/[pdpsoft]/nl.nikhef.pdp.dynsched-pbs-plugin/branches/RB-2.2.2/RELEASE
ViewVC logotype

Contents of /nl.nikhef.pdp.dynsched-pbs-plugin/branches/RB-2.2.2/RELEASE

Parent Directory Parent Directory | Revision Log Revision Log


Revision 2258 - (show annotations) (download)
Tue Apr 5 12:34:37 2011 UTC (11 years, 6 months ago) by templon
File size: 6349 byte(s)
had forgotten to document path stuff.

1 This file contains release notes and a change history for the
2 pbs/torque and maui plugins for the lcg-info-dynamic-scheduler information provider.
3 The notes are most recent first.
4
5 Release 2.2.2
6
7 A bug was fixed in parsing the output of "qstat -f"; the error condition had
8 not been previously reported, and was due to job staging errors being reported
9 via the "sched_hint" attribute. Empty lines in this error output broke the
10 parsing algorithm.
11
12 The lrmsinfo-pbs previously had the location of the python modules determined at
13 build time and embedded in the program. This has been removed as in EMI, everything
14 will go in system default locations (see also below about Makefile changes). For
15 local (non RPM) installs, you'll need to have PYTHONPATH set accordingly.
16
17 A testing framework was introduced (available in the svn, not in the RPM) with right now
18 a single test defined; which is whether the above situation can be correctly parsed.
19
20 The Makefile was changed in order to have the python library path be compliant with system
21 defaults.
22
23 Release 2.2.1
24
25 Dependencies changed back to the original set, so that this version's RPM
26 dependencies are compatible with those of previous versions (and also
27 with the lcg-info-dynamic-scheduler-generic package).
28
29 Release 2.2.0
30 Moved to new repository format and changed RPM name. Rather large changes in two areas. First, in the logFileParser
31 part of the pbsServer.py code. The new "account" field printed in the torque accounting logs broke the old parser,
32 this is fixed in the current release. The second set of changes have to do with how the LiveServer code parses
33 the 'qstat -f' output. When present (recent versions of torque), the code now uses 'startTime' as the start time
34 of the job, and uses the 'euser' and 'egroup' as the user and group of the job. If these fields are not present,
35 the code reverts to the old behavior. Using startTime will give slightly different results, since startTime
36 is present immediately upon job start, whereas walltime (used in the old calculation) is only updated once per
37 mom update interval, meaning there were always a few jobs in the system that were in state "running" but that had
38 no startTime. This will no longer be the case.
39
40 Three user-visible changes: there are two new job attributes in the "Job" object. One is the walltime used by the job,
41 the other is the 'startAnchor' field which tells you whether the code has used the 'startTime' to find the start time,
42 or has done it the old way by subtracting the walltime used from the current time. These data are both printed
43 by lrmsinfo-pbs. Those two changes are in principle user-visible, but only for those users directly using lrmsinfo-pbs.
44 Finally, vomaxjobs-maui accepts an argument '-k' to provide a key-file to the diagnose command. This is needed for
45 e.g. installations where the diagnose client cmd is on RHEL4 and the maui server is on RHEL5 (different build = different key).
46
47 Information below this point refers to older versions of the package, under the older naming
48 scheme.
49
50 Release 2.1.0
51 - Remove support for specifying hostname on command line (savannah bug 35662)
52
53 Release 2.0.1
54 - fix LICENSE, otherwise no change.
55
56 Release 2.0.0
57 - uses new lrms.py, otherwise no changes.
58
59 Release 1.6.1
60 - update documentation and examples to be consistent with 2.0 release
61
62 Release 1.6.0
63
64 - some changes to build system (three targets increases aggravation)
65 - some changes to pbsServer classes to assist in debugging.
66 - some changes to vomaxjobs-maui to assist in debugging/testing;
67 also fixed various unreported bugs discovered during testing.
68 - Change mapping of pbs/torque job states in pbs classes; up til now
69 was either queued (Q) or running (any other states). Now we have:
70
71 From the qstat (torque 2.0.0p4) man page:
72
73 C - Job is completed after having run (mapped to 'done')
74 E - Job is exiting after having run. (mapped to 'running')
75 H - Job is held. (mapped to 'pending')
76 Q - job is queued, eligible to run or routed. (mapped to 'queued')
77 R - job is running. (mapped to 'running')
78 T - job is being moved to new location. (mapped to 'pending')
79 W - job is waiting for its execution time (mapped to 'queued')
80
81 Release 1.5.2:
82
83 pbs package: Fix to vomaxjobs-maui to deal with cases where there is
84 extra 'warning' output near the top of the command output from diagnose -g.
85
86 Release 1.5.1:
87
88 fix dependency problems with RPMs.
89
90 Release 1.5.0
91
92 in vomaxjobs-maui, adapt to handle MAXPROC specifications like
93 MAXPROC=soft,hard The code reports the 'hard' limit, since
94 this is relevant when the system is not full, and this is when
95 it's needed. Maui uses the soft limit on a full system, but
96 in this case the info provider will drop FreeSlots to zero as
97 soon as jobs remain in the queued state instead of executing
98 immediately.
99
100 Release 1.4.2
101
102 in pbsServer.py: included Steve Traylen's patch to deal with jobs for which the
103 uid/gid printed by 'qstat' is not listed in the in running machine's
104 pw DB. This can happen when the CE is not the same physical
105 machine as the actual LRMS server.
106
107 Estimated Response Time Info Providers (v 1.4.1)
108 ------------------------------------------------
109
110 This information provider is new in LCG 2.7.0 and is
111 contained in two RPMs, lcg-info-dynamic-scheduler-generic
112 and lcg-info-dynamic-scheduler-pbs. Sites using torque/pbs
113 as an LRMS and Maui as a scheduler are fully supported by
114 this configuration; those using other schedulers and/or
115 LRMS systems will need to provide the appropriate back-end
116 plugins.
117
118 For sites meeting the following criteria, the system should
119 work out of the box with no modifications whatsoever:
120
121 LRMS == torque
122 scheduler == maui
123 vo names == unix group names of that vo's pool accounts
124
125 Documentation on what to do if this is not the case can be
126 found in the file
127
128 lcg-info-dynamic-scheduler.txt
129
130 in the doc directory
131
132 /opt/lcg/share/doc/lcg-info-dynamic-scheduler
133
134 There is also documentation in this directory indicating
135 the requirements on the backend commands you will need to
136 provide in the case that you are using a different
137 scheduler or LRMS. Tim Bell at CERN can help for people
138 using LSF.

Properties

Name Value
svn:keywords Id URL

grid.support@nikhef.nl
ViewVC Help
Powered by ViewVC 1.1.28