dcgm_sampler: fix handling of DCGM string fields

Use correct ldms function to add string metrics. Add error messages. Fixes #1321
ovis-hpc · Jan 20, 2024 · 36ddbf8 · 36ddbf8
1 parent a2bdf82
commit 36ddbf8
Showing 1 changed file with 17 additions and 2 deletions.
diff --git a/ldms/src/sampler/dcgm_sampler/dcgm_sampler.c b/ldms/src/sampler/dcgm_sampler/dcgm_sampler.c
@@ -318,9 +318,24 @@ static int gpu_schema_create()
 
                 field_meta = DcgmFieldGetById(conf.fields[i]);
                 ldms_type = dcgm_to_ldms_type(field_meta->fieldType);
-                rc = ldms_schema_metric_add(sch, field_meta->tag, ldms_type);
-                if (rc < 0)
+                if (ldms_type == LDMS_V_NONE) {
+                        ovis_log(mylog, OVIS_LERROR,
+                                 "DCGM field %d has a DCGM type %d, which is not supported by this sampler\n",
+                                 conf.fields[i], field_meta->fieldType);
+                        goto err2;
+                }
+                if (ldms_type == LDMS_V_CHAR_ARRAY) {
+                        rc = ldms_schema_metric_array_add(sch, field_meta->tag,
+                                                          ldms_type, field_meta->valueFormat->width+1);
+                } else {
+                        rc = ldms_schema_metric_add(sch, field_meta->tag, ldms_type);
+                }
+                if (rc < 0) {
+                        ovis_log(mylog, OVIS_LERROR,
+                                 "failed adding ldms metric to set for DCGM field %d, DCGM type %d\n",
+                                 conf.fields[i], field_meta->fieldType);
                         goto err2;
+                }
                 translation_table[conf.fields[i]].ldms_index = rc;
                 translation_table[conf.fields[i]].ldms_type = ldms_type;
         }