Conversation
|
The second patch is not needed since there is no change to the ioengine ops struct. |
Oh, ok, I've understood something wrong, so I reverted commit |
|
Isn't it normal that CI didn't work for Windows and Macos? |
We do tend to see the odd failure but this macOS error looks to be a permanent build error: (from https://github.com/axboe/fio/actions/runs/18996147708/job/54256207771?pr=2009 ). Windows also looks like a build failure: (from https://github.com/axboe/fio/actions/runs/18996147708/job/54256207773?pr=2009#step:8:183 ). |
|
If you're introducing a platform specific engine you will have to add a configure probe to stop it being built on platforms that are not supported. See what |
|
What @sitsofe said, but also please sanitize your git history and commit messages too. Your commit message does not follow the required format (signed-off-by, for example), and it should also have an actual description of why this commit exists. All it has is "add xattr ioengines" which is pretty obvious. Explain what the engines are, and what they do. Basically duplicate some of the manual additions you have. And don't add a commit you don't need (the version bump) and then add a revert of that. I don't want useless commits like that in the git history. Just rebase your branch, dropping those two patches, and then force push a new one for the PR. Outside of that, I do think the engine makes sense, and it mostly looks good. There are some useless additions like variables that you don't really need. Things like |
290d2b1 to
61b483a
Compare
The patch introduce ioengines that allow to measure extended attributes access: filesetxattr for xattrs value setting, filegetxattr - for lookup. Signed-off-by: Sergei Truschev <s.truschev@yadro.com>
|
Fixed compilation errors, git history and commit message |
|
Hello! Did you have a time to look my PR? |
|
Please replace all the ifdef for WIN32 with a proper CONFIG* flag, as generated by either configure, or in this case just set for windows. Also please look at reducing the number of them in the first place. Having them littered throughout the code makes for a fragile and hard to maintain code base. There are numerous examples in the code base for how to do that, either with stubs that return an error for the non-CONFIG_FOO case or similar. |
Hello!
I've prepared support of xattr operations measurement into fio. The only thing, I think, is that multiple operations in one IOPS doesn't quite fit the fio philosophy. I would like to know your opinion on this matter.