thiell

Thanks @luxiaoyong for the report. I am going to try to reproduce your problem and keep you posted. You confirm your 1000+ nodes cannot be folded at all? An example...

Hi @luxiaoyong – glad to hear you found a workaround for now, thanks for sharing. Sorry I have been slow with this until now. But I've made some progress and...

@luxiaoyong ah, that's weird, I see a big difference: without the patch: ``` $ time clush -o-q -w @test uptime ... real 1m59.129s user 1m50.360s sys 0m0.169s ``` with the...

Hi @piste-jp and thanks for your response! Sorry, here are more logs, we can see that LTFS does actually complain about READ_ATTR and LOGSENSE timing out, so it must issue...

And BTW, this is a rare occurrence, we see that maybe only once or twice per week when writing non-stop using 4 LTO-9 drives. Our tape vendor support analyzed the...

See this one for example that I already prepared for our support (file too big for github so I put it on Google Drive): [elm-ent-dm02.fulllogs.txt.gz](https://drive.google.com/file/d/1hAvyXa2rseX8Gjnit780QtOLmfOKwZoc/view?usp=sharing) It was a previous occurrence...

@piste-jp super interesting!! Thank you very much! I guess the LTO-9 drive should likely report a higher timeout value then... I will let our tape support folks know about that!

Hi @piste-jp I got a response from IBM via our support channel. Looks like they are realizing something is wrong... > …Read Attribute command generally does not require tape motion....

@piste-jp Thank you, you are so helpful! I really appreciate it. I think it's a great solution for now: either READ_ATTR will return almost immediately (like most of the time)...

@piste-jp just a quick update, after more than a week running with your patch (slightly adapted as we have FH drives) and archiving literally millions of files, I've **not** seen...