Gleb Natapov

In the `test_replace_active_node` test such error is expected and suppose to be filtered out by the test if `consistent-topology-changes` is enabled. Probably a fallout from making consistent topology default @patjed41.

> Therefore, at the moment of the test trying to do a CL=THREE query through node 2, the node thinks that there is a pending replica, so it requires 4...

@juliayakovlev @roydahan how is this rolling restart nemesis decides that it can restart next node?

> There is -- through node_ops RPCs and ring_delay sleeps. Where exactly? There is no ring delay with node_ops between write_both_read_old and write_both_read_new state because there is no write_both_read_new state....

Reproduces more easily with: ``` diff --git a/service/storage_proxy.cc b/service/storage_proxy.cc index 21ec5f8ba3..f9f23c17ee 100644 --- a/service/storage_proxy.cc +++ b/service/storage_proxy.cc @@ -5842,6 +5842,7 @@ storage_proxy::query_partition_key_range_concurrent(storage_proxy::clock_type::t co_await coroutine::maybe_yield(); } } + co_await sleep(std::chrono::milliseconds(100)); slogger.trace("creating range...

Why do you have to use maintenance socket as opposite to regular connection?

Not sure. Here it looks like bootstrapping failed because streaming failed. And streaming failed because it was aborted.

> @kbr-scylla , it is another issue. Here decommission was expected to be aborted, but bootstrap of new node expected to be successfull, while it failed So what happened is...

This log message: ``` 2024-04-10T17:10:57.911+00:00 parallel-topology-schema-changes-mu-db-node-7c79549b-17 !ERR | scylla[7463]: [shard 0: gms] raft_topology - raft_topology_cmd failed with: std::runtime_error ({shard 0: std::runtime_error (repair[44489ed4-6f7e-42c5-869e-e961be26b71a]: 3 out of 7745 ranges failed, keyspace=keyspace1, tables={standard2,...