CERN Accelerating science

Zero size files can't be read from fuse mount


(Franck Eyraud) #1

Hello,

I wanted to report on JIRA the issue we met today on Citrine, but it seems I don’t have the correct access any more, gives an error, so I post it here.

The following behaviour breaks the functioning of some programs (the file is non-existent at the beginning), and is clearly incorrect. It can easily be reproduced this way :

/eos/path/to/file $ touch test-file
/eos/path/to/file $ cat test-file
cat: test-file: Numerical result out of range
/eos/path/to/file $ truncate -s0 test-file
/eos/path/to/file $ cat test-file
/eos/path/to/file $
/eos/path/to/file $ cp test-file test-file-new
/eos/path/to/file $ cat test-file-new
cat: test-file-new: Numerical result out of range

So even copying a readable empty file creates an unreadable one.

The reason might be because the replicas are not present on the FSTs :

Correct file :

$ eos file check /eos/path/to/file/test-file
path="/eos/path/to/file/test-file" fid="0001a762" size="0" nrep="2" checksumtype="adler" checksum="0000000100000000000000000000000000000000"
nrep="00" fsid="5" host="s-jrciprcids93v.cidsn.jrc.it:1095" fstpath="/data02/0000000a/0001a762" size="0" statsize="0" checksum="0000000100000000000000000000000000000000"
nrep="01" fsid="2" host="s-jrciprcids92v.cidsn.jrc.it:1095" fstpath="/data02/0000000a/0001a762" size="0" statsize="0" checksum="0000000100000000000000000000000000000000"

Incorrect file :

eos file check /eos/path/to/file/test-file-new
path="/eos/path/to/file/test-file-new" fid="0001a763" size="0" nrep="2" checksumtype="adler" checksum="0000000100000000000000000000000000000000"
nrep="00" fsid="6" host="s-jrciprcids93v.cidsn.jrc.it:1095" fstpath="/data03/0000000a/0001a763" size="0" statsize="18446744073709551615" checksum="0000000100000000000000000000000000000000"
nrep="01" fsid="3" host="s-jrciprcids92v.cidsn.jrc.it:1095" fstpath="/data03/0000000a/0001a763" size="0" statsize="18446744073709551615" checksum="0000000100000000000000000000000000000000"

Observed with MGM 4.2.12 and FST 4.2.18, and reproduced with an instance freshly fully upgraded to 4.2.19


(Maria Arsuaga Rios) #2

Hi Frank,

as it was mentioned in EOS Communication Workflows, now you can just send an email to eos-support@cern.ch reporting the issue and requesting the issue to be registered in JIRA with the corresponding description. Then, the EOS team will create it accordingly and follow up.

I hope it helps,
Maria


(Franck Eyraud) #3

Dear Maria,

You are right, thank you for reminding this, it sure helps! I will do as foreseen. Feel free to remove this post, if needed.

Franck


(Yaodong Cheng) #4

Hi Frank,

Did you resolve the problem? We also observed it with EOS server 4.2.16. If you have fixed it, please help us.

Thank,
yaodong@IHEP


(Veselin Vasilev) #5

Hi Yaodong,

That was registered as a bug, and fixed later on. We got it fixed in JRC by upgrading our MGMs and FSTs to the 4.2.20 version.

Greetings,
Veselin


(Yaodong Cheng) #6

Hi Veselin,

Yes, I tested it in Version 4.2.22, and the bug was fixed.

Thanks,
Yaodong