Re: Error determining LDAP server type: Timed out

On 8/04/25 06:42, Reinhard Westerholt wrote:
> Hello everybody,
> 
> I have configured Squid to use Kerberos authentication and employed 
> ext_kerberos_ldap_group_acl to limit HTTP access. This setup works fine 
> in most cases. However, randomly (and rarely), there seems to be an 
> issue with the LDAP query, causing the Active Directory groups to fail 
> to resolve, which subsequently denies user access. When this occurs, the 
> issue persists for about an hour. I suspect this might be due to caching 
> or TTL.

Very likely.

> 
> Here is a relevant part of the configuration:
> 
> auth_param negotiate program /usr/lib/squid/negotiate_kerberos_auth -s 
> HTTP/server.fqdn@domain
> auth_param negotiate children 500
> auth_param negotiate keep_alive on
> 
> external_acl_type INetAccess ttl=3600 children-max=1000 %LOGIN /usr/lib/ 
> squid/ext_kerberos_ldap_group_acl -g "INetAccess" -a -i -l ldap:// 
> server.fqdn:389 -D "DOMAIN"

There you have it " ttl=3600 "  (ie 1 hour, in seconds). If this is the 
only TTL configured it will be used for both OK and ERR results.


...
> 
> I manually attempted to retrieve the subschemasubentry using ldapsearch. 
> As far as I can see, there is no issue, or I could not reproduce the 
> problem manually via ldapsearch.
> 
> Does anyone have any ideas on how to further debug this issue? 
> Additionally, are there any recommendations regarding the TTL value? If 
> I reduce the TTL, the duration of the issue might be shorter for 
> individual users.


There are three alternatives to changing "ttl=" itself.

In order of my recommendation:

1) Avoiding the external helper entirely if you can.

The auth helper you are using should be providing Squid with a list of 
the group SSID's the logged in user is a member of. With those, Squid 
can use the 'note' type ACL to check the groups quickly instead of 
needing a separate helper lookup.

Like so:
   acl INetAccess note group ...[SSID]...


2) Setting "grace=" which will attempt the re-lookup before the TTL 
finishes. That way Squid will use the already cached value at the exact 
time the failure occurs, and may be able to re-try the failed lookup at 
least once before it has any effects.
   This grace= option should ideally be only a few seconds and must be 
strictly less than both ttl= and negative-ttl= values.


3) Add a shorter value for the ERR results with "negative-ttl=".

Just be aware that lowering either of these TTL values will impact 
transaction speed for all the traffic where the user is (ttl=), or not 
(negative-ttl=) a member of that group.


HTH
Amos