Hi, I have tested Cassandra 3.0.10 and 3.0.17 and I have found three potential bugs there.
The test steps with CCM are as follows: 1) Start 3 node cluster 2) Create enough data with cassandra-stress 3) Kill node3 4) Create + Start node4 with replace_address=node3 5) Wait for streaming to start (Check node4 logs for "Prepare completed. Receiving N files”) 6) While node is streaming, drop keyspace1.standard1 Based on my findinds what happens after this are following not-so-great things: 1) Bootstrapping fails (which really shouldn’t because schema changes in bigger always online multi user clusters can happen anytime, same with bootstrapping new nodes / replacing old nodes) 2) At least in my test scenario bootstrap fails / halts and is left in UJ state BUT CQL port is opened and clients can connect normally which is weird. At the same time it is both joining and it allows CQL connections. 3) If I run “bootstrap resume” twice after failure or once before bootstrap has failed, nodetool is reporting that bootstrap is already done (same goes to logs). Which is not true. Bootstrap is ongoing. What do you think, should I open tickets about these? Best regards, Hannu Kröger --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org For additional commands, e-mail: user-h...@cassandra.apache.org
