On 20/03/2019 06:37, Walker, Andreas wrote:
Hi Andy,
unfortunately (in terms of reproducing the error, not in terms of my project),
starting with a fresh database seems to have solved the problem. It is well
possible that the previous database was corrupted due to an abort transaction,
and that this only happened once - I assumed that dropping the graph would also
get rid of any corrupted data, but if I understand you correctly, that may not
have been the case, and the error appearing again and again may just have been
a consequence of the initial corruption.
Yes, that is what is happening. RDF terms aren't removed from the node
table on delete unless the database is compacted. Delete graph (no refs
to the RDF Term left) and compact, then it should get rid of the problem
if that's the cause.
That's a big "if" at the moment. I'm guessing that abort might be
involved because it is not a common code path. Equally, it may be
something else.
I'll do
https://issues.apache.org/jira/browse/JENA-1685
which reinforces the code (acts even more cautiously!) but I can see how
any of the changes definitely explain the situation.
Andy
If you still want me to run a test, I'd be happy to, but I guess this may not
be necessary in that case.
Best,
Andreas
________________________________
Von: Andy Seaborne <a...@apache.org>
Gesendet: Mittwoch, 20. März 2019 00:15:48
An: users@jena.apache.org
Betreff: Re: AW: AW: AW: Error 500: No conversion to a Node: <RDF_Term >
Hi Andreas,
Do you have a reproducible test case even if it happens only occasionally?
Having looked at the code, I can't see a risk point and certainly not
like earlier problems with TDB1.
Have you had any abort transactions - these happen in the Fuseki UI only
if you browse away from a file upload or there is a data error.
I can build a special for you with a layer of node write caching removed
if its easier to run an test case in your environment rather than try to
extract one.
Andy
On 11/03/2019 21:59, Andy Seaborne wrote:
Hi Andreas,
On 11/03/2019 14:37, Walker, Andreas wrote:
Hi Andy,
the database was created from the web interface, and I've only used
the web interface to add data to it, so no other version has ever
touched it.
OK - so you have only run v3.10.0.
If I understand you correctly, the problem is with the database as a
whole, so dropping and reloading the graph might not solve the
problem. I have now switched to a fresh database and am currently
reloading the data, so I can see whether the problem persists beyond
the original database.
If it does then we have a reproducible test case. That said, I can
think of a way of a single load or a single load and a sequence of
updates not in a parallel can break the node table.
The email of Osma's is a compaction - have you compacted the database?
(Fuseki must not be running at the time - this is supposed to be caught
using OS file locks but ... I'm told VMs can get this wrong (but I don't
know which and when).
The only backup I have done is by making snapshots of the entire
virtual server this is running on, so I don't think that is related in
any way.
Probably not related but is it an instantaneous backup of the
filesystem? If not, then it isn't a reliable backup (in the same way
that copying all the files isn't a safe backup procedure).
The problem is that if the copy is done while a write transaction is
running, some files may be copied before the commit point and some
after, which risks chaos.
Andy
Thanks again for your help,
Andreas
________________________________
Von: Andy Seaborne <a...@apache.org>
Gesendet: Freitag, 8. März 2019 10:50:28
An: users@jena.apache.org
Betreff: Re: AW: AW: Error 500: No conversion to a Node: <RDF_Term >
Hi Andreas,
Is this a database that has only ever been used with 3.10.0 or was the
data loaded with a previous version at some time in the past?
The problem occurs silently during loading. There is no sign of the
problem at the time and the system works just fine while the RDF term,
or terms, are also in the node table cache.
Then the system is restarted.
Then the RDF term is needed for a query and the errors are reported.
But the problem originated back when the data was loaded or updated, may
be several restarts ago.
Of course, it may be a different issue in which case, but the error
message is consistent with the known bug.
Have you been backing up the server on a regular basis? A backup is
NQuads so it is pulling every RDF term from disk (subject to already
being cached).
Andy
On 07/03/2019 20:47, Walker, Andreas wrote:
Hi Andy,
I am running Version 3.10.0. The problem with reloading the database
is the regular (multiple times a day) recurrence of the problem, so
if there are any strategies to avoid it, I'd appreciate any advice.
Best,
Andreas
________________________________
Von: Andy Seaborne <a...@apache.org>
Gesendet: Donnerstag, 7. März 2019 21:12
An: users@jena.apache.org
Betreff: Re: AW: Error 500: No conversion to a Node: <RDF_Term >
Hi Andreas - which version are you running?
It does not look like the corruption problem, which is now fixed.
The best thing to do is reload the database again. Whatever terms were
messed up are permanently damaged I'm afraid.
Andy
On 07/03/2019 10:49, Walker, Andreas wrote:
Dear all,
as a quick follow-up which might be helpful in identifying the
error; I can currently run a SPARQL query (just listing any triples)
with LIMIT 80, but no higher, before I run into the error, so it
seems like there might indeed be a particular part of the database
that is corrupted.
Best,
Andreas
________________________________
Von: Walker, Andreas <andreas.wal...@sub.uni-goettingen.de>
Gesendet: Mittwoch, 6. März 2019 10:42:32
An: users@jena.apache.org
Betreff: Error 500: No conversion to a Node: <RDF_Term >
Dear all,
from time to time, my Fuseki server starts throwing the following
error message on any SPARQL query I pose to one of my graphs:
"Error 500: No conversion to a Node: <RDF_Term >"
Unfortunately, I couldn't find any explanation of this error
message, beyond a discussion of a corrupted TDB2 database.
(https://users.jena.apache.narkive.com/LF4XE801/corrupted-tdb2-database)<https://users.jena.apache.narkive.com/LF4XE801/corrupted-tdb2-database>
Once this happens, the only thing I could do so far is to drop the
entire afflicted graph and rebuild it, but of course that isn't
going to be a viable solution in the long term.
The only way I interact with Fuseki is by starting and stopping it,
querying it via the SPARQL endpoint (and sometimes through the web
interface, e.g. when troubleshooting my application), and uploading
new triples (as turtle files) via the web interface. I haven't been
able to find a pattern in when the error appears so far.
Any insights into why this error appears, and what to do in order to
avoid it? I'd appreciate any help.
Best,
Andreas
