-
Notifications
You must be signed in to change notification settings - Fork 2.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Auth Manager API part 3: OAuth2 Manager #11844
Conversation
6c3453c
to
e502b8b
Compare
e502b8b
to
b88d99d
Compare
@@ -42,6 +57,9 @@ public static AuthManager loadAuthManager(String name, Map<String, String> prope | |||
case AuthProperties.AUTH_TYPE_BASIC: | |||
impl = AuthProperties.AUTH_MANAGER_IMPL_BASIC; | |||
break; | |||
case AuthProperties.AUTH_TYPE_OAUTH2: | |||
impl = AuthProperties.AUTH_MANAGER_IMPL_OAUTH2; | |||
break; | |||
default: | |||
impl = authType; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can we also define something like AUTH_IMPL = "rest.auth.type";
that could be used here for any custom auth manager/provider we want to bring in?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That's precisely the intent here; if you have a custom manager, you would activate it with rest.auth.type=my.custom.AuthManager
. Is that OK for you?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes that works! I wasn't sure whether you were planning on having authType for as a name of the type
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There will be aliases for built-in managers, e.g. you can activate the built-in oauth2 manager with either:
rest.auth.type = oauth2
rest.auth.type = org.apache.iceberg.rest.auth.OAuth2Manager
But for custom ones, you must provide the FQDN of your implementation.
* @param executor the executor to use for eviction tasks; if null, the cache will use the | ||
* {@linkplain ForkJoinPool#commonPool() common pool}. The executor will not be closed when | ||
* this cache is closed. | ||
* @param nanoTimeSupplier the supplier for nano time; if null, the cache will use {@link |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is there a reason we would need to expose this?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No, not really, I can turn this constructor into package-private. It's meant only for tests.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
+1
core/src/main/java/org/apache/iceberg/rest/auth/AuthConfig.java
Outdated
Show resolved
Hide resolved
core/src/main/java/org/apache/iceberg/rest/auth/RefreshingAuthManager.java
Show resolved
Hide resolved
core/src/test/java/org/apache/iceberg/rest/auth/TestAuthSessionCache.java
Outdated
Show resolved
Hide resolved
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
* @param sessionTimeout the session timeout. Sessions will become eligible for eviction after | ||
* this duration of inactivity. | ||
*/ | ||
public AuthSessionCache(Duration sessionTimeout) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It looks like we should also make this package private. The other issue is that we shouldn't be using the common pool by default. This should probably be provided by the OAuth2Manager
or if we expect there to be only one per manager, then just create a named pool for handling refreshes. We definitely don't want to rely on or possibly tie up the common pool.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@danielcweeks we are already using the common pool. The session cache currently is created as follows:
iceberg/core/src/main/java/org/apache/iceberg/rest/RESTSessionCatalog.java
Lines 1144 to 1154 in a2b8008
return Caffeine.newBuilder() | |
.expireAfterAccess(Duration.ofMillis(expirationIntervalMs)) | |
.removalListener( | |
(RemovalListener<String, AuthSession>) | |
(id, auth, cause) -> { | |
if (auth != null) { | |
auth.stopRefreshing(); | |
} | |
}) | |
.build(); | |
} |
As you can see, we are not providing an executor explicitly, so evictions are being done on the common pool.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Re: turning this constructor package-private: this won't fly, as SigV4 will also need to create caches, so this constructor needs to be public.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@danielcweeks I searched for all Caffeine
caches in Iceberg's repo and found 25 occurrences. Only one occurrence sets the executor, the one in CachingCatalog
:
.executor(Runnable::run) // Makes the callbacks to removal listener synchronous |
Consequently, all the others are using the common pool for asynchronous tasks such as eviction.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@danielcweeks do you maintain that you want me to use a different pool here?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, we previously had refresh in a dedicated threadpool, so I don't think the other caffeine examples are what we should point to.
Also, we don't want to rely on the common pool for something important like token refresh. If something ties up the thread pool, we may miss the window to refresh. Also, we don't want to tie up the thread pool with refresh threads if something goes wrong.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is not about token refresh, this is about auth session eviction.
But fair enough. I'll do it.
@adutra Just one last comment on the thread pool, but other than that this look good to go. |
.setNameFormat(name + "-auth-session-evict-%d") | ||
.setDaemon(true) | ||
.build(); | ||
return Executors.newCachedThreadPool(threadFactory); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this is the best executor for this kind of task, as we don't want a core thread kept alive if there isn't anything to do.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should be using a dedicated pool from our ThreadPools
utility. I would suggest: ThreadPools::newExitingWorkerPool(String namePrefix, int poolSize)
with a named pool like we had previously.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
OK, switched to ThreadPools.newExitingWorkerPool
.
3rd PR for the Auth Manager API. Previous ones:
This PR introduces the
OAuth2Manager
along with session caching and credentials refreshing. It is still "unplugged" though. Most of theOAuth2Manager
code reflects one-to-one what's currently hard-coded inRESTSessionCatalog
(and it's still there).\cc @nastra @danielcweeks