Hi, I am trying to create a TPC-DS database on Kudu with a scale factor of 1000. My kudu version has been built using the 1.0.0-SNAPSHOT source. My cluster has 4 kudu-tserver instances. The tables use a replication factor of 3.
I am using simple Impala insert statements to load the tables: insert into kudu-table select * from parquet-table. This insert overwrites the contents of several block_manager_instance files in the cluster. These files get filled with messages of this type: E1020 10:25:05.947288 19199 consensus_queue.cc:349] T 5acb121ecb9945f79fe25e30d49df4e3 P 5ba624b4123a464fa95a11bfdbef9210 [LEADER]: Error trying to read ahead of the log while preparing peer request: Incomplete: Op with index 405031 is ahead of the local log (next sequential op: 405031). Destination peer: Peer: 7aad1905ae144a029788546e2a104909, Is new: false, Last received: 304.405031, Next index: 405032, Last known committed idx: 405031, Last exchange result: ERROR, Needs remote bootstrap: false The tserver cannot be restarted after this because it reads these files on startup and fails with an "invalid magic number" message. There is no recovery possible. I have to delete all files in the database and recreate all tables. This corruption does not occur when I set the replication factor to 1. Is anyone aware of this problem? Is there a fix available? I have attached one of the block_manager_instance files, if it helps. Thanks, Ravi
