improve performance for keys with expiration time

[judeng]

This change only affects keys with expiry time.
For SETEX, the average improvement is 5%, and for GET with expiation key, we gain a improvement of 13%.

When keys have expiration time, Redis has an assertion to look up the main dict every time when it touches the expires.
This comes with a performance const, especially during rehash. the damage will be double.

It looks like that assert was added some ten years old, maybe out of paranoia, and there's probably no reason to keep it at that cost.

[judeng]

here benchmark result with duplicate 10 times, for setex, the average improvement is 5%, and for get command, we gain a improvement of 13%

command：

memtier_benchmark -s 127.0.0.1 -p 7000 -t 5 -c 10  --hide-histogram  --key-maximum=100000 --command="setex __key__ 10000 __data__" --test-time=60  --pipeline=50
memtier_benchmark -s 127.0.0.1 -p 7000 -t 5 -c 10  --hide-histogram  --key-maximum=100000 --command="get __key__" --test-time=60  --pipeline=50

results:

setex	this thread	519482.8	511375.7	502968	514473.1	510386.9	510380.8	506815.5	520807.6	516531.3	516756.6	5129978
setex	unstable	457644.7	493901.4	485095.7	483967.4	492119.1	485894.5	477765.4	483789.1	466823.1	488208	4815208
get	this thread	853503	864664.4	856952.4	822671.4	856694.9	839800.9	860599.7	846762.3	873183.1	847229.9	8522062
get	unstable	755178.8	749558.7	760419.1	765058.5	726172	785763.1	703592.7	769812	763690.4	768107	7547352

[judeng]

the ci faied with message "Expected '29375' to be less than '29236' (context: type eval line 17 cmd {assert_lessthan $el_sum2 [expr $el_sum1+5000] } proc ::test)", It looks like same with #12169, @CharlesChen888 hi, could you have a look? thank you

[CharlesChen888]

That is a new test for some new metrics in INFO STATS, see #11963. I already allowed a big tolerance in the test, but I guess I need to make it even bigger...

[oranagra]

i think i'd rather just drop the assertion if it causes so much damage.
maybe it was needed during development time, but probably not needed anymore.
the complication of adding a server variable, and a debug command doesn't seem worth it.
please include some statement about the measured improvement in the top comment.

[judeng]

i think i'd rather just drop the assertion if it causes so much damage.

Totally agree. I just don't know why this assert was added so had to be conservative :-)