https://bugs.openldap.org/show_bug.cgi?id=9291
--- Comment #2 from Markus markus@objectbox.io --- (In reply to Howard Chu from comment #1)
re: (2) pervasive consistency checks will probably not be implemented since they would break LMDB's #1 reason to exist - fast read performance.
Just to clarify, (2) would be about simple checks that should have no measurable performance impact, e.g. "page number is not zero" where appropriate, not the expensive fsck-like ones from (1).
Would you say that the described situation could be detected efficiently? (Basically my question (3)...)
(Although, we already do have a few such checks when compiled with higher values of MDB_DEBUG.
Thanks for that hint, I just found mdb_audit when looking for MDB_DEBUG. On a first glance it may be something similar to what I had in mind with (1). Will have a closer look at that and other MDB_DEBUG related code.
You always have the option of compiling with your own debug checks as well.) Ultimately, it's not our responsibility to detect broken hardware or filesystems, it is the OS's.
This could be read as you suggest forking for those cases? Again, to clarify, does that include checks that have no measurable performance impact?