dlmarion commented on PR #3375: URL: https://github.com/apache/accumulo/pull/3375#issuecomment-1536174194
I remembered that at one point I had made this configurable. I went back and looked and I did in an early [commit](https://github.com/apache/accumulo/pull/1818/commits/d66ba42bc3ac227d85aaa0cbce98cf179dd01be8#diff-deac7e822d8841e7105b78917860aae5a0bfc2760f30aaa8f2d74dd9c5fa4761) in the PR that introduced this behavior. We also had this discussion on that PR as well ([ref](https://github.com/apache/accumulo/pull/1818/files#r543599320)). If we take VFS out of the equation - if a user mistakenly specifies that classpath for a classloader context, then runs a scan on table that uses that context to load iterator classes, then they can take down a good portion or the entire cluster. However, I believe that the client makes a call to the Tablet Server to test that it can load the iterator classes. I believe the issue here is the class can only be loaded by a portion of the TabletServers, due to VFS. Given that we have users using the VFSClassLoader, I think we should resurrect the configuration property. If we can somehow scope that to apply only to scan threads and only to classloading Errors, that might be appropriate. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
