Repository navigation
Description
In attempting to reproduce something else on demo.dataverse.org, I ran into intermittent publishing failures from the DataCite Test MDS API:
java.io.IOException: Response from postMetadata: 500, <!doctype html>
<h3>Error ID:</h3>
<span class=“error”>f508a192</span>
<h3>Details:</h3>
<p>Web application could not be started by the Phusion Passenger(R) application server.</p>
<p class=“bold”>Please read <a href=“https://www.phusionpassenger.com/library/admin/log_file/” class=“plain”>the Passenger log file</a> (search for the Error ID) to find the details of the error.</p>
<p>You can also get a detailed report to appear directly on this page, but for security reasons it is only provided if Phusion Passenger(R) is run with <i>environment</i> set to <i>development</i> and/or with the <i>friendly error pages</i> option set to <i>on</i>.</p>
<p>For more information about configuring environment and friendly error pages, see:</p>
<ul>
<li><a href=“https://www.phusionpassenger.com/library/config/nginx/reference/#passenger_friendly_error_pages”>Nginx integration mode</a></li>
<li><a href=“https://www.phusionpassenger.com/library/config/apache/reference/#passengerfriendlyerrorpages”>Apache integration mode</a></li>
<li><a href=“https://www.phusionpassenger.com/library/config/standalone/reference/#--friendly-error-pages---no-friendly-error-pages-friendly_error_pages”>Standalone mode</a></li>
</ul>
Payara logs: Identifier not created: create failed but the database says identifierregistered = 't':
# select * from dvobject where identifier like '%FT4YBA%';
id | dtype | createdate | indextime | modificationtime | permissionindextime | permissionmodificationtime | publicationdate | creator_id | owner_id | releaseuser
_id | previewimageavailable | storageidentifier | authority | globalidcreatetime | identifierregistered | identifier | protocol | previewimagefail | separator
---------+----------+-------------------------+-------------------------+-------------------------+-------------------------+----------------------------+-------------------------+------------+----------+------------
----+-----------------------+-----------------------------------------+-----------+-------------------------+----------------------+-------------------+----------+------------------+-----------
2737782 | Dataset | 2026-11-11 13:40:28.852 | 2026-11-11 13:41:01.405 | 2026-11-11 13:41:01.199 | 2026-11-11 13:41:01.359 | 2026-11-11 13:40:28.852 | 2026-11-11 13:40:38.323 | 361 | 2737774 |
361 | f | s3dir://10.70122/FK2/FT4YBA | 10.70122 | 2026-11-11 13:40:41.52 | t | FK2/FT4YBA | doi | f | /
Dataverse also logs that it's unlocking the dataset, but 24+ hours later:

As of yesterday evening there were 91 datasetlocks on demo.dataverse.org, all with reason Workflow
Leonid could produce the Phusion Passenger response from the DataCite Test MDS API with
$ curl -X POST "https://mds.datacite.org/metadata"
I'm only concerned with Dataverse thinking the identifier is registered when it isn't and leaving the dataset locked; each of those conditions will require corrective API calls from an archivist / administrator.
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsNo status
Activity
wshahn commentedon Nov 13, 2026
If I curl from api.test.datacite, I can see the dataset got registered. Because the state is findable below, it seems like
identifierregistered = 't'is correct and something is wrong with the dataset lock workflow here.landreev commentedon Nov 19, 2026
@qqmyers That specific dataset may be registered, but the real issue is that DataCite's test/dev service (https://mds.test.datacite.org) was genuinely dead on that day (500 in response to all/any requests); and Dataverse does not appear to be able to handle that very well - each such failure appears to leave a lock behind.
This keeps happening apparently - not seeing any of these errors in server.log today; but they kept popping up approx. every other day since Don noticed the above.
As Don mentioned above, when the dev. mds service is in this state, a
% curl -X POST "https://mds.datacite.org/metadata"results in a 500 and the html message quoted. When the service is working, this empty POST results in some meaningful message about missing or invalid arguments.
Last time I observed this in r/t, I can't remember if I tried the above from outside of demo (like my own macbook); to eliminate the possibility that it was demo specifically being placed on the naughty list. But the main issue remains regardless - the app does not handle this error gracefully, whatever the reason.
landreev commentedon Nov 19, 2026
I may know what it is. The key detail in Don's report is that all these locks are of type Workflow.
If an exception is thrown during the DataCite exchange, this is how it's caught in the FinalizePublication command:
... but not
DatasetLock.Reason.Workflow- ?qqmyers commentedon Nov 19, 2026
That makes sense - this line removes the workflow lock, but never called if you throw up top due to DataCIte:
dataverse/src/main/java/edu/harvard/iq/dataverse/engine/command/impl/FinalizeDatasetPublicationCommand.java
Line 251 in e6d7a10
Maybe a try/finally since I don't think DataCite is the only thing that can throw a CommandException (e.g. validateFiles)
landreev commentedon Nov 19, 2026
And also, early on in the command we attempt to register the identifier for the dataset (and possibly files) if it has not been registered yet; only the finalizePublication lock is removed there as well if that throws an exception.
So yes a try/finally around the whole thing.
donsizemore commentedon Nov 20, 2026
potentially related: #7395 and #10992
landreev commentedon Nov 20, 2026
The older one is almost certainly about the dedicated
finalizePublicationlock proper, that we must have made Dataverse clean up properly since. This issue is specific toworkflowlocks (it became a problem on demo due to a combo of having a workflow added instance-wide and the test mds service instability). But we should use the opportunity to close it as well.Not sure about #10992, it looks like there's a lot going on there. At least some of it appears to be on account of the checksum validation of a giant file genuinely taking a long time.
landreev commentedon Nov 20, 2026
Something like this probably: cf15cf7
That said, it took more changed lines than I expected to try and have these locks deleted in just one place, with try/finally.
Wondering if it's actually going to be simpler and safer to try to remove that workflow lock in 3 separate
catch()statements.