So this is a simple one, if I run the following code on Ubuntu 18.04 (4.15.0-29-generic ) and against Ubuntu 16.04 ( 4.4.0-87-generic) I get two different results ; The question is why? It seems that fsync may be broken in the latest kernel and may well be broken as early as 4.10 ? This has huge implications around performance sensitive applications like mysql. (Yes I know it can be turned off in mysql but thats only any good if you like your databases corrupted).


4.4.0-87-generic - # time ./fsync
real 0m7.537s <-----
user 0m0.000s
sys 0m1.060s <-----


4.15.0-29-generic - # time ./fsync
real 1m38.299s <-----
user 0m0.013s
sys 0m0.893s <-----


Source stolen from else where. ;-)

#include <stdio.h>
#include <fcntl.h>
#include <time.h>
#include <unistd.h>


void withSync() {
int f = open( "/tmp/t8" , O_RDWR | O_CREAT );
lseek (f, 0, SEEK_SET );
int records = 10*1000;
clock_t ustart = clock();
for(int i = 0; i < records; i++) {
write(f, "012345678901234567890123456789" , 30);
fsync(f);
}
clock_t uend = clock();
close (f);
printf(" sync() seconds:%lf writes per second:%lf\n", ((double)(uend-ustart))/(CLOCKS_PER_SEC), ((double)records)/((double)(uend-ustart))/(CLOCKS_PER_SEC));
}


void withoutSync() {
int f = open( "/tmp/t10" , O_RDWR | O_CREAT );
lseek (f, 0, SEEK_SET );
int records = 10*1000;
clock_t ustart = clock();
for(int i = 0; i < records; i++) {
write(f, "012345678901234567890123456789" , 30 );
}
clock_t uend = clock();
close (f);
printf("no sync() seconds:%lf writes per second:%lf \n", ((double)(uend-ustart))/(CLOCKS_PER_SEC), ((double)records)/((double)(uend-ustart))/(CLOCKS_PER_SEC));
}


int main(int argc, const char * argv[])
{
/* withoutSync();*/
withSync();
return 0;
}

You haven't said what filesystem or what mount flags you are using. Specifically barrier.

So this is using ext4 with the following mount options rw,noatime,nodiratime,nobarrier,data=ordered