Add first performance options

[phvalguima]

Some of the early conclusions I've found. These options will not use memory maps for files but rather allow "direct" access. The advantage: mysql has its own engines to cache data and does not need the kernel to manage file caches itself.

[taurus-forever]

IMHO, good default as we are deploying DB.
@delgod to decide on the architecture level... should we sit on DB default or tune them as we think we should.
Tnx Pedro!

[paulomach]

lgtm, just remember to bump the libpatch value and fix unit tests

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 65.90%. Comparing base (6bcd952) to head (41f6ea0).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #400   +/-   ##
=======================================
  Coverage   65.90%   65.90%           
=======================================
  Files          17       17           
  Lines        3103     3103           
  Branches      411      411           
=======================================
  Hits         2045     2045           
  Misses        929      929           
  Partials      129      129           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[7annaba3l]

Looks good to me also. Let's make those the defaults and not allow them to be changed.
We will review our auto-tuned in the next 24.10 cycle, hopefully.

[taurus-forever]

Hi,

JFYI, I have made 1 hour tests on localhost for default and proposed parameters (twice).
Env: Laptop=>Multipass QEMU=>LXD.
For the small numbers (like 6K qpt) the difference is not really visible:

I still believe we should go with new defaults based on the previous test results done by Pedro.

P.S. Ignore significant spikes/drops on the localhost test, it looks like some laptop background tasks. No issues notices on charm/qemu itself.