ldms_ls option to print decomposition templates

[tom95858]

This feature dumps a decomposition template for each unqique schema published by an ldmsd. Although this template can be used directly, it is intended to be edited by the user to define indices of interest and rename/remove columns.

While developing this change it was discovered that the type names produced by ldms_metric_type_to_str() were not symmetric with those consumed by ldms_metric_str_to_type(). This change addresses this discrepency.

Also added is a #define for the minimum SHA256 schema digest string length LDMS_DIGEST_STR_LENGTH.

[tom95858]

@morrone, @narategithub Please take a look at this change and try it out if you can. I have confirmed that the template files created are successfully consumed by an ldmsd. This change does not bother to guess what the "indices" section might be, but creates it nonetheless for the user to edit. This mitigates the pain somewhat of discovering the digests and stuffing them in the digest section so that they can be mapped to schema decomposition rules.

Here's some example output:

{
  "type" : "flex",
  "decomposition" : {
     "slurm_job_data_decomp" : {
       "type" : "static",
       "rows" : [
          {
            "schema" : "slurm_job_data",
            "cols" : [
              { "src" : "context", "dst" : "context", "type" : "char[]", "array_len" : 255 },
              { "src" : "data", "rec_member" : "uid", "dst" : "uid", "type" : "s64" },
              { "src" : "data", "rec_member" : "total_tasks", "dst" : "total_tasks", "type" : "s64" },
              { "src" : "data", "rec_member" : "nodeid", "dst" : "nodeid", "type" : "s64" },
              { "src" : "data", "rec_member" : "local_tasks", "dst" : "local_tasks", "type" : "s64" },
              { "src" : "data", "rec_member" : "ncpus", "dst" : "ncpus", "type" : "s64" },
              { "src" : "data", "rec_member" : "job_id", "dst" : "job_id", "type" : "s64" },
              { "src" : "data", "rec_member" : "gid", "dst" : "gid", "type" : "s64" },
              { "src" : "data", "rec_member" : "nnodes", "dst" : "nnodes", "type" : "s64" }
,
              { "src" : "event", "dst" : "event", "type" : "char[]", "array_len" : 255 },
              { "src" : "timestamp", "dst" : "timestamp", "type" : "s64" },
              { "src" : "schema", "dst" : "schema", "type" : "char[]", "array_len" : 255 }
          ]
        }
      ],
      "indices" : [
      ]
    },
     "slurm_step_data_decomp" : {
... snip ...
     },
  },
  "digest" : {
   "AD6D601CB7089CB18BE36D7C993B1D0865C8A4BC53B2F01B673C22F0EC88E452" : "slurm_task_data_decomp",
    "F655F30DE2E43104080D8B1262A536F1D9BF5567E701519D6ED341C38E96F393" : "slurm_job_data_decomp",
    "F8853EF57772A98F70DEF3FE90A628BB0DC882998AC10E2982510A85C61DA7D1" : "slurm_step_data_decomp"
  }
}

[tom95858]

@morrone, @narategithub sorry forgot the command syntax...

ldms_ls -h localhost -p 20001 -a munge -d

The command honors all the set filter syntax so, if you only want a single set/schema, you could do this:

ldms_ls -h localhost -p 20001 -a munge -d orion-01/meminfo

... or all the schema on a particular node ...

ldms_ls -h localhost -p 20001 -a munge -d -E orion-01/.*

[morrone]

Looks good to me from what I can see here. I probably won't have a chance to try it until some time next week, but I'll try to fit it in then.

[narategithub]

@tom95858 I tested this and found 2 issues:

1. the 'timestamp' is missing from the decomposition dump, and
2. the "indices" attribute was misplaced. It supposed to be a member of a row object.

I'm sending you a patch in another email.