Hitting some troubles with moving existing envs to s3fs. The gameplan:
Step 6 is interrupted unfortunately, because I'm seeing IOErrors when opening and uploading files:
[#|2023-10-11T14:35:32.161+0200|WARNING|Payara 5.2022.3|edu.harvard.iq.dataverse.dataaccess.ImageThumbConverter|_ThreadID=95;_ThreadName=http-thread-pool::jk-connector(1);_TimeMillis=1697027732161;_LevelValue=900;|
caught IOException trying to open an input stream for file://18b13f045f2-ef595f3a5aa6java.io.IOException: Failed to open local file file:///data/dataverse/files/10.5072/TAR/Y05ZDI/18b13f045f2-ef595f3a5aa6|#]
[#|2023-10-11T14:35:45.004+0200|SEVERE|Payara 5.2022.3|edu.harvard.iq.dataverse.util.FileUtil|_ThreadID=98;_ThreadName=http-thread-pool::jk-connector(4);_TimeMillis=1697027745004;_LevelValue=1000;|
Failed to create filesTempDirectory: /data/dataverse/files/temp|#]
[#|2023-10-11T14:35:45.004+0200|WARNING|Payara 5.2022.3|edu.harvard.iq.dataverse.EditDatafilesPage|_ThreadID=98;_ThreadName=http-thread-pool::jk-connector(4);_TimeMillis=1697027745004;_LevelValue=900;|
Failed to process and/or save the file blank_template_melting_bard.jpg; Temp directory is not configured.|#]
Thoughts? : )
It kind of looks like that the S3 file sytem cannot create a folder for the temp file dir
Is it writable?
Can you actually create a directory in s3fs without any content? (In S3 AFAIK folders are just a concept like we do it in Dataverse, it's metadata of a file object)
I'm not sure about the first one
That could come from different things...
Yeap, can confirm if I just use mkdir on the mount directly.
Not sure how your config and your mounts look like...
Also as user dataverse?
(Or payara, as we might talk containers)
I'm debugging with umask at 000 at present, shouldn't be a permission issue (though can be). It's a thread to pull at.
I am tempted to move this to #troubleshooting ... It's not a dev topic (yet?)
Yes, please move.
This topic was moved here from #dev > S3fs woes by Oliver Bertuch.
Apologies for creating it here in the first place : )
Apologies for starting in dev in the first place.
I briefly am about to get yelled at, brb after this.
A message was moved here from #dev > S3fs woes by Oliver Bertuch.
Thomas van Erven said:
I'm debugging with umask at 000 at present, shouldn't be a permission issue (though can be). It's a thread to pull at.
Well what happens in the background is that if you upload a file and the temp dir does not yet exist, it will create it for you before storing the upload for further processing
So if Dataverse cannot create this directory for you, but you can directories on the mounted FS, this certainly indicates some sort of permission problem
Temp directory exists; it booms at creating the file.
Side note; when in a discussion with management, do not say "of course management idea quality is poor, I wouldn't expect anything else from management". :smiling_face_with_tear:
Thomas van Erven said:
Temp directory exists; it booms at creating the file.
That's not what the error message is telling us...? :thinking:
The error sits on a throne of lies:
[#|2023-10-11T16:15:03.480+0200|SEVERE|Payara 5.2022.3|edu.harvard.iq.dataverse.util.FileUtil|_ThreadID=96;_ThreadName=http-thread-pool::jk-connector(2);_TimeMillis=1697033703480;_LevelValue=1000;|
Failed to create filesTempDirectory: /data/dataverse/files/temp|#]
[#|2023-10-11T16:15:03.480+0200|WARNING|Payara 5.2022.3|edu.harvard.iq.dataverse.EditDatafilesPage|_ThreadID=96;_ThreadName=http-thread-pool::jk-connector(2);_TimeMillis=1697033703480;_LevelValue=900;|
Failed to process and/or save the file DeanSpencer-spotcol-dwarfforge.jpg; Temp directory is not configured.|#]
[root@tar31 temp]# pwd
/data/dataverse/files/temp
However, you gave me an excellent idea. Ownership.
As concept, not as a file property, you understand )
See that's why I am introducing the startup checks in https://github.com/IQSS/dataverse/pull/9819
I'm glad the yelling was brief! :grinning:
It wasn't permission issue either.
But; I've got a slightly different error, which suggests it might be rights orientated.
You were correct; it works now. Topic may be resolved.
(As a side note; the dataverse file behaviours for temp/draft/final take quite long in terms of the interface that keeps loading, for large files and larger sets of files. I suspect I'm going to consult your migration scripts and use the Keystone implementation directly, rather than a fuse mount).
Thomas van Erven has marked this topic as resolved.
Last updated: Oct 30 2025 at 06:21 UTC