[Swift-devel] Dynamic profiles and count
Justin M Wozniak
wozniak at mcs.anl.gov
Tue Sep 18 21:40:54 CDT 2012
Just to clarify- this is based on changes already in svn regarding
dynamic profiles. That's why the map is there.
http://www.ci.uchicago.edu/swift/guides/trunk/userguide/userguide.html#_dynamic_profiles
David is trying to solve a specific issue with using dynamic profiles
and "count": we need to know all about "count" to solve this.
This is a paste of my recent email to David about dynamic profiles and
issues with "count":
Ok, the place to look is in the Swift repo: svn diff -r5206:5207
Apparently, no changes were made to CoG.
The dynamic profiles should be visible in the KML.
From there, you really need to know your Karajan semantics. The
attributes are known to execute2() and thus execute() and Swift's Execute.
Now Java. You should be able to see the attributes in GridExec. They
are applied to the Task's JobSpecification. These good places to add
some trace-level logging.
Then, the Task is sent to PBSExecutor. You could add logging here too
to make sure the attributes made it.
One thing to consider is the possibility that the attribute is being
overwritten by some other component. If you add logging to
JobSpecification.setAttribute(), you might be able to find that.
On 9/18/2012 4:58 PM, Mihael Hategan wrote:
> I'm not sure. The way I understand things, your change modifies the way
> execute is invoked from:
>
> execute(executable, ..., count=n)
>
> to:
>
> execute(executable, ..., attributes=map(map:entry("count", n), ...))
>
> But the semantics are the same. The task object, in both cases, will
> have an attribute named "count" equal to n.
>
> Can you send me the full diff of your changes?
>
> Mihael
>
> On Tue, 2012-09-18 at 13:33 -0500, David Kelly wrote:
>> Hello,
>>
>> I have been working on a namd scaling test using Swift. I am using the plain PBS provider at the moment (no coasters). The swift script sets a minimum number of nodes, a maximum number of nodes, iterates through those values, then uses dynamic profiles to change the value of 'count' to modify the number of nodes to request. Here is the script:
>>
>> ---
>> type file;
>>
>> app (file out, file err) namd_wrapper (int numnodes, file psf_file, file pdb_file, file coord_restart_file,
>> file velocity_restart_file, file system_restart_file)
>> {
>> profile "count" = numnodes;
>> namd_wrapper @psf_file @pdb_file @coord_restart_file @velocity_restart_file @system_restart_file stdout=@out stderr=@err;
>> }
>>
>> # Range of nodes to test on
>> int minNodes=1;
>> int maxNodes=2;
>> int delta=1;
>>
>> # Files
>> file psf <"h0_solvion.psf">;
>> file pdb <"h0_solvion.pdb">;
>> file coord_restart <"h0_eq.0.restart.coor">;
>> file velocity_restart <"h0_eq.0.restart.vel">;
>> file system_restart <"h0_eq.0.restart.xsc">;
>>
>> foreach nodes in [minNodes:maxNodes:delta] {
>> file output <single_file_mapper; file=@strcat("logs/scaling-", nodes, ".out.txt")>;
>> file error <single_file_mapper; file=@strcat("logs/scaling-", nodes, ".err.txt")>;
>> (output, error) = namd_wrapper(nodes, psf, pdb, coord_restart, velocity_restart, system_restart);
>> }
>> ---
>>
>> In sites.xml, I also set the jobtype to "single" so it doesn't start a worker on each node (namd uses MPI).
>>
>> The problem that I'm running into is that, as is, dynamic profiles do not seem to allow you to modify the value for count. I have a workaround for this which involves removing references to "count" in GridExec.java, Execute.java, and TCProfile.java. This works for me in terms of this script, and it works with a few other simple catsn type scripts I've tested. I just wanted to double check to make sure this wouldn't cause any other issues before committing. Here are the changes:
>>
>> Index: modules/karajan/src/org/globus/cog/karajan/workflow/nodes/grid/GridExec.java
>> ===================================================================
>> --- modules/karajan/src/org/globus/cog/karajan/workflow/nodes/grid/GridExec.java (revision 3472)
>> +++ modules/karajan/src/org/globus/cog/karajan/workflow/nodes/grid/GridExec.java (working copy)
>> @@ -56,7 +56,7 @@
>> public static final Arg A_STDIN = new Arg.Optional("stdin");
>> public static final Arg A_PROVIDER = new Arg.Optional("provider");
>> public static final Arg A_SECURITY_CONTEXT = new Arg.Optional("securitycontext");
>> - public static final Arg A_COUNT = new Arg.Optional("count");
>> + // public static final Arg A_COUNT = new Arg.Optional("count");
>> public static final Arg A_HOST_COUNT = new Arg.Optional("hostcount");
>> public static final Arg A_JOBTYPE = new Arg.Optional("jobtype");
>> public static final Arg A_MAXTIME = new Arg.Optional("maxtime");
>> @@ -86,7 +86,8 @@
>> static {
>> setArguments(GridExec.class, new Arg[] { A_EXECUTABLE, A_ARGS, A_ARGUMENTS, A_HOST,
>> A_STDOUT, A_STDERR, A_STDOUTLOCATION, A_STDERRLOCATION, A_STDIN, A_PROVIDER,
>> - A_COUNT, A_HOST_COUNT, A_JOBTYPE, A_MAXTIME, A_MAXWALLTIME, A_MAXCPUTIME,
>> + // A_COUNT,
>> + A_HOST_COUNT, A_JOBTYPE, A_MAXTIME, A_MAXWALLTIME, A_MAXCPUTIME,
>> A_ENVIRONMENT, A_QUEUE, A_PROJECT, A_MINMEMORY, A_MAXMEMORY, A_REDIRECT,
>> A_SECURITY_CONTEXT, A_DIRECTORY, A_NATIVESPEC, A_DELEGATION, A_ATTRIBUTES,
>> C_ENVIRONMENT, A_FAIL_ON_JOB_ERROR, A_BATCH, C_STAGEIN, C_STAGEOUT, C_CLEANUP,
>> @@ -346,7 +347,7 @@
>> }
>> }
>>
>> - protected final static Arg[] MISC_ATTRS = new Arg[] { A_COUNT, A_HOST_COUNT, A_JOBTYPE,
>> + protected final static Arg[] MISC_ATTRS = new Arg[] { A_HOST_COUNT, A_JOBTYPE,
>> A_MAXTIME, A_MAXWALLTIME, A_MAXCPUTIME, A_QUEUE, A_PROJECT, A_MINMEMORY, A_MAXMEMORY };
>>
>> protected void setMiscAttributes(JobSpecification js, VariableStack stack)
>>
>> Index: src/org/griphyn/vdl/karajan/lib/TCProfile.java
>> ===================================================================
>> --- src/org/griphyn/vdl/karajan/lib/TCProfile.java (revision 5930)
>> +++ src/org/griphyn/vdl/karajan/lib/TCProfile.java (working copy)
>> @@ -63,7 +63,6 @@
>>
>> static {
>> PROFILE_T = new HashMap<String, Arg>();
>> - PROFILE_T.put("count", GridExec.A_COUNT);
>> PROFILE_T.put("jobtype", GridExec.A_JOBTYPE);
>> PROFILE_T.put("maxcputime", GridExec.A_MAXCPUTIME);
>> PROFILE_T.put("maxmemory", GridExec.A_MAXMEMORY);
>>
>> Index: src/org/griphyn/vdl/karajan/lib/Execute.java
>> ===================================================================
>> --- src/org/griphyn/vdl/karajan/lib/Execute.java (revision 5930)
>> +++ src/org/griphyn/vdl/karajan/lib/Execute.java (working copy)
>> @@ -47,7 +47,7 @@
>> static {
>> setArguments(Execute.class, new Arg[] { A_EXECUTABLE, A_ARGS, A_ARGUMENTS, A_HOST,
>> A_STDOUT, A_STDERR, A_STDOUTLOCATION, A_STDERRLOCATION, A_STDIN, A_PROVIDER,
>> - A_COUNT, A_HOST_COUNT, A_JOBTYPE, A_MAXTIME, A_MAXWALLTIME, A_MAXCPUTIME,
>> + A_HOST_COUNT, A_JOBTYPE, A_MAXTIME, A_MAXWALLTIME, A_MAXCPUTIME,
>> A_ENVIRONMENT, A_QUEUE, A_PROJECT, A_MINMEMORY, A_MAXMEMORY, A_REDIRECT,
>> A_SECURITY_CONTEXT, A_DIRECTORY, A_NATIVESPEC, A_DELEGATION, A_ATTRIBUTES,
>> C_ENVIRONMENT, A_FAIL_ON_JOB_ERROR, A_BATCH, A_REPLICATION_GROUP,
>> _______________________________________________
>> Swift-devel mailing list
>> Swift-devel at ci.uchicago.edu
>> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel
>
> _______________________________________________
> Swift-devel mailing list
> Swift-devel at ci.uchicago.edu
> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel
--
Justin M Wozniak
More information about the Swift-devel
mailing list