[Ready-to-review] Adding caching to PyPerfNativeStackTrace

[IzabellaRaulin]

Resolving the issue #379

• in-progress: Updating the performance results for the latest commit

1. Summary of changes

• Upgrade the PyPerf to the version containing cache implementation (see Added caching in PyperfNativeStack Granulate/bcc#36)

2. Results [this section will be updated soon]

Methodology:

• 10 python applications (doing while 1: pass) were running for all this time
• Running dockerized gprofiler in DWARF mode and set profiling frequency to 99 Hz, see command below:

  docker run --network host --name <name> --restart=always -d --pid=host --userns=host -v 
  /var/run/docker.sock:/var/run/docker.sock --privileged <my-image-name:tag> -cu --token=<my_token> --service-name= 
  <my_service_name>" -f 99 --perf-mode dwarf
  

• Data uploaded to services (7th Oct' 22):
  • gprofiler-official-1.6.0 (collection_start: 5:50 PM CET, collection_end: 6:20 PM CET)
  • gprofiler-caching-revision-1.2 (collection_start: 6:30 PM CET, collection_end: 7:00 PM CET)

For granulate/gprofiler:1.6.0
_Ux86_64_get_proc_name	Total time 3.52% Own time: 0.0% Samples: 19,700
ebpf::pyperf::NativeStackTrace::NativeStackTrace	Total time 4.43% Own time: 0.0% Samples: 24,302
ebpf::pyperf::NativeStackTrace::cache_eviction	N/A

For pyPerfNativeStack with enabled caching
_Ux86_64_get_proc_name	Total time 3.15% Own time: 0.0% Samples: 15,108
ebpf::pyperf::NativeStackTrace::NativeStackTrace	Total time 4.1% Own time: 0.0% Samples: 21,357
ebpf::pyperf::NativeStackTrace::cache_eviction	Total time 0.13% Own time: 0.% Samples: 727

3. Screenshot for the scenario with official docker image (granulate/gprofiler:1.6.0)

3.1. _Ux86_64_get_proc_name matches

3.2. _Ux86_64_get_proc_name matches for filtered Pyperf process

4. Screenshots for the scenario with enabled caching in PyPerfNativeStack

4.1. _Ux86_64_get_proc_name matches

4.2. _Ux86_64_get_proc_name matches for filtered Pyperf process

4.3. cache_exiction matches

4.4. cache_exiction matches for filtered Pyperf process

5. Comparison

Before (docker image granulate/gprofiler:1.6.0):

After enabling caching

[Jongy]

@IzabellaRaulin , the comparison methology looks good.

Going forward - I put a few more comments on the BCC PR. Then we can merge it.
Before we merge this PR, I'll get #478 merged so that PyPerf & native unwinding is tested again against our full matrix. After thart we're good to go :) The Python tests matrix PR will be merged this week.

[Jongy]

@IzabellaRaulin please merge from master as I have merged #478 which adds more tests for PyPerf, I want your changes to be tested by them as well.