Vincent Kang Fu
Vincent Kang Fu
There have been two ~~32-bit~~ Windows latency test failures where there is an inconsistency between latency percentiles reported by fio and corresponding percentiles calculated from latency logs: 1. https://ci.appveyor.com/project/vincentkfu/fio/builds/33008676/job/ywsyeaymrb3htht2 2....
Fixing experimental verify revealed a failure on Windows with the newly added t/jobs/t0026.fio. This job is currently disabled on Windows. I'm opening this ticket so that we have a reminder...
The http ioengine no longer builds on newer distributions because it uses deprecated functions. Here is the related `config.log` output: ``` Compiling test case curl-new-ssl gcc -D_GNU_SOURCE -include config-host.h -Werror...
For Windows `pread` and `pwrite` are defined in `posix.c` and currently look like: ``` ssize_t pwrite(int fildes, const void *buf, size_t nbyte, off_t offset) { int64_t pos = _telli64(fildes); ssize_t...
Three failures in the past two months: https://github.com/vincentkfu/fio/actions/runs/14414832872/job/40429681779 https://github.com/vincentkfu/fio/actions/runs/13680647374/job/38251879040 https://github.com/fiotestbot/fio/actions/runs/14914258938/job/41895873567 Related issue: https://github.com/axboe/fio/issues/890
When fio loads an external ioengine it needs to find a `struct ioengine_ops`. It first checks for a symbol matching the ionegine's name and if that fails for an `ioengine`...
A recent [paper](https://www.vldb.org/pvldb/vol18/p4295-haas.pdf) suggests that fio's zipf distribution generator is flawed: I [modified](https://github.com/vincentkfu/fio/tree/zipf) `t/genzipf.c` to add some options to better assess fio's Zipf distribution generator. I added a `-d` option...
Currently fio does a `malloc/free` each time it verifies a block with a `%o` in its `verify_pattern` when `thread=1` or `verify_async > 1`. This is slow and should be improved...