Hi, I don't use the gluster gfapi ioengine but this looks like a subtlety with ioengines that set .get_file_size() but don't set .setup(). If you have an ioengine which defines its own .get_file_size(), it has to be prepared for that function to be called BEFORE the call to .init() is done when there is no .setup() (this happens prior to file layout in the scenario where the user doesn't explicitly set size with the fio job). So in your case because add_job() over in init.c does f->real_file_size = -1ULL for diskless ioengines and from get_file_sizes() fio ends up calling fio_gf_get_file_size() which returns 0 (making fio think the file_get_size call worked) and when fio gets to fio_gf_open_file(), f->real_file_size still has the maximum value and the ioengine goes on to extend the file. I guess most are setting the size option when using the gfapi ioengine - notice how it is set in the example gfapi.fio job (https://github.com/axboe/fio/blob/master/examples/gfapi.fio ). On 24 November 2017 at 10:21, qingwei wei <tchengwee@xxxxxxxxx> wrote: > Hi, > > I am currently using glusterfs 3.12.2 and when i use FIO 3.2, the file > is extended automatically to 16E when doing read. > > Steps to produce > > 1. Perform sequential write > sudo fio -group_reporting -ioengine gfapi -volume test -brick > ubuntu24D -fallocate none -direct 1 -nrfiles 1 -filesize 1G -openfiles > 1 -bs 256k -numjobs 1 -iodepth 1 -name test6 -rw write > > 2. Perform sequential read > sudo fio -group_reporting -ioengine gfapi -volume test -brick > ubuntu24D -fallocate none -direct 1 -nrfiles 1 -filesize 1G -openfiles > 1 -bs 256k -numjobs 1 -iodepth 1 -name test6 -rw read > > 3. Mount the gluster volume using FUSE and can see the file size is now 16E > -rw-r--r-- 1 root root 16E Nov 24 18:02 test6.0.0 > > > I also run in debug mode and you can see the file is extended to 16E > > file 12546 fio file test8.0.0 open mode rw td rw read > file 12546 fio extend file test8.0.0 from 1073741824 to 18446744073709551615 > > I also do some printf and i can see that f->real_file_size = > buf.st_size; is never executed and the function return on if (!g || > !g->fs) { > > int fio_gf_get_file_size(struct thread_data *td, struct fio_file *f) > { > struct stat buf; > int ret; > struct gf_data *g = td->io_ops_data; > > dprint(FD_FILE, "get file size %s\n", f->file_name); > > if (!g || !g->fs) { > return 0; > } > if (fio_file_size_known(f)) > return 0; > > ret = glfs_lstat(g->fs, f->file_name, &buf); > if (ret < 0) { > log_err("glfs_lstat failed.\n"); > return ret; > } > > f->real_file_size = buf.st_size; > fio_file_set_size_known(f); > > return 0; > > } > > Anyone can help to see what goes wrong? Thanks. > > Cw > -- > To unsubscribe from this list: send the line "unsubscribe fio" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- Sitsofe | http://sucs.org/~sits/ -- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
