| 1 |
--source include/have_ndb.inc |
| 2 |
--source include/have_ndb_debug.inc |
| 3 |
|
| 4 |
--echo Testing for bug#24444908 / bug#25128512 |
| 5 |
################################################# |
| 6 |
# Testcase for Bug#24444908: |
| 7 |
# |
| 8 |
# CLUSTER CRASHED DURING RESTART WITH AN ASSERTION |
| 9 |
# !CALLBACKOBJ->HAS_DATA_TO |
| 10 |
# |
| 11 |
# We had no thread concurrency safe way of checking for |
| 12 |
# disconnected nodes while sending to them. Thus there |
| 13 |
# could be race conditions where a thread passed the |
| 14 |
# 'is connected to node <n>' check while another thread |
| 15 |
# disconnected and cleared the send buffers to node <n>. |
| 16 |
# If the sending thread then was 'slow', it could append |
| 17 |
# more data to the just cleared send buffers. |
| 18 |
# |
| 19 |
# Upon reconnect we then hit an assert as we assumed the |
| 20 |
# send buffers to still be empty. |
| 21 |
# |
| 22 |
# With release built binaries this resulted in problems as in |
| 23 |
# Bug#25128512: ..'WE(1) HAVE BEEN DECLARED DEAD BY 2 (VIA 4)' |
| 24 |
# Due to messages created as part of the disconnects not being |
| 25 |
# cleared, and was sent after we reconnected again. Thus it |
| 26 |
# was interpretted as being yet another disconnect. |
| 27 |
################################################# |
| 28 |
|
| 29 |
--echo Restart node 1 while other node(s) keps try sending to it |
| 30 |
--exec $NDB_MGM -e "all dump 2355 1" >> $NDB_TOOLS_OUTPUT |
| 31 |
|
| 32 |
--echo Wait for node 1 to go down. |
| 33 |
--exec $NDB_WAITER --wait-nodes=1 --no-contact |
| 34 |
|
| 35 |
--echo Waiting for node 1 to get started again |
| 36 |
--exec $NDB_WAITER |
| 37 |
|
| 38 |
--echo End of test for bug#24444908 / bug#25128512 |
| 39 |
--remove_file $NDB_TOOLS_OUTPUT |