mempool: fix lack of pool names in mempool:dump output for JSON forma…

[ifed01]

…tter.

Signed-off-by: Igor Fedotov ifed@mail.ru

[tchaikov]

@ifed01 i think we could add a dummy section here, like

f->open_object_section("mempool");
f->open_array_section("pools");
for (...) {
  ...
}
f->close_section();
f->dump_object("total", total);
f->close_section();

as the pools are of the same type (pool_t), while total is a stats_t. what do you think?

[ifed01]

makes sense. will do

[ifed01]

@tchaikov - unfortunately it looks like top level blocks inside an array can't have names as well - hence we either need yet another dummy nested level or should print pool names as a field within a block not as a block name. Consider:

{
"pools" : [
   {
     "pool1" : {
        "a":"1"
      }
   },
   {
     "pool2" : {
       "a":"1"
     }
   }
]
}
vs.
{
"pools" : [
 {
   "name" : "pool2"
   "a":"1"
 },
 {
   "name" : "pool2"
   "a":"1"
 }
]
}

Not sure adding array section worths such changes... Please note that these 'dummy' block names will be visible in other formats hence wasting the output...

[tchaikov]

no, elements in an array does not have their names. so, right, the output would look like the latter.

IMHO, the problem of mempool::dump() are two folded:

1. no top level data structure. that's what your change is addressing
2. "total" is not a pool, but it is printed in line with other pools.

[ifed01]

@tchaikov: made an object section for pools instead of array one to avoid additional nested level. Please take a look.

[ifed01]

@tchaikov: ping

[tchaikov]

http://pulpito.ceph.com/kchai-2017-10-31_13:59:11-rados-wip-kefu-testing-2017-10-31-1919-distro-basic-mira/

failed test tracked at http://tracker.ceph.com/issues/21993