Type::GetByName(): cache results for faster lookup

[Al2Klimov]

Especially since #9543

"Blocked by" (not really, but preferably)

• Separately handle apply rules targetting only specific parent objects #9545

[julianbrost]

I just had a really quick look at this and find it strange doing it this way around.

How much work would it be to make the primary source of types a std::unordered_map<String, Type::Ptr> and make globals.Types a view into this?

[Al2Klimov]

Too much for what this change shall improve.

[Al2Klimov]

Also globals.System.Types is initialised very early and never changes.

[Al2Klimov]

Before/after, time icinga2 daemon -C

f59f361
real 36m38.277s
user 196m31.058s
sys 71m49.837s

real 35m14.356s
user 190m16.353s
sys 68m23.075s

real 35m35.668s
user 192m36.516s
sys 68m44.489s

979a918
real 33m37.158s
user 185m54.210s
sys 61m20.227s

real 32m31.482s
user 182m6.965s
sys 58m28.920s

real 32m7.106s
user 176m20.685s
sys 60m13.250s

In short, I'm proud of this bridge technology.

[Al2Klimov]

Would you prefer if the register macros also fill a second, non-DSL, map (which is preferably accessed unlocked)?

[Al2Klimov]

In short, I'm proud of this bridge technology.

We have caches also e.g. in #9408 after all.

[Al2Klimov]

• 56 cores/threads (IDK)
• 256G RAM
• Config of our customer 22470
• time icinga2 daemon -C

master before #9627

real 54m20,817s
user 337m24,969s
sys 2459m21,103s

master before #9627 + #9553

real 49m3,936s
user 345m6,072s
sys 2208m49,892s

master with #9627

real 23m15,077s
user 1039m6,028s
sys 198m59,482s

master with #9627 + #9553

real 21m52,338s
user 963m19,549s
sys 203m45,879s