It is not intuitive that in order to get all defined databases in an environment we must use a write lock to prevent any other platform thread or virtual thread or effect... to do the same operation at the same time in order to avoid random crashes under heavy concurrent situation which is my case.

I'm using lmdbjava 0.8.2 (from scala 3 with ZIO 2 library), with an environment configured with the following flags :

                               Env
                                 .create()
                                 .setMapSize(100_000_000_000L)
                                 .setMaxDbs(10_000)
                                 .setMaxReaders(100)
                                 .open(
                                   databasesPath,
                                   EnvFlags.MDB_NOTLS,
                                   EnvFlags.MDB_NOLOCK, // Locks managed using ZIO ReentrantLock
                                   EnvFlags.MDB_NOSYNC
                                 )

So in my databases list implementation, I replaced reentrantLock.withReadLock by reentrantLock.withWriteLock:

  def databases(): Task[List[String]] = {
    reentrantLock.withWriteLock(
      ZIO.logSpan("databases") {
        for {
          databases <- Task
                         .attempt {
                           env
                             .getDbiNames()
                             .asScala
                             .map(bytes => new String(bytes))
                             .toList
                         }
                         .tap(l => ZIO.log(s"${l.size} databases found : ${l.mkString(",")}"))
        } yield databases
      }
    )
  }

which allows to only have one simultaneous call to getDbiNames and prevent all the crashes I had until now. After taking a look to getDbiNames() implementation I've understood why such a user side write lock must be implemented.

  public List<byte[]> getDbiNames() {
    final List<byte[]> result = new ArrayList<>();
    final Dbi<T> names = openDbi((byte[]) null);
    try (Txn<T> txn = txnRead();
    ....

this is related to the use of openDbi((byte[])null) which requires only one simultaneous access. So what can we do about this ? I spent something like more than ~20 hours to identify the issue as I think it is quite easy to fall into this trap. But thanks to it a performancement enhancement has been made to the ZIO library (zio/zio#6674).

• I'm not sure that it is easily fixable, is there an other way to get all defined databases within an environment ?
• I can prepare a Pull Request with various information in order to help to deal with concurrency (in the wiki : https://github.com/lmdbjava/lmdbjava/wiki/Concurrency) ?

thanks in advance for your response,
David.