On Wed, 25 Jul 2012 15:57:58 +0300
Alex Netes <ale...@mellanox.com> wrote:
> Hi Ira,
>
> On 16:53 Tue 03 Jul , Ira Weiny wrote:
> >
> >
> > When a node drops from the fabric the default behavior was to leave the node
> > data in the PerfMgr DB. Add the option (with default set to TRUE) to delete
> > these "inactive" or missing nodes from the DB.
> >
> > Signed-off-by: Ira Weiny <wei...@llnl.gov>
> > ---
[snip]
> > +}
> > +
> > /**********************************************************************
> > * Dump a reading vs the previous reading to stdout
> > **********************************************************************/
> > diff --git a/opensm/osm_subnet.c b/opensm/osm_subnet.c
> > index 662e591..06efa9c 100644
> > --- a/opensm/osm_subnet.c
> > +++ b/opensm/osm_subnet.c
> > @@ -493,6 +493,7 @@ static const opt_rec_t opt_tbl[] = {
> > { "perfmgr_max_outstanding_queries",
> > OPT_OFFSET(perfmgr_max_outstanding_queries), opts_parse_uint32, NULL, 0 },
> > { "perfmgr_ignore_cas", OPT_OFFSET(perfmgr_ignore_cas),
> > opts_parse_boolean, NULL, 0 },
> > { "event_db_dump_file", OPT_OFFSET(event_db_dump_file),
> > opts_parse_charp, NULL, 0 },
> > + { "perfmgr_rm_nodes", OPT_OFFSET(perfmgr_rm_nodes), opts_parse_boolean,
> > NULL, 0 },
>
> I guess that this option can be enabled to be changed on the fly, as it might
> be changed also via the console.
yes.
>
> > #endif /* ENABLE_OSM_PERF_MGR */
> > { "event_plugin_name", OPT_OFFSET(event_plugin_name), opts_parse_charp,
> > NULL, 0 },
> > { "event_plugin_options", OPT_OFFSET(event_plugin_options),
> > opts_parse_charp, NULL, 0 },
> > @@ -985,6 +986,7 @@ void osm_subn_set_default_opt(IN osm_subn_opt_t * p_opt)
> > OSM_PERFMGR_DEFAULT_MAX_OUTSTANDING_QUERIES;
> > p_opt->perfmgr_ignore_cas = FALSE;
> > p_opt->event_db_dump_file = NULL; /* use default */
> > + p_opt->perfmgr_rm_nodes = TRUE;
>
> Here you change the default behavior. Any thoughts what is the added value to
> leave disconnected nodes in the DB?
Originally I was thinking that leaving nodes in the DB would reduce the load
on the perfmgr when nodes disappears/reappears for "normal" reasons. For
example a full reboot of all compute nodes.
As we start to finally ramp up to use this I don't think the load is that
great and we are having issues with hardware replacements leaving nodes in the
DB and causing user confusion (for example they have 2 nodes of the same name
being reported). So for users I think the default is better set at TRUE.
Ira
>
> > #endif /* ENABLE_OSM_PERF_MGR */
> >
> > p_opt->event_plugin_name = NULL;
> > @@ -2009,12 +2011,15 @@ int osm_subn_output_conf(FILE *out, IN
> > osm_subn_opt_t * p_opts)
> > "perfmgr_sweep_time_s %u\n\n"
> > "# Max outstanding queries\n"
> > "perfmgr_max_outstanding_queries %u\n"
> > - "perfmgr_ignore_cas %s\n\n",
> > + "perfmgr_ignore_cas %s\n\n"
> > + "# Remove missing nodes from DB\n"
> > + "perfmgr_rm_nodes %s\n",
> > p_opts->perfmgr ? "TRUE" : "FALSE",
> > p_opts->perfmgr_redir ? "TRUE" : "FALSE",
> > p_opts->perfmgr_sweep_time_s,
> > p_opts->perfmgr_max_outstanding_queries,
> > - p_opts->perfmgr_ignore_cas ? "TRUE" : "FALSE");
> > + p_opts->perfmgr_ignore_cas ? "TRUE" : "FALSE",
> > + p_opts->perfmgr_rm_nodes ? "TRUE" : "FALSE");
> >
> > fprintf(out,
> > "#\n# Event DB Options\n#\n"
> > --
> > 1.7.1
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> > the body of a message to majord...@vger.kernel.org
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Ira Weiny
Member of Technical Staff
Lawrence Livermore National Lab
925-423-8008
wei...@llnl.gov
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
