too many open files (IDLE socket types) (ITS#3011)

[jmcclure@lucent.com]

Full_Name: James McClure
Version: 2.1.21
OS: Solaris 9
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (65.247.141.10)


Environment is:
Sun Netra 20 (t4), Solaris 9, OpenLDAP 2.1.21, BDB Berkely 4.1.25

My particular LDAP server deployment contains about 500K entries, and typically
services a dozen or so clients (both application/daemon and user types).  After
approx. a week, slapd hangs reporting the following error message: "Mar  9
04:20:36 mtlecm01 slapd[10366]: [ID 285381 local4.debug] daemon: accept(7)
failed errno=24 (Too many open files)".

Since the number of clients is small, I don't feel that that upper limit of 256
descriptors is the underlying cause in this situation.  Especially when I find
that a large percentage of the file descriptors in use are IDLE state sockets
that are not assocaited with anything (lsof output):
...
slapd   10366 root   25u  IPv4 0x300179501b8     0t1646    TCP
mtlecm01:*->mtlecm01:* (IDLE)
slapd   10366 root   26u  IPv4 0x300179516b8     0t1650    TCP
mtlecm01:*->mtlecm01:* (IDLE)
slapd   10366 root   27u  IPv4 0x30045379cf0     0t1650    TCP
mtlecm01:*->mtlecm01:* (IDLE)
...
slapd     10366   root  255u  IPv4 0x30017c979c8       0t48    TCP
mtlecm01:*->mtlecm01:* (IDLE)

I could not find reference to an issue thread about slapd having trouble
determining that a client died or exited without gracefully closing the
connection.  This is all that I can think of at the moment that could be
happening, since the sockets appear unconnected to any known endpoint.

Thanks.
James McClure
jmcclure@lucent.com