Hi Howard,

Many thanks for your suggestions. I am about to try what you've suggested (download and compile the latest master version of lmdb from git using master branch of https://github.com/LMDB/lmdb).

However, just to note, I am running the latest version of zimbra which uses pretty recent version of openldap:

ii zimbra-lmdb 2.4.46-1zimbra8.7b amd64 LMDB Binaries ii zimbra-lmdb-lib:amd64 2.4.46-1zimbra8.7b amd64 LMDB Libraries ii zimbra-openldap-client 2.4.46-1zimbra8.7b amd64 OpenLDAP Client Binaries ii zimbra-openldap-lib:amd64 2.4.46-1zimbra8.7b amd64 OpenLDAP Libraries ii zimbra-openldap-server 2.4.46-1zimbra8.7b amd64 OpenLDAP Server Binaries

The problem that I am describing occurred with the above version of openldap.

I will keep you posted with any updates.

Cheers Andrei

----- Original Message -----

From: "Howard Chu" hyc@symas.com To: "Andrei Mikhailovsky" andrei@arhont.com, "openldap-technical" openldap-technical@openldap.org Sent: Thursday, 7 February, 2019 02:42:40 Subject: Re: help with mdb database recovery after crash

Andrei Mikhailovsky wrote:

...

Hello everyone,

I have a bit of an issue with my ldap database. I have a Zimbra community edition which uses openldap. A server crashed and I am unable to start the ldap services after the reboot. The description of my problem, after some digging about is:

the initial error indicated problem with the ldap

Starting ldap...Done. Search error: Unable to determine enabled services from ldap. Enabled services read from cache. Service list may be inaccurate.

Having investigated the issue, I noticed the following errors in the zimbra.log

*slapd[31281]: mdb_entry_decode: attribute index 560427631 not recognized*

I also noticed that the /opt/zimbra/data/ldap/mdb/db/data.mdb is actually 81Gb in size and had reached the limit imposed by the ldap_db_maxsize variable. so over the weekend, the LDAP mdb file became no longer sparse.

I tried following the steps described in https://syslint.com/blog/tutorial/solved-critical-ldap-primary-mdb-database-... but with no success, as the slapcat segfaults with the following message.

/opt/zimbra/common/sbin/slapcat -ccc -F /opt/zimbra/data/ldap/config -b "" -l /opt/zimbra/RECOVERY/SLAPCAT/zimbra_mdb.ldiff 5c583982 mdb_entry_decode: attribute index 560427631 not recognized Segmentation fault (core dumped)

the mdb_copy produces a file of 420 mb in size, but it still contains the "mdb_entry_decode: attribute index 560427631 not recognized" error. I've also tried mdb_dump, but had the same issues after using the mdb_load command.

I found a post ( http://www.openldap.org/its/index.cgi/Software%20Bugs?id=8360 ) in the openldap community that the mdb gets corrupted if it reaches the maximum defined size. but no solution of how to fix it.

That's from over 3 years ago and has subsequently fixed. If you're running on such an old release, there's likely not much that can be done. Ordinarily it's possible to back up to the immediately preceding transaction, in case the last transaction is corrupted, but with that particular bug it's likely that the corruption occurred in an earlier transaction and has been carried forward in all subsequent ones.

...

any advice on getting the database recovered and working again?

You could try using the preceding transaction and see if it's in any better shape. The code for this is not released in LMDB 0.9. You can compile the mdb.master branch in git to obtain it. Then use the "-v" option with mdb_copy and see if that copy of the database is usable.

-- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/