Dovecot supports caching the results of passdb and userdb lookups. The following rules apply to using the authentication cache:
Data is used from the cache if it's not expired (auth_cache_ttl setting)
If a database lookup fails because of some internal error, but data still exists in the cache (even if expired), the cached data is used. This allows Dovecot to log in some users even if the database is temporarily down.
The authentication cache can be flushed by sending a SIGHUP to dovecot-auth.
Sending SIGUSR2 to dovecot-auth makes it log the number of cache hits and misses. You can use that information for tuning the cache size and TTL.
TIP
It should be pretty safe to set very high TTLs, because the only field that usually can change is the user's password, and Dovecot attempts to catch those cases (see the rules above).
auth_cache_negative_ttl| Default | 1hour |
|---|---|
| Value | time |
This sets the time to live for negative hits to passdb or userdb (i.e., when the user is not found or there is a password mismatch).
The value 0 completely disables caching of these hits.
auth_cache_size| Default | [None] |
|---|---|
| Value | size |
The authentication cache size (e.g., 10M).
auth_cache_size = 0 disables use of the authentication cache.
A typical passdb cache entry is around 50 bytes and a typical userdb cache entry is around 100-200 bytes, depending on the amount of information your user and password database lookups return.
auth_cache_ttl| Default | 1hour |
|---|---|
| Value | time |
Time to live for cache entries.
After the TTL expires, the cached record is no longer used, unless the main database look-up returns internal failure.
Entries are removed from the cache only when the cache is full and a new entry is to be added.
auth_cache_verify_password_with_worker| Default | no |
|---|---|
| Value | boolean |
The auth master process by default is responsible for the hash
verifications. Setting this to yes moves the verification to auth-worker
processes. This allows distributing the hash calculations to multiple CPU
cores, which could make sense if strong hashes are used.
Usually only the username uniquely identifies a user, but in some setups you may need something more, for example the remote IP address.
For SQL and LDAP lookups Dovecot figures this out automatically by using all the used Config variables as the cache key. For example, if your SQL query contains %s, %u, and %r the cache entry is used only if all of them (service name, username and remote IP) match for the new lookup.
With other databases Dovecot doesn't know what could affect caching, so you have to tell Dovecot manually. The following databases require specifying the cache key (via cache_key):
For example if the PAM lookup depends on username and service, you can use:
passdb {
driver = pam
args = cache_key=%s%u *
}User logs in with password X. The password X is added to cache and login succeeds.
Password is changed to Y.
User logs in with password Y. The cached password X doesn't match Y, but since the previous authentication was successful Dovecot does another backend passdb lookup to see if the password changed. It did, so the password Y is cached and login succeeds.
User logs in with password X. The password X is added to cache and login succeeds.
Password is changed to Y.
User logs in with password X. The cached password X matches X, so login succeeds.
User logs in with password X. The password X is added to cache and login succeeds.
User logs in with password Y. The cached password X doesn't match Y, but since the previous authentication was successful Dovecot does another backend passdb lookup to see if the password changed. It didn't, so the login fails.
Password is changed to Y.
User logs in with password Y. The cached password X doesn't match Y and the previous authentication was unsuccessful, so Dovecot doesn't bother doing another backend passdb lookup (until cache TTL expires). The login fails.