|
40 | # |
---|
41 | if len(nl) == 1: |
---|
42 | width = len( nl[0] ) |
---|
43 | else: |
---|
44 | width = len( nl[-1] ) |
---|
45 | |
---|
46 | # Determine what format we have to use |
---|
47 | # |
---|
48 | if width == 3: |
---|
49 | step = end = 19 |
---|
50 | format = '%3s' |
---|
51 | else: |
---|
52 | step = end = 25 |
---|
53 | format = '%2s' |
---|
54 | |
---|
55 | start = 0 |
---|
56 | |
---|
57 | items = len(nl) |
---|
58 | |
---|
59 | # Sanity check |
---|
60 | if end > items: |
---|
61 | end = items |
---|
62 | |
---|
63 | while start < items: |
---|
64 | |
---|
65 | print ' ', |
---|
66 | for j in range(start,end): |
---|
67 | print format %(nl[j]) , |
---|
68 | |
---|
69 | print '\n ', |
---|
70 | for j in range(start,end): |
---|
71 | print format %(sl[j]) , |
---|
72 | |
---|
73 | print '\n' |
---|
74 | |
---|
75 | start = end |
---|
76 | end = end + step |
---|
77 | |
---|
78 | if end > items: |
---|
79 | end = items |
---|
80 | |
---|
81 | # Now some statistics |
---|
82 | # |
---|
83 | n = 0 |
---|
84 | for key in translate_state.keys(): |
---|
85 | value = translate_state[key] |
---|
86 | print "%3s %-21s : %d\t |" %( value, key, sl.count(value) ), |
---|
87 | if n%2: |
---|
88 | print '' |
---|
89 | n = n + 1 |
---|
90 | |
---|
91 | def main(): |
---|
92 | state_list = [] |
---|
93 | node_list = [] |
---|
94 | node_nr = 0 |
---|
95 | |
---|
96 | if len(sys.argv) > 1: |
---|
97 | pbs_server = sys.argv[1] |
---|
98 | else: |
---|
99 | pbs_server = pbs.pbs_default() |
---|
100 | if not pbs_server: |
---|
101 | print "No default pbs server, usage: pbsmon [server] " |
---|
102 | sys.exit(1) |
---|
103 | |
---|
104 | con = pbs.pbs_connect(pbs_server) |
---|
105 | if con < 0: |
---|
106 | errno, text = pbs.error() |
---|
107 | print errno, text |
---|
108 | sys.exit(1) |
---|
109 | |
---|
110 | # We are only interested in the state and jobs of a node |
---|
111 | # |
---|
112 | attrl = pbs.new_attrl(2); |
---|
113 | attrl[0].name='state' |
---|
114 | attrl[1].name = 'jobs' |
---|
115 | |
---|
116 | |
---|
117 | nodes = pbs.pbs_statnode(con, "", attrl, "NULL") |
---|
118 | |
---|
119 | # Some is het None dan weer NULL, beats me |
---|
120 | # |
---|
121 | for node in nodes: |
---|
122 | |
---|
123 | # display_node_status(batch_info) |
---|
124 | node_attr = node.attribs |
---|
125 | |
---|
126 | # A node can have serveral states, huh. We are only |
---|
127 | # interested in first entry. |
---|
128 | # |
---|
129 | temp = string.splitfields(node_attr[0].value, ',') |
---|
130 | state = temp[0] |
---|
131 | |
---|
132 | # look if on a free node a job is scheduled, then mark it |
---|
133 | # as other state |
---|
134 | # |
---|
135 | if state == pbs.ND_free: |
---|
136 | if len([x for x in node_attr if x.name == 'jobs']): |
---|
137 | state_list.append(translate_state[pbs_ND_free_and_job]) |
---|
138 | else: |
---|
139 | state_list.append(translate_state[state]) |
---|
140 | else: |
---|
141 | state_list.append(translate_state[state]) |
---|
142 | |
---|
143 | |
---|
144 | re_host = re.compile(r""" |
---|
145 | |
---|
146 | (?P<name>\d+) |
---|
147 | |
---|
148 | """, re.VERBOSE) |
---|
149 | |
---|
150 | result = re_host.search(node.name) |
---|
151 | if result: |
---|
152 | node_list.append( result.group('name') ) |
---|
153 | else: |
---|
154 | node_nr = node_nr + 1 |
---|
155 | node_list.append( str(node_nr) ) |
---|
156 | |
---|
157 | display_cluster_status(node_list, state_list) |
---|
158 | |
---|
159 | main() |
---|