murblanc commented on a change in pull request #2373:
URL: https://github.com/apache/lucene-solr/pull/2373#discussion_r576411785



##########
File path: 
solr/core/src/java/org/apache/solr/cluster/placement/plugins/AffinityPlacementFactory.java
##########
@@ -280,25 +250,31 @@ public void verifyAllowedModification(ModificationRequest 
modificationRequest, P
 
     private void verifyDeleteCollection(DeleteCollectionRequest 
deleteCollectionRequest, PlacementContext placementContext) throws 
PlacementModificationException, InterruptedException {
       Cluster cluster = placementContext.getCluster();
-      Set<String> colocatedCollections = 
colocatedWith.getOrDefault(deleteCollectionRequest.getCollection().getName(), 
Set.of());
-      for (String primaryName : colocatedCollections) {
-        try {
-          if (cluster.getCollection(primaryName) != null) {
-            // still exists
-            throw new PlacementModificationException("colocated collection " + 
primaryName +
-                " of " + deleteCollectionRequest.getCollection().getName() + " 
still present");
-          }
-        } catch (IOException e) {
-          throw new PlacementModificationException("failed to retrieve 
colocated collection information", e);
+      Set<String> colocatedWith = new HashSet<>();
+      for (SolrCollection collection : cluster.collections()) {

Review comment:
       We could document to the user that in order to support safe deletes, the 
secondary collection must reference the primary one. Then up to the user to 
decide if the two collections are updated or not...
   At least this leaves a scalable option available.
   The other "cleaner" solution is to make collection references a first class 
citizen managed correctly (so we can maintain mapping and reverse mapping), but 
then the scope increases greatly and changes quite a few things in how 
collection are manages (and `withCollection` would then be just one specific 
usage of this collection reference mechanism, but maybe the only one...).
   
   No perfect solution here, but once we get to very large numbers of 
collections (with collection lists not stored in node memory) the iteration 
will be prohibitively expensive (that's why I didn't want having such an 
iterator in the first place BTW).




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
[email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to