source: trunk/CHANGES @ 226

=========== 4.0.0
  * pbs.py added some new functions for 2.4 version:
    - pbs_fbserver(void);
    - pbs_get_server_list(void);
    - pbs_sigjobasync(int connect, char *job_id, char *signal, char *extend);
    - pbs_alterjob_async(int connect, char *job_id, struct attrl *ATTRL, char *extend);
    - pbs_checkpointjob(int connect, char *job_id, char *extend);
    - log_ext(int,char *,char *,int);
    - log_init(char *, char *);
    - log_remove_old(char *,unsigned long);

    Note: version 2.1.X is still supported

  * autotools environment
    - Added support for finding pbs-config (borrowed for mpiexec source)
    - Detecting which version of torque is installed
    - only support installations that have pbs-config installed. 
    - Removed old code for installations that do not a have pbs-config
      installed.

  * PBSQuery.py
    - There was a bug in the new data interface (Resource_List for jobs)
      Reported & Patch by: Mark Roberts ( mark at gingergeeks dot co dot uk)
      Applied by with some minor changes: Bas van der Vlies

    - The new data structure is the default. You can switch back to the old
      one with:
            p = PBSQuery()
            p.old_data_structure()

      Author: Bas van der Vlies

   * examples/ha_server.py
     An example how to use the High Availabilty torque functions
     Author: Bas van der Vlies & Mark Roberts ( mark at gingergeeks dot co dot uk )

=========== 3.6.0
  * pbs, new generated pbs_wrap.c, pbs.py fixes a bug in pbs_runjob()
    Reported by: Dennis Stam
    Fixed by: Bas van der Vlies 

=========== 3.5.0
  * PBSQuery
        The class functions of node, job and queue support old and new data
        structure.

        Changed the behaviour of the new data stucture, We can use it as
        dictionary and as class attribute, this is equivalent, eg:
          - print node['np'] and print node.np

        for a node we parse the 'status' line and split on '=' char, You now can
        use these statements, eg
          - print node.status.arch     (node['status'].arch or node['status']['arch'])
          - print node.status.nsession

        for a job we parse the 'Variable_List' line and split on '=' char, You now can
        use the statements, eg:
          - print job.Variable_List.PBS_O_WORKDIR
          - print job.Variable_List.PBS_O_HOME

        For more info see examples/new_interface.py

    Author: Bas van der Vlies

 * new_rack_pbsmon.py
        Rewrite to new data structure and automatically determine how many nodes
        and racks cluster has and skip printing of empty racks (default), use -w/--wide
        for old behaviour.

   Author: Bas van der Vlies


=========== 3.2.0
        - PBSQuery:
          New data structure. In the old structure it is a dictionary
          with a value and the value is a string. This is changed
          that dictionary values are now of type list or dictionary depends
          on the value of the keyword, eg for a node:
            - np = 2:
              * node['np'] = [ '2' ]  

            - properties = cores2, mem4gb, parallel 
              * node['properties'] = [ 'cores2', 'mem4gb', 'parallel' ]

            - status = arch=x86_64,sessions=22599,,size=70627864kb, ...
              * node['status']['arch'] = [ 'x86_64' ]
              * node['status']['sessions'] = [ '222599' ]
              * ...

         The data structure is activated by the function:
          * new_data_structure()

        In a future release it will be come the default.
        example:
                p = PBSQuery()
                p.new_data_structure()

                nodes = p.getnodes()
                print nodes.np, nodes['np']

        Author: Bas van der Vlies

        - PBSQuery
          For old and new data structure we now can use another syntax:
           * node['np'] and node.np are equivalent

          This new syntax works for all keywords.

        Author: Bas van der Vlies

        - Added iter object for job, node, queue and server objects, eg:
          node = p.getnode('gb-r1n1')
          print node.name
          for attrib in node:
                print '%\t%s = %s' %(attrib, node[attrib]) 
        Author: Bas van der Vlies

        - fixed an error in getnode, getqueue and getjob, return empty
          dictionary if not found
        Author: Bas van der Vlies
        
        - New build system for rpm packages, make -f Makefile.rpm
          Author: Michel Jouvin <jouvin add lal dot in2p3 dot fr>

        Applied: Bas van der Vlies

=========== 3.0.1
        - PBSQuery
                * Removed a testing code line
                Reported By: sam @ vpac dot org
                Fixed by: Bas van der Vlies

=========== 3.0.0
        - PBSQuery
                * Cleaned up some old lines of code which didn't do anything.
                * Changed the returnvalue of getqueue, getnode and getjob. Now
                  they return an instance of the object, instead of a dict.
                  This is API change so we bumped the version to 3.0.0
                * Now the getnode-function accepts the short name as well as
                  the long name as an argument.

                Author: Sil Westerveld <Sil.Westerveld@sara.nl>

        - pbs.py, _pbs.so, PBSQuery.py
                * Fixed a memory leak when used in daemon mode
                Author: Bas van der Vlies <basv@sara.nl>

=========== 2.9.8 
        - pbs_python.spec file patch added libdir and python defines so it 
          will build for RHEL5.1/CentOS 5.1
          by        : Michael Sternberg <sternberg add anl dot gov>
          applied by: Bas van der Vlies

        - Updated the header files for pbs_python. So it can use some new 
          defines variables/defines, eg:
                * ATTR_NODE_status
                * ATTR_NODE_note
                * ...
          Added by: Dennis Stam <dennis.stam@sara.nl>, Bas van der Vlies

        - Two new functions to log.h:
                log_roll(int size);
                log_size(void);
          Added by: Bas van der Vlies

        - We now have proper destructors (free memory) in SWIG for attrl and 
          attropl attributes.
          Added by: Bas van der Vlies

        - PBSQuery
                * Added getnodes_with_property() function. This will get all nodes with
                  a certain property, eg:
                   - get all nodes with property 'cores8' and return all attributes:
                      getnodes_with_property('cores8')
                
                   - get all nodes with property 'cores2' and only return the 
                     attributes 'state' and 'properties':
                      attrl = [ 'state', 'properties' ]
                      getnodes_with_property('cores2', attrl)
                
                     Only for torque version > 2.1.0 
             
                Added by: Bas van der Vlies

                * Added to class node the function
                 - get_jobs(self, unique=None)
                   Returns a list of the currently running job-id('s) on the node.
                Added by: Sil Westerveld <Sil.Westerveld@sara.nl>

                * Added get_server_name() function.
                  This will return the PBS-server's name.
                Added by: Sil Westerveld <Sil.Westerveld@sara.nl>

                * Added to class _PBSobject the function
                  - uniq(self, list)
                    Filters out unique items of a list.
                Added by: Sil Westerveld <Sil.Westerveld@sara.nl>

                * Added to class job the function
                  - get_nodes(self, unique=None)
                    Returns a list of the nodes which run the job.
                Added by: Sil Westerveld <Sil.Westerveld@sara.nl>

=========== Version 2.9.4
        -  Function pbs.pbs_geterrmsg does not exits any more is
           replaced by pbs.error():
                errno, text = pbs.error()
                print errno, text
           Reported by: pk at q-leap dot com
           Fixed by   : Bas van der Vlies

        - PBSQuery.py
           * Added get_version() for server class, returns the version of 
             the batch server
             Added by    : Bas van der Vlies

           * Added get_nodes() for job class, returns a list of nodes on 
             which the job is run
             Added by    : Bas van der Vlies

           * PBSQuery.py rewrite make use of UserDict module.
             Suggested by: Ramon Bastiaans
             Fixed by    : Bas van der Vlies

=========== Version 2.9.2
        - The function pbs_rescquery() did not work. Fixed it
        - The function avail(), did not work fixed it
        - The functions pbs_rescreserve and pbs_rescrelease did not work
          fixed it

        - Fixed a bug in new_rack_pbsmon.py. Counting of serial
          nodes was wrong and state down has topmost priority

=========== Version 2.9.0
        - Added support for new torque build system 2.1.0
        - Added new_rack_pbsmon.py
          Authors: Walter de Jong & Bas van der Vlies

=========== Version 2.8.2

Configure has a new option:
        --with-pbsdir=PATH

If set it will use PATH specified as argument to
find the pbs/torque libraries. If unset it will use
some useful defaults to find the libaries.
Suggested by: Troy Baer

PBSQuery.py:
        - If we can not make a connection with the server
          then raise an exception and let the programmer 
          decides what to do.

pbs_python.spec:
        - It can now handle mode bits
          Author: Martin Pels from SARA
        
=========== Version 2.8.0

setup.py:
        - Added /usr/lib to PBS_LIB_DIR

pbs.py:
        - Added new function "pbs.error()". This function checks if 
          an error has occured with a pbs function, eg:
            task_id = pbs.pbs_submit(c, attrl, "A.tsk", 'NULL', 'NULL')
            error_number, err_txt = pbs.error()
            if error_number:
                print error_number, err_txt

          prints the following message if script A.tsk does not exists:
            15042 (qsub) cannot access script file

pbs_python.spec:
        - to make rpm packages from the source package, initial version
        Author: Martin Pels from SARA

README:
        - Now requires python 2.1 and higher
        - Explain how to build DEBIAN package

=========== Version 2.7.10

PBSQuery.py:
        Fixed an error. You always got all resources from eg nodes
        even when you supplied an attribute list where you only
        requested the 'state' of the node.

        Updated the inline python documentation

=========== Version 2.7.9

Fixed an error the previous version was in DEBUG mode. So
you get an lot undesired output on the screen.

debian package now also generated a package with the write 
version info.

=========== Version 2.7.8

Setup.py:
        Can also handle ROCKS installations + the version
        info is now the same as pbs.py
        Thanks to: Roy Dragseth

=========== Version 2.7.7

PBSQuery module: 
  It can now be used in daemon programs. The pbs_server closes
  the connection after an certain amount of time. This is fixed,
  before every query there is an new connection and if the query 
  is finished the connection is closed.
  
=========== Version 2.7.6

Now setup.py can also handle old Oscar Installations. 
Thanks To: Robin Humble

Fixed an print statement fix in new_interface.py:
Thanks To: Robin Humble

Made some improvements for PBSQuery module:
        - Updated the documentation
        - Added the has_key() function to all
          PBSObjects. So that the behaviour is similar to 
          an dictionary
          Suggested By: Ramon Bastiaans
        
=========== Version 2.7.5
 Added PBSQuery module.  This module requires pbs.py and it a simple module 
 for querying the pbs server. The documentation is in the module. Use for 
 example ipython to read it.

 Written By: Roy Dragseth 
             Bas van der Vlies

 eg: ( see also examples/new_interface.py)

 from PBSQuery import PBSQuery
 p = PBSQuery()

 jobs = p.getjobs()
 for name, job in jobs.items()
        print job

=========== Version 2.7.4
 New versions fixes an bug in function pbs.pbs_statnode(). 
 Reported by: Keith Poirier 

 Configure support:
 Contributed By: Yaroslav Halchenko

 Debian package support:
 Contributed By: Yaroslav Halchenko

 Added Support for Debuging the interface, You must edit pbs_wrap.c 
 and search for SARA_DEBUG:
        #define SARA_DEBUG 1
 Implemented by: Ramon Bastiaans

 examples/pbsmon.py:
        - Fixed an error when regex fails to determine node number, 
          no status was displayed
        - It will now display and 'j' if the node is free for the batch
          system and a job runs on the node (SMP-systems).
 Changed by: Bas van der Vlies

 examples/rack_pbsmon.py:
        - An pbsmon that display node info per rack
        - Edit the rack_pbsmon.py to adjust the values. It requires
          that hostname contains rack and node id's, 
          eg: gb-r<number>n<number>
 Contributed by: Walter de Jong

=========== Version 2.7.3
 The name SPBS is changed to the new name TORQUE (Tera-scale Open-source
 Resource and QUEing manager). 

 The interface support OPENPBS and TORQUE.

=========== Version 2.7.2
 Some minor changes to include files. Now the Scalable PBS keywords
 are also supported ( server attribute names: node_ping_rate and
 node_check_rate). This does not interfere with the openpbs software.

 pbsmon.py could not handle an one node cluster. Fixed it.

=========== Version 2.7
 Forgot to wrap the pbs_statfree() function. So we could not free
 allocated memory from functions that return 'struct batch_status *' 
 like pbs_statjob().

 pbsmon.py can now handle 2 and 3 digit hostnames. Patch supplied by
 Daniel Olson

=========== Version 2.6.1
 Setup.py now checks if all openpbs libraries are installed to compile the 
 package.

=========== Version 2.6

 Fixed a bug in the pbs python module. Forgot to map the pbs_statjob()
 function. This bug prevented to pass 'struct attrl' variables to this
 function. Thanks to Evelyn Shiu for reporting this bug.


=========== Version 2.5
 Added the OpenPbs logging functions (log.h and liblog.a). There
 is an example in the examples directory: logpbs.py

 Fixed a bug in examples/resmom_info.py used the default port for
 pbs_resmon if getservbyname fails.

 Fixed a bug in resmom code. If the user has no permissions to query
 the pbs_mom daemon a empty string is returned. Now we check for
 the empty string. So we do not get a python exception.

 Fixed a bug in setup.py forgot the exit statement when we did not
 find the pbs libraries.

=========== Version 2.3
 Bugfix in the pbsnodes-a.py. Forgot to import the sys module.

 Removed some obsolete code. 

 As for now we make use of the distutils module. So the package
 is automatically compiled and installed in proper directories, usage:
    python setup.py install
 This requires that the package 'distutils' must be installed. As for
 version 1.6 and higher it is in the distribution.

 Added a function that prints the pbs python interface version. So we
 can check if we have the right interface version:
 import pbs
 print pbs.version()
 

=========== Version 2.2

In this release we can query the pbs_mom daemon with the aid of
resource management functions:
   pbs.openrm()
   pbs.closerm()
   pbs.downrm()
   pbs.configrm()
   pbs.addreq()
   pbs.allreq()
   pbs.flushreq()
   pbs.activereq()
   pbs.fullresp()
   pbs.getreq()

There is also a python function:
   pbs.get_mom_values(id, [list]):
     id   - connection id of pbs.openrm(node-name, port-number)
     list - Is a optional paramter. A list of resource keywords

   If 'list' is not supplied then this function will get the values
   for the standard resource keywords, eg uname, loadave, .. + 'arch' 
   depended keywords. The arch feature is only implemented for linux, 
   but it can easily be ported to other oses.

   if 'list' is supplied then we only fetch values for the keywords
   that are in the list.

   The function returns a dictonary. The keys are the resource keywords.

   See 'resmom_info.py' for a example of this new feature.

=========== Version 2.0

The previous version was a simple interface above the PBS C API LIB.  This 
interface has gone a major changes. The most noticeable change is  that
the functions accept/return Python lists instead of C-lists.  So you can now 
use standard Python syntax for manipulating Python lists. If you are familar 
with the Python syntax then you will appreciate this new interface.

NOTE:
  This interface is NOT compatible with the old one

Here are some guidelines to convert your code to the new one:

The constructors for the struct has gone a major change:
  old code:
    temp = pbs.new_attrl();
    attrl_p = pbs.attrlPtr(temp)
    attrl_p.name = 'state'

  new_code:
     attr_l = pbs.new_attrl(1)   // Creates a list of attrl structs length 1
     attr_l[0].name = 'state'


The pbs_stat functions returns Python lists instead of C-lists. There is
NO next field anymore:
  old_code:
    temp = pbs.new_attrl();
    attrl_p = pbs.attrlPtr(temp)
    attrl_p.this = 'NULL'

    nodes = pbs.pbs_statnode(con, "", attrl_p, "NULL")

    while nodes.this != 'NULL':
      print nodes.name

      node_attrl = nodes.attribs
      while node_attrl.this != 'NULL':
        print '\t', node_attrl.name, '=', node_attrl.value
        node_attrl = node_attrl.next

    nodes = nodes.next 


  new_code:
    nodes = pbs.pbs_statnode(con, "", "NULL", "NULL")
    for node in nodes:
      print node.name
      for attrib in node.attribs:
        print '\t', attrib.name, '=', attrib.value

Another advantage is you can use the print statement to show the
connect of attrl and attropl classes:
   attr_l = pbs.new_attrl(2)
   attr_l[0].name = 'bas'
   attr_l[0].value = 'van der Vlies'

   print  attr_l[0]
   >> (bas,,van der Vlies)
    
I hope these examples illustrate the changes. If you specify a wrong type
for a function then function wil raise a Python exception.