https://bugs.openldap.org/show_bug.cgi?id=10477

Issue ID: 10477 Summary: Reopening LMDB environment clears MDB_MAP_FULL error Product: LMDB Version: 0.9.33 Hardware: x86_64 OS: Linux Status: UNCONFIRMED Keywords: needs_review Severity: normal Priority: --- Component: liblmdb Assignee: bugs@openldap.org Reporter: david.komarek@whalebone.io Target Milestone: ---

Hello,

We are using LMDB from multiple processes. One process performs writes, while several other processes (typically 4–12, depending on the number of CPUs) read from the same environment. The environment contains multiple (6) databases, and on average we write approximately 150 records per second.

After some time, we begin encountering MDB_MAP_FULL errors on every write operation. Interestingly, after reopening the environment in the writer process, the error disappears and writing resumes successfully.

Our configured map size is 4 GB, and we have not implemented any resizing strategy. Based on our observations, this does not seem strictly necessary, as there appears to be sufficient free space after reopening the environment.

Reader processes do not keep read transactions open for long periods. We reset transactions using mdb_txn_reset after each read operation and reuse them via mdb_txn_renew.

The environment is open with following flags for writing: MDB_WRITEMAP | MDB_NOTLS | MDB_NORDAHEAD | MDB_NOSYNC and for reading: MDB_RDONLY | MDB_NOTLS | MDB_NORDAHEAD.

Could you please help us understand what might be causing this behavior and how we can prevent MDB_MAP_FULL errors without needing to reopen the environment?

Thank you in advance.