SamuelBoerlin commented on issue #3488:
URL: https://github.com/apache/jena/issues/3488#issuecomment-3384658453

   > Do you happen to know what the CONSTRUCT query is?
   Have other CONSTRUCT queries succeeded?
   I think this is different problem but made likely to happen around the time 
of compaction finishing. It may (I'm still investigating) leave the system in a 
broken state but something triggered it in the first place.
   
   Unfortunately not. But queries in general work fine and don't seem to be 
affected by the compaction getting stuck.
   The DB has been in this state for days several times now and we haven't 
observed any problems other than more and more disk space getting used up and 
memory usage not going down (usually it drops quite a lot after a successful 
compaction).
   
   > It's inside the JVM to get a `java.lang.InternalError` - it is accessing a 
memory mapped file.
   Would it be possible to run without `--add-modules=jdk.incubator.vector` and 
with a 2G heap `-Xmx2G`?
   
   Ah sorry, I had copied the wrong line. We are already running with `-Xmx2G`:
   
   ```
   /opt/java/openjdk/bin/java -Xmx2G -Dlog4j2.formatMsgNoLookups=true 
--enable-native-access=ALL-UNNAMED --add-modules=jdk.incubator.vector 
-Dlog4j.configurationFile=/jena-fuseki/log4j2.properties -cp 
/jena-fuseki/fuseki-server.jar org.apache.jena.fuseki.main.cmds.FusekiServerCmd
   ```
   
   The `--add-modules=jdk.incubator.vector` was only added when we upgraded 
from 5.2.0 to 5.5.0.
   This problem was also happening already before that (though less frequently) 
on 5.2.0 and without `--add-modules=jdk.incubator.vector`, so I don't think it 
is the root cause.
   I'll give it a try, but it'll take a while.
   
   Some additional points that are interesting:
   - I tried reproducing the problem by running the compaction script in a loop 
for quite a while on a less busy server with the same data and was unable to 
reproduce it this way. So it seems that it might only be triggered when other 
stuff is happening at the same time as compaction.
   - Starting a backup via `GET http://db:3030/dsp-repo` with `Accept: 
application/trig` ca. half an hour after starting the compaction and before it 
has finished, led to compaction getting stuck and the .trig file being 
truncated at some random point and being smaller than expected (and yet no 
error was detected by `curl --fail ...`). This led us to change the schedules 
such that backups and compaction don't overlap, which seems to have solved the 
occasional corrupted/truncated backups, but didn't solve the compaction getting 
stuck.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to