Checkbox remote master cost over 5min during Bootstrap after test plan selected

[Ray Chen]

While using checkbox-cli master to connect slave at first time, after selected test plan, the master cost over 5 min to connected success and continue to tests selection

Step:

1. Run Checkbox-cli slave on DUT
2. Run checkbox-cli master <IP_ADDER>
3. Select test plan on master
4. Master start bootstrap as below

============[ Bootstrap com.canonical.certification::device (1/3) ]=============
==========[ Bootstrap com.canonical.certification::executable (2/3) ]===========
=============[ Bootstrap com.canonical.certification::fwts (3/3) ]==============

and it will cost over 5 min to tests selection(sometime over 10 min)

=========================================
~$ apt-cache policy plainbox checkbox
plainbox:
  Installed: (none)
  Candidate: 0.39.0~ppa~ubuntu16.04.1
  Version table:
     0.39.0~ppa~ubuntu16.04.1 500
        500 http://ppa.launchpad.net/hardware-certification/public/ubuntu xenial/main amd64 Packages
        500 http://ppa.launchpad.net/hardware-certification/public/ubuntu xenial/main i386 Packages
     0.25-1 500
        500 http://archive.ubuntu.com/ubuntu xenial/universe amd64 Packages
        500 http://archive.ubuntu.com/ubuntu xenial/universe i386 Packages
checkbox:
  Installed: (none)
  Candidate: 1.2.4-0ubuntu1
  Version table:
     1.2.4-0ubuntu1 500
        500 http://archive.ubuntu.com/ubuntu xenial/universe amd64 Packages
        500 http://archive.ubuntu.com/ubuntu xenial/universe i386 Packages

[Maciej Kisielewski (kissiel)]

This got fixed by a few optimization I and Sylvain did.
Until proven otherwise I consider this to be fixed.

[Sylvain Pineau (sylvain-pineau)]

Reproduced!

Slave is running dev ppa versions of checkbox-ng and p-p-c. I've cloned p-p-tpm2 on the slave and manually `manage.py develop` it.

From master it takes 70s to see the test selection screen (one of the tpm2 test plan):

#############################

[launcher]
app_id = com.canonical.certification:checkbox-test
launcher_version = 1
stock_reports = text, submission_files

[test plan]
unit = com.canonical.certification::tpm2.0_4.1.1-automated
forced = yes

[test selection]
forced = no

#############################

A bit of debug at rpyc level shows a huge traffic coming from the slave with job details inside the data.

In order to benefit from 1. stream compression and 2. no netrefs we absolutely have to prefer very simple data model when sending structures of data to master, in other words, list and strings.

For everything else, do json serialization. Keeping Netrefs (https://rpyc.readthedocs.io/en/latest/docs/theory.html#object-proxying) on all known units costs a lot.

[Pierre Equoy (pieq)]

Tested using checkbox-snappy 20.15 rev2035 from latest/beta channel and the launcher described in comment 2.

The test plan screen is shown within 2 seconds, and I can execute it without any problem.

-> tested OK!