CORE: return false when view exists endpoint isn't supported

[danielcweeks]

No description provided.

[danielcweeks]

(cc @ajreid21)

[rdblue]

Wasn't the previous behavior to attempt to load the view if the HEAD request is not supported? I think it would be safer to check V1_LOAD_VIEW and use that if it is supported before returning false.

[nastra]

I think we might actually need to fall back to loading the view as we did previously. Also btw the same issue applies for namespaceExists and tableExists as those three are all now doing a HEAD request. We might need something like

--- a/core/src/main/java/org/apache/iceberg/rest/RESTSessionCatalog.java
+++ b/core/src/main/java/org/apache/iceberg/rest/RESTSessionCatalog.java
@@ -444,12 +444,14 @@ public class RESTSessionCatalog extends BaseViewSessionCatalog

   @Override
   public boolean tableExists(SessionContext context, TableIdentifier identifier) {
-    Endpoint.check(endpoints, Endpoint.V1_TABLE_EXISTS);
-
     try {
       checkIdentifierIsValid(identifier);
-      client.head(paths.table(identifier), headers(context), ErrorHandlers.tableErrorHandler());
-      return true;
+      if (endpoints.contains(Endpoint.V1_TABLE_EXISTS)) {
+        client.head(paths.table(identifier), headers(context), ErrorHandlers.tableErrorHandler());
+        return true;
+      } else {
+        return super.tableExists(context, identifier);
+      }
     } catch (NoSuchTableException e) {
       return false;
     }
@@ -665,13 +667,15 @@ public class RESTSessionCatalog extends BaseViewSessionCatalog

   @Override
   public boolean namespaceExists(SessionContext context, Namespace namespace) {
-    Endpoint.check(endpoints, Endpoint.V1_NAMESPACE_EXISTS);
-
     try {
       checkNamespaceIsValid(namespace);
-      client.head(
-          paths.namespace(namespace), headers(context), ErrorHandlers.namespaceErrorHandler());
-      return true;
+      if (endpoints.contains(Endpoint.V1_NAMESPACE_EXISTS)) {
+        client.head(
+            paths.namespace(namespace), headers(context), ErrorHandlers.namespaceErrorHandler());
+        return true;
+      } else {
+        return super.namespaceExists(context, namespace);
+      }
     } catch (NoSuchNamespaceException e) {
       return false;
     }
@@ -1239,12 +1243,14 @@ public class RESTSessionCatalog extends BaseViewSessionCatalog

   @Override
   public boolean viewExists(SessionContext context, TableIdentifier identifier) {
-    Endpoint.check(endpoints, Endpoint.V1_VIEW_EXISTS);
-
     try {
       checkViewIdentifierIsValid(identifier);
-      client.head(paths.view(identifier), headers(context), ErrorHandlers.viewErrorHandler());
-      return true;
+      if (endpoints.contains(Endpoint.V1_VIEW_EXISTS)) {
+        client.head(paths.view(identifier), headers(context), ErrorHandlers.viewErrorHandler());
+        return true;
+      } else {
+        return super.viewExists(context, identifier);
+      }

[danielcweeks]

We have some inconsistency about how we're handling this. The listViews takes the same approach as what I have here, which also aligns with our DEFAULT_ENDPOINTS, so it's strange that it would call load view, but list view would return nothing.

[ebyhr]

+1 for fallback to older logic as @nastra suggested. I sent #12294 with some tests.

I decided to skip Iceberg 1.8.0 in Tirno due to this endpoint's issue and S3 strong integrity checksum issue #12264.

[ajantha-bhat]

Is it possible to add the testcase in TCK?
Compatibility suite was suppose to catch this?

[snazy]

I think this PR should not just partly fix the issue for views, but also for tables and namespaces.

[nastra]

Closing this one in favor of #12314