Re: [dpdk-dev] [PATCH v5] lib/metrics: add unregister api for metrics

[Remy Horton <remy.horton@intel.com>]

The tests for metrics themselves now all pass, but there is failure with 
bitrate statistics (bitrate stats uses metrics underneath). Latency 
statistics, which also uses metrics, seems ok though.

RTE>>bitratestats_autotest
  + ------------------------------------------------------- +
  + Test Suite : BitRate Stats Unit Test Suite
port in ring setup : 0
  + ------------------------------------------------------- +
  + TestCase [ 0] : test_stats_bitrate_create succeeded
  + TestCase [ 1] : test_stats_bitrate_reg failed
  + TestCase [ 2] : test_stats_bitrate_reg_invalidpointer succeeded
  + TestCase [ 3] : test_stats_bitrate_calc_invalid_bitrate_data succeeded
Invalid port_id=33
  + TestCase [ 4] : test_stats_bitrate_calc_invalid_portid_1 succeeded
Invalid port_id=65535
  + TestCase [ 5] : test_stats_bitrate_calc_invalid_portid_2 succeeded
Invalid port_id=31
  + TestCase [ 6] : test_stats_bitrate_calc_non_existing_portid succeeded
  + TestCase [ 7] : test_stats_bitrate_calc succeeded
  + ------------------------------------------------------- +
  + Test Suite Summary
  + Tests Total :        8
  + Tests Skipped :      0
  + Tests Executed :     8
  + Tests Unsupported:   0
  + Tests Passed :       7
  + Tests Failed :       1
  + ------------------------------------------------------- +
Test Failed


Personally if I was reimplementing metrics I would be inclined to 
replace the single fixed-size metadata[] array that is carved up with 
something more dynamic, perhaps a 2d linked-list. Doing so would 
probably negate the need for a bitmap.



On 27/02/2019 17:19, Junjie Wan wrote:
> From: junka <wan.junjie@foxmail.com>
>
> The bitmap will help maintain the metrics. We can dynamically
> add and remove metrics data. For example, after uninit latency lib,
> it could remove itself from the metrics. This could make the result
> from rte_metrics_get_names much more simple to display the wanted
> metrics data only.
>
> Signed-off-by: Junjie Wan <wan.junjie@foxmail.com>