Debian Codegraph @

mysql-test/suite/ndb/t/ndb_sendbuffer.test

raw
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