LMDB dumps the entire db if running with debug logging

[nguyenm100]

Current Behavior

looks like lmdb dumps the entire valuestore if debug logging is enabled on startup. this becomes an issue for running against any decently size db.

https://github.com/eclipse-rdf4j/rdf4j/blob/main/core/sail/lmdb/src/main/java/org/eclipse/rdf4j/sail/lmdb/ValueStore.java#L243

Expected Behavior

dump should either be removed or put under TRACE

Steps To Reproduce

1. set logging to DEBUG
2. select from an lmdb store

Version

5.0.3

Are you interested in contributing a solution yourself?

None

Anything else?

No response

[nguyenm100]

@kenwenzel we can make the fix. let us know whether we should remove or put under TRACE (or other option). tx

[kenwenzel]

I think this was introduced by @hmottestad to fix some errors.
I would propose to remove the logging as it is problematic with large dbs.

[nguyenm100]

should we remove the whole if (logger.isDebugEnabled()) {} block?

[kenwenzel]

I'm actually a bit puzzled why the logic for gc is triggered inside the block and not outside before checking the log level. I'll have to take a look If this is intended or just an error.

[kenwenzel]

I propose something alonog the following lines:

// trigger deletion of values marked for GC
startTransaction(true);
commit();

if (LOG_VALUES_ON_STARTUP) {
	// print current values in store
	logValues();
}

The initial transaction needs to be moved outside the if condition to evict unused values on startup.

[nguyenm100]

does it make sense to log the values via an external tool or log just the first X values (x=1000 or something) when logging is on trace? the other thought is to disable debug/trace for this function in logback.xml as the default and anyone that needs it can turn it on there?

though how useful is this function beyond people debugging a specific problem in lmdb thus they could always manually enable it locally.

[kenwenzel]

A good approach could be to just add a main method to the ValueStore that can be used to log the values and completely remove the logging from the constructor. I'm not sure if RDF4J already depends on picocli or something to ease the handling if command line arguments.