nisplusLDAPentryTtl Attribute

Since the rpc.nisd daemon's local database (in memory and on disk) functions as a cache for LDAP data, the nisplusLDAPentryTtl attribute allows you to set the time-to-live (TTL) values of entries in that cache. There are three TTLs for each database ID. The first two control the initial TTL when the rpc.nisd first loads the corresponding NIS+ object data from disk, and the third TTL is assigned to an object when it is read or refreshed from LDAP.

For example the following results in the rpc.org_dir table object getting an initial TTL randomly selected in the range 21600 to 43200 seconds.

nisplusLDAPentryTtl	rpc_table:21600:43200:43200

When that initial TTL expires and the table object is refreshed from LDAP, the TTL will be set to 43200 seconds.

Similarly the following will assign an initial TTL between 1800 and 3600 seconds to the entries in the rpc.org_dir table when it is first loaded.

nisplusLDAPentryTtl	rpc:1800:3600:3600

Each entry gets its own randomly selected TTL in the range specified. When a table entry expires and is refreshed, the TTL is set to 3600 seconds.

When selecting TTL values, consider the trade-off between performance and consistency. If the TTLs used for LDAP data cached by the rpc.nisd are very long, performance is the same as if rpc.nisd was not mapping data from LDAP at all. However, if the LDAP data is changed (by some entity other than rpc.nisd), it can also take a very long time before that change is visible in NIS+.

Conversely, selecting a very short (or even zero) TTL means that changes to LDAP data are quickly visible in NIS+, but can also impose a significant performance penalty. Typically, an NIS+ operation that also reads data from or writes data to LDAP will take at least two to three times longer (plus the LDAP lookup overhead) than the same operation without LDAP communication. Although performance can vary greatly depending on the hardware resources, scanning a large (tens of thousands or hundreds of thousands of entries) LDAP container to identify NIS+ entries that should be refreshed can take a long time. The rpc.nisddaemon performs this scan in the background, continuing to serve possibly stale data while it is running, but the background scan still consumes CPU and memory on the NIS+ server.

Carefully consider how critical it is to have NIS+ data in close synchronization with LDAP, and select the longest TTL that is acceptable for each NIS+ object. The default (when no nisplusLDAPentryTtl is specified) is 1 hour. The template mapping file /var/nis/NIS+LDAPmapping.template changes this to 12 hours for objects other than table entries. However, there is no auto-recognition of non-entry objects, so if you add mapping for a non-entry object, the TTL will default to 1 hour.

There are no TTLs for nonexistent objects. Hence, no matter which TTLs are in effect for LDAP-mapped entries in an NIS+ table, a request for an entry that does not exist in NIS+ will query LDAP for that entry.