[BUGFIX] Art: fix dm_only frontend

[cphyc]

PR Summary

Removes tests that are intended for particle datasets. They were incorrectly run on ART, an octree dataset.

[matthewturk]

Looks good to me. I think @munkm pointed out somewhere that there's a different DM-only ART sim we could test.

[munkm]

Yes I did in the slack. i'll update the test. It's the DMonly dataset.

[Xarthisius]

Looks like DMonly dataset is not recognized as a valid ART file

[cphyc]

Note: cbd1a68 and b5cfe9f are merely an attempt at making things work ; I am not entirely sure about them, so please feel free to overwrite them.

[Xarthisius]

I had to kill the test run. It was stuck in some sort of an infinite loop and consuming 108GB of RAM...
Strace showed only this:

openat(AT_FDCWD, "/mnt/yt/DMonly/PMcrs0.0100.DAT", O_RDONLY|O_CLOEXEC) = 15
fstat(15, {st_mode=S_IFREG|0664, st_size=402653184, ...}) = 0
ioctl(15, SNDCTL_TMR_TIMEBASE or SNDRV_TIMER_IOCTL_NEXT_DEVICE or TCGETS, 0x7ffd320aaf50) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(15, 0, SEEK_CUR)                  = 0
lseek(15, 0, SEEK_END)                  = 402653184
lseek(15, 0, SEEK_CUR)                  = 402653184
close(15)

[cphyc]

I had to kill the test run. It was stuck in some sort of an infinite loop and consuming 108GB of RAM...
Strace showed only this:

openat(AT_FDCWD, "/mnt/yt/DMonly/PMcrs0.0100.DAT", O_RDONLY|O_CLOEXEC) = 15
fstat(15, {st_mode=S_IFREG|0664, st_size=402653184, ...}) = 0
ioctl(15, SNDCTL_TMR_TIMEBASE or SNDRV_TIMER_IOCTL_NEXT_DEVICE or TCGETS, 0x7ffd320aaf50) = -1 ENOTTY (Inappropriate ioctl for device)
lseek(15, 0, SEEK_CUR)                  = 0
lseek(15, 0, SEEK_END)                  = 402653184
lseek(15, 0, SEEK_CUR)                  = 402653184
close(15)

The tests should now pass, but if anyone has some experience with ART datasets, that would be valuable as I am unsure about the correctness of my fixes.

[cphyc]

@yt-fido test this please

[matthewturk]

So I think this is part of a bigger problem, which is that not all of the frontends have had their range arguments implemented. But as long as this works, it's fine. In #2658 I tried to make this more formal, by making it obvious, but that one's not yet in.

[cphyc]

Agreed. But even with #2658 in, we'd still have the problem that particle frontends with no implementation of chunking will fail, even with CHUNKSIZE=∞

[matthewturk]

Good point. All this kind of makes me hope we can figure out a better chunking solution soon.

…

On Mon, Jun 29, 2020 at 12:23 PM Corentin Cadiou ***@***.***> wrote: Agreed. But even with #2658 <#2658> in, we'd still have the problem that particle frontends with no implementation of chunking will fail, even with CHUNKSIZE=∞ — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2688 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAVXO3Q6A2HSGX2KMHL6HDRZDEXVANCNFSM4OIOQXKQ> .