Intensive disk write after long-running LDAP activity terminated

Hello,

we had a long-running (several days) process doing add and modify operations in OpenLDAP 2.4.41 + LMDB backend + SLES 11.3.

When the process was stopped, we experienced minutes of intensive disk write by LMDB.

The LMDB option dbnosync is not set.

Checking git we see that there should be a transaction commit in LMDB after every LDAP add/modify operation, and the updating of on-disk contents should happen after each LDAP add/modify operation.

Furthermore, at the OS level, dirty pages should be generally flushed every 5 seconds.

So we wonder what could possibly cause such a significant I/O burst?

Regards,

tamas