Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] using hnp_always_use_plm
From: Ralph Castain (rhc_at_[hidden])
Date: 2009-12-18 11:59:59


Yikes - you most definitely do -not- want to do that. It -might- work in some circumstances, but it will lead to considerable confusion in the system in most cases.

We have had that accidentally happen to users - took a long time to debug what was going on. Definitely don't want to go back to that day!

On Dec 18, 2009, at 9:10 AM, Damien Guinier wrote:

> Sorry , I am not clear. That true, the same PLM is use on all node.
> my parameter name is incorrect , "mpirun_not_as_orted" is better. My problem is simple:
> - I want "mpirun" haven't the "orted" launch feature.
> - To create processes on the "mpirun node", I want launch by "plm" an "orted" ( on this "mpirun node"), and ask orted, to create processes.
>
> With this way, process tracker, debug tools have no difference between nodes.
>
> Sorry for this confusion.
> Damien
>
> Ralph Castain a écrit :
>> It isn't necessary. The orted already will open and use the local plm if you simply set OMPI_MCA_plm=foo in its environment. The rsh, tm, and slurm plm modules already do this so that they can execute a tree-like spawn (for rsh) and because I needed ssh on the backend nodes to locally launch "slaves" on RoadRunner and other machines.
>>
>> The required code (already in those modules) is:
>>
>> /* enable local launch by the orteds */
>> var = mca_base_param_environ_variable("plm", NULL, NULL);
>> opal_setenv(var, "rsh", true, &env);
>> free(var);
>>
>>
>> You don't want the orted using the hnp ess module as it will then try to track its own launches and totally forget that it is a remote orted with slightly different responsibilities.
>>
>> If you need it to execute a different plm on the backend, please let me know - it is a trivial change to allow specification of remote launch agents, and we should do it for them all if we do.
>>
>> Ralph
>>
>> On Dec 18, 2009, at 7:43 AM, Damien Guinier wrote:
>>
>>> Hi Ralph
>>>
>>> On Openmpi, I working on a new little feature: hnp_always_use_plm.
>>> - To create final application , mpirun use on remote "orted via plm: Process lifecycle managment module" or localy "fork()". So the first compute node haven't the same methode than other compute node. Some debug tools(padb ...) and management tools (squeus -s ...) are impacted by this difference.
>>> To simplify this cluster tools usage, I propose to add the possibility to use "orted via plm" on remote and localy.
>>>
>>> I make a patch to add the parameter "OMPI_MCA_ess_hnp_always_use_plm", to use the "plm" module everywhere. On my patch , by default nothing is changed ( no impact).
>>>
>>> Can you say to me , if this feature( and the patch) is good ?
>>>
>>> thank you
>>>
>>> Damien
>>>
>>> diff orte/mca/ess/hnp/ess_hnp.h
>>> --- a/orte/mca/ess/hnp/ess_hnp.h Tue Dec 15 15:31:24 2009 +0100
>>> +++ b/orte/mca/ess/hnp/ess_hnp.h Tue Dec 15 18:19:18 2009 +0100
>>> @@ -27,7 +27,7 @@
>>> int orte_ess_hnp_component_open(void);
>>> int orte_ess_hnp_component_close(void);
>>> int orte_ess_hnp_component_query(mca_base_module_t **module, int *priority);
>>> -
>>> +extern int mca_ess_hnp_always_use_plm;
>>>
>>> ORTE_MODULE_DECLSPEC extern orte_ess_base_component_t mca_ess_hnp_component;
>>>
>>> diff orte/mca/ess/hnp/ess_hnp_component.c
>>> --- a/orte/mca/ess/hnp/ess_hnp_component.c Tue Dec 15 15:31:24 2009 +0100
>>> +++ b/orte/mca/ess/hnp/ess_hnp_component.c Tue Dec 15 18:19:18 2009 +0100
>>> @@ -33,6 +33,7 @@
>>> #include "orte/mca/ess/hnp/ess_hnp.h"
>>>
>>> extern orte_ess_base_module_t orte_ess_hnp_module;
>>> +int mca_ess_hnp_always_use_plm = 0;
>>>
>>> /*
>>> * Instantiate the public struct with all of our public information
>>> @@ -63,6 +64,10 @@
>>> int
>>> orte_ess_hnp_component_open(void)
>>> {
>>> + mca_base_param_reg_int(&mca_ess_hnp_component.base_version,
>>> + "always_use_plm",
>>> + "Used to force plm on all machine",
>>> + false,false, mca_ess_hnp_always_use_plm ,&mca_ess_hnp_always_use_plm);
>>> return ORTE_SUCCESS;
>>> }
>>>
>>> diff orte/mca/ess/hnp/ess_hnp_module.c
>>> --- a/orte/mca/ess/hnp/ess_hnp_module.c Tue Dec 15 15:31:24 2009 +0100
>>> +++ b/orte/mca/ess/hnp/ess_hnp_module.c Tue Dec 15 18:19:18 2009 +0100
>>> @@ -442,9 +442,12 @@
>>> * node object
>>> */
>>> OBJ_RETAIN(proc); /* keep accounting straight */
>>> + if(mca_ess_hnp_always_use_plm==0)
>>> + {
>>> node->daemon = proc;
>>> node->daemon_launched = true;
>>> node->state = ORTE_NODE_STATE_UP;
>>> + }
>>>
>>> /* record that the daemon job is running */
>>> jdata->num_procs = 1;
>>>
>>
>