Node fails to shut down gracefully during INSERT load

[Alex Yurchenko]

Start insert load like:

./sysbench --mysql-host=127.0.0.1 --mysql-port=3305 --max-requests=0 --num-threads=16 --test=oltp --oltp-test-mode=complex --oltp-table-size=1000000 prepare

Try to shut down one of the nodes and you'll get:

110504 18:18:56 [Note] WSREP: applier has been aborted, skipping apply_rbr: 25
110504 18:18:56 [ERROR] WSREP: Failed to apply app buffer: Yn�M, seqno: 25, status: WSREP_FATAL
  at galera/src/replicator_smm.cpp:apply_data():77
  at galera/src/replicator_smm.cpp:apply_trx_ws():216
110504 18:18:56 [Note] WSREP: recv thread exiting (code:6)
110504 18:18:56 [ERROR] WSREP: node consistency compromised, aborting

After which node hangs.

[Alex Yurchenko (ayurchen)]

looks like the check thd->wsrep_exec_mode==REPL_RECV in is_client_connection() fails for wsrep applier (wsrep_exec_mode is for some reason LOCAL_STATE) and it is closed as a client connection.

While it is not entirely clear why wsrep_exec_mode in applier is LOCAL_STATE, adding a check for thd->wsrep_applier seems to fix this particular situation.

[Alex Yurchenko (ayurchen)]

fix committed in r3052