LMDB read handles can remain exhausted after timed-out server queries

[hmottestad]

Problem

When many RDF4J Server queries against an LMDB repository time out, LMDB read handles can accumulate until the environment reaches MDB_READERS_FULL. After that, the repository may keep failing new reads even though only a small number of queries are running concurrently.

Observed stack trace starts at:

java.io.IOException: MDB_READERS_FULL: Environment maxreaders limit reached
	at org.eclipse.rdf4j.sail.lmdb.LmdbUtil.E(LmdbUtil.java:67)
	at org.eclipse.rdf4j.sail.lmdb.TxnManager.startReadTxn(TxnManager.java:65)
	at org.eclipse.rdf4j.sail.lmdb.TxnManager.createReadTxnInternal(TxnManager.java:152)
	at org.eclipse.rdf4j.sail.lmdb.TxnManager.createReadTxn(TxnManager.java:136)
	at org.eclipse.rdf4j.sail.lmdb.LmdbSailStore$LmdbSailDataset.<init>(LmdbSailStore.java:1105)

Reproducer

A Spring Boot RDF4J Server test reproduces this with default RDF4J LMDB settings by:

• creating an LMDB repository through the server
• loading a small dataset
• sending many concurrent POST queries with short server-side timeout values
• asserting that at least one query times out
• asserting that no MDB_READERS_FULL or maxreaders failures are logged or returned
• asserting that a later health query still succeeds

Expected

When LMDB reports MDB_READERS_FULL, RDF4J should close inactive/stale read transactions that no longer represent active query work, check for dead LMDB readers, retry the read transaction acquisition, and keep the repository usable.