Daniel Gergely created AMBARI-15646:
---------------------------------------
Summary: Audit Log Code Cleanup & Safety
Key: AMBARI-15646
URL: https://issues.apache.org/jira/browse/AMBARI-15646
Project: Ambari
Issue Type: Bug
Components: ambari-server
Reporter: Daniel Gergely
Assignee: Daniel Gergely
As a follow-up to AMBARI-15241, there are concerns brought up in review which
should be addressed but didn't need to hold up the feature being committed.
These can be further broken out to into separate Jiras if needed:
- When initializing a ThreadLocal, you can specify an initial value. This code
is unnecessary:
{code}
private ThreadLocal<DateFormat> dateFormatThreadLocal = new ThreadLocal<>();
if(dateFormatThreadLocal.get() == null) {
//2016-03-11T10:42:36.376Z
dateFormatThreadLocal.set(new
SimpleDateFormat("yyyy-MM-dd'T'HH:mm:ss.SSSX"));
}
{code}
- There are no tests for a majority of events and event creators.
- Using a multibinder is fine to be able to inject a {{Set<Foo>}}, but it's not
clear to developers adding code that this is what must be done.
-- We either need to document the super interface to make it clear how to have
new creators bound
-- Or annotate creators with an annotation which then be automatically picked
up by the {{AuditLoggerModule}} and bound without the need to explicitly define
each creator.
- {code}
// binding for audit event creators
Multibinder<RequestAuditEventCreator> auditLogEventCreatorBinder =
Multibinder.newSetBinder(binder(), RequestAuditEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(DefaultEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ComponentEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ServiceEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(UnauthorizedEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ConfigurationChangeEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(UserEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(GroupEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(MemberEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(PrivilegeEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(BlueprintExportEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ServiceConfigDownloadEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(BlueprintEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ViewInstanceEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ViewPrivilegeEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(RepositoryEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(RepositoryVersionEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(AlertGroupEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(AlertTargetEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(HostEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(UpgradeEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(UpgradeItemEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(RecommendationIgnoreEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(ValidationIgnoreEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(CredentialEventCreator.class);
auditLogEventCreatorBinder.addBinding().to(RequestEventCreator.class);
bind(RequestAuditLogger.class).to(RequestAuditLoggerImpl.class);
{code}
- Event creators have nested invocations which is not only hard to read, but
can potentially cause NPE's; it's a dangerous practice. As an example:
{code:title=AlertGroupEventCreator}
String.valueOf(request.getBody().getNamedPropertySets().iterator().next().getProperties().get(PropertyHelper.getPropertyId("AlertGroup",
"name")));
{code}
-- Additionally, this references properties by building them, instead of by
their registration in the property provider. If the property name ever changed,
this could easily be missed.
- Some of the {{auditLog}} methods check to ensure that the logger is enabled
first. This is very good, as building objects which won't be logged is a waste
and potential performance problem. However, not all of them do. All
{{auditLog}} methods should check this first, and return if not enabled. You
can do this using AOP or just brute-force every method.
{code}
private void auditLog(HostRoleCommandEntity commandEntity, Long requestId) {
if(!auditLogger.isEnabled()) {
return;
}
{code}
- The temporary status caches in {{ActionDBAccessorImpl}} are never cleared.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
