[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Server Hangs

On 8/17/11 2:31 PM, Howard Chu wrote:
David Engeset wrote:
I upgraded and updated four of our OpenLDAP servers that we have back in
May to run the latest stable version of OpenLDAP (2.4.23) along with BDB
(4.8.30).  Everything was running with no issues until a little over a
month later one of the servers slapd processes hung, the only way I
could restart the process was to use kill -9, all other kill options
failed.  Over the next month and a half the issue reoccurred on the same
server and occurred on two of the other servers.  There was nothing in
the logs to indicate an issue with running out of file descriptors, dead
locks or anything else.  I set out to see if I could recreate the issue
and I found if I had around 20000 entries, which our database is roughly
around 21000, and ran a script to randomly query, one a time, the
entries in the database and then run another script that added 1000
entries, one at a time, then deleted them in reverse order, one at a
time, and will continue to do so infinitely.  When I ran the two scripts
simultaneously they would hang after 3 to 16 deletes were completed.  I
attempted to use the latest version of OpenLDAP (2.4.26) to see if any
of the bug fixes in it would help and I still get the same results, I
even tried to run it with all of the supported versions of BDB, 4.4,
4.5, 4.6, 4.7, 5.0 and 5.1 with the same results.  I ran it with full
logging on and I was not able to find any thing that pointed to the problem.

Perhaps you should also post your test script so we can reproduce the problem ourselves. Looking at your db_stat output I see a read-lock is dangling out there without an active thread that owns it. Need to be able to inspect each of the server threads under gdb to find which thread the lock belonged to.

We have been running OpenLDAP 2.2 and 2.3 for years (many servers
without any restarting of slapd for over a year) without any lockups, so
I decided to test with OpenLDAP 2.3.43 with BDB 4.2.52 (with patches)
and loaded the same exact database and the same exact tests and it runs
literally for hours with no issues.  I attempted to upgrade the version
of BDB to 4.4 and I started to experience the hanging again, so it
appears to be a BDB issue.  I searched for related issues with no
success and considering that others are running 2.4 with newer versions
of BDB for a couple of years now I find it odd that I am running into
this issue on my first use of 2.4.

I tested all of this on CentOS 5.4, 5.6 and Fedora 17 with the same
results.  Does anyone have any ideas or suggestions on what I can try to
do to fix this issue?

Below are some of the configs I am using on my last attempts to resolve
the issue:

set_cachesize 0 536870912 1
set_lg_regionmax 10485760
set_lg_max 104857600
set_lg_bsize 2097152
set_lg_dir /var/log/bdb
set_tmp_dir /var/log/bdb
# This one I added recently to see if it might help.
set_lk_detect DB_LOCK_DEFAULT

include         /usr/local/etc/openldap/schema/cosine.schema
include         /usr/local/etc/openldap/schema/nis.schema
include         /usr/local/etc/openldap/schema/misc.schema
include         /usr/local/etc/openldap/schema/inetorgperson.schema

pidfile         /usr/local/var/run/slapd.pid
argsfile        /usr/local/var/run/slapd.args
conn_max_pending 1000

database        bdb
cachesize       20000
suffix          "dc=example,dc=net"
checkpoint      5120 30
rootdn          "cn=Manager,dc=example,dc=net"
rootpw         secrect
directory       /usr/local/var/openldap-data

# Indices to maintain
index default pres,eq
index cn,uid
#index WhidNetCustID,CustID,ID
index sn pres,eq,sub
index   objectClass     eq
index   uidNumber eq
index   gidNumber eq
index   memberUid eq

# database access control definitions
access to attrs=userPassword
          by self write
          by anonymous auth
          by dn="cn=Admin,dc=example,dc=net" write
          by * none
access to *
          by self write
          by dn="cn=Admin,dc=example,dc=net" write
          by * read

I can send out the LDIF I am using and the perl scripts that I run to
break it for anyone who is interested.
Thank you,

Attached is the configs I used as well as the two perl scripts. The first perl script searchldap.pl just calls the ldapsearch command to lookup a random selected user in the database, and does so infinitely. The second perl script adddelldap.pl uses the Net::LDAP module and connects to the database and adds 1000 records and then removes them, each complete cycle I call a round and echo that out to the display and it will continue to do so for 1000 rounds. I had the script set to close and open new connections after X number of random actions have been completed, to simulate a typical set of changes in a single connection, it is limited to a maximum of 10 operations for a single connection. I am unable to attach the LDIF I used due to size, but it had roughly 20000 records under the ou=people branch.
Thank you,

Whidbey Telecom Internet and Broadband
Software Engineer

Attachment: configs.scripts.tgz
Description: GNU Zip compressed data