]
ehsavoie Hugonnet updated WFCORE-2978:
--------------------------------------
Summary: Verify that WFCORE-2923 fix is valid (was: Verify that JBEAP-11343 fix is
valid)
Verify that WFCORE-2923 fix is valid
-------------------------------------
Key: WFCORE-2978
URL:
https://issues.jboss.org/browse/WFCORE-2978
Project: WildFly Core
Issue Type: Bug
Components: Logging, Security
Reporter: ehsavoie Hugonnet
Assignee: ehsavoie Hugonnet
Priority: Blocker
[1:39 PM] Emmanuel Hugonnet: @BrianStansberry hi, could you take a look at
https://github.com/wildfly/wildfly-core/pull/2514
[1:40 PM] Emmanuel Hugonnet: I've a bit of a doubt because i couldn't create a
dependency on the credentialstore since auditlog handlers are not services
[1:41 PM] Brian Stansberry: ok. it does seem a bit nasty because of that
[1:41 PM] Brian Stansberry: at a glance
[1:41 PM] Emmanuel Hugonnet: yes
[1:41 PM] Brian Stansberry: a very quick glance
[1:42 PM] Brian Stansberry: ah , but SyslogAuditLogHandler is not an OSH so I won't
comment until I really understand :)
[1:43 PM] Emmanuel Hugonnet: yes ;)
[2:04 PM] Brian Stansberry: I don't think that will be reliable; there's no
guarantee that store will be started
[2:07 PM] Kabir Khan: I don't think the syslog handler tries to write until boot is
done
[2:07 PM] Kabir Khan: could it be possible to lazy init those suppliers?
[2:11 PM] Emmanuel Hugonnet: I guess we would only need the attribute value and a
serviceregistry
[2:12 PM] Emmanuel Hugonnet: I'm wondering if this is not lazy per default
[2:12 PM] Emmanuel Hugonnet: as the service is only called when the credential is
required as far as I can see
[2:13 PM] Brian Stansberry: yes, it is lazy
[2:43 PM] Kabir Khan: @ehsavoie I think the problem @BrianStansberry mentions is whether
the services have stabilised so the CR is ready by the time the syslog write happens
[2:44 PM] Emmanuel Hugonnet: @KabirKhan yes but I don't have any service in the audit
log tree to be able to require for the CR to be ready
[2:45 PM] Kabir Khan: @ehsavoie I can try to discuss it on the pm call, perhaps we can do
without it for the beta
[2:47 PM] Emmanuel Hugonnet: or I could add a service to get this one on
[2:48 PM] Emmanuel Hugonnet: a bit like what is done for security realms
[2:53 PM] Kabir Khan: That could be good for the future, but I think for Beta it should
be ok as it is
[2:57 PM] Brian Stansberry: @KabirKhan @ehsavoie +1
[2:57 PM] Brian Stansberry: I think it is fine at boot as no logging will happen until
MSC has stabilized
[2:57 PM] Brian Stansberry: and probably post boot too
[2:57 PM] Emmanuel Hugonnet: well the boot test is ok
[2:58 PM] Kabir Khan: but we should have a blocker jira to investigate whether our
assumptions are correct
[2:58 PM] Emmanuel Hugonnet: ok
[2:58 PM] Brian Stansberry: the scenario is a management op adds the credential store and
the syslog handler, and then the syslog handler wants to log before the store service is
up
[2:58 PM] Brian Stansberry: but, for audit logging we don't log until op
commit/rollback
[2:58 PM] Brian Stansberry: and by then MSC is going to be stablized
[3:06 PM] Kabir Khan: that should be: 'we think we don't log', so that needs
checking :)