slapd hungs after being up for over a day under load (ITS#2952)

[masneyb@gftp.org]

Full_Name: Brian Masney
Version: 2.1.25 (20031217)
OS: Debian GNU/Linux
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (216.12.23.12)


There is a bug in slapd that it will hang whenever it's up for more than a day.
It will accept a TCP connection but it will not perform any kind of reads and
writes. 
On our main LDAP master server, after slapd hung on the slave, the entry it hung
on was a delete request. I initially did a strace on the hung slapd process and
it showed this:

futex(0x40e3ec18, FUTEX_WAIT, 14751, NULL <unfinished ...>

Here is a gdb backtrace:

(gdb) bt
#0  0x080501a1 in ber_memcalloc ()
#1  0x080696f5 in ch_calloc ()
#2  0x40556b5e in idl_alloc () from /usr/lib/ldap/back_ldbm.so
#3  0x40556b8b in idl_allids () from /usr/lib/ldap/back_ldbm.so
#4  0x40556c9a in idl_free () from /usr/lib/ldap/back_ldbm.so
#5  0x40557d7a in idl_delete_key () from /usr/lib/ldap/back_ldbm.so
#6  0x4055ce4f in dn2id_delete () from /usr/lib/ldap/back_ldbm.so
#7  0x40561853 in ldbm_back_delete () from /usr/lib/ldap/back_ldbm.so
#8  0x080684bb in do_delete ()
#9  0x08056517 in connection_done ()
#10 0x40026c58 in ldap_pvt_thread_pool_destroy () from /usr/lib/libldap_r.so.2
#11 0x40166964 in start_thread () from /lib/tls/libpthread.so.0
#12 0x42857854 in ?? ()

Also, even though I'm using this on Debian, the source that I am using is the
official 2.1.25 (20031217) source without any of the Debian GNU/TLS patches
applied.