Add missing mergeStats in garbage collector and cache writer threads (fix for #8391)

[Noremos]

No description provided.

[Noremos]

This is a fix for #8391

[hvlad]

Just to be sure: how this patch affects performance ? Did you compare restore time, for example ?

[Noremos]

Just to be sure: how this patch affects performance ? Did you compare restore time, for example ?

I didn't measure it, to be honest. The changes affect background threads, so I didn’t think that it would somehow decrease noticeably. I will measure the performance as soon as I have the opportunity

[Noremos]

Just to be sure: how this patch affects performance ? Did you compare restore time, for example ?

I have tested BR on a 1 GB database. It doesn't seem like there's any noticeable difference.

--- without fix:
time ./gbak -user SYSDBA -b soruce.fdb mybd_m.fbk
real    0m14,806s
user    0m12,665s
sys     0m0,951s

time ./gbak -user SYSDBA -c mybd_m.fbk ./mydb_m.fdb
real    0m18,429s
user    0m18,633s
sys     0m0,891s

--- with fix:
time ./gbak -user SYSDBA -b soruce.fdb mybd.fbk

real    0m14,550s
user    0m12,526s
sys     0m0,979s
time ./gbak -user SYSDBA -c mybd.fbk ./mydb.fdb

real    0m18,024s
user    0m18,108s
sys     0m0,919s


ll -h ./mydb.fdb
-rw-rw---- 1 art art 1,1G янв 20 11:09 ./mydb.fdb
ll -h ./mybd.fbk
-rw-rw-r-- 1 art art 865M янв 20 11:09 ./mybd.fbk


--without fix without -par 8
-- (run 3 times)
real    0m8,572s
user    0m23,811s
sys     0m1,782s

real    0m7,667s
user    0m23,901s
sys     0m1,623s

real    0m8,675s
user    0m23,499s
sys     0m1,819s

--with fix with -par 8
-- (run 3 times)

real    0m7,495s
user    0m22,909s
sys     0m1,618s


real    0m7,982s
user    0m22,490s
sys     0m1,753s

real    0m7,337s
user    0m24,557s
sys     0m1,454s

[dyemanov]

What versions should it be backported into? Only v5 or older versions are also affected?

[Noremos]

What versions should it be backported into? Only v5 or older versions are also affected?

Yes, the bug was discovered in version 3.0.