| 1 |
--source include/have_ndb.inc |
| 2 |
--source include/have_ndb_debug.inc |
| 3 |
|
| 4 |
--echo A client holding a row lock can cause node restart |
| 5 |
--echo (CopyFrag) to stall until the row lock is released. |
| 6 |
--echo |
| 7 |
--echo Check that a stalled CopyFrag does not lead to |
| 8 |
--echo TC takeover stall by : |
| 9 |
--echo - Restarting a node |
| 10 |
--echo - Stalling CopyFrag due to a rowlock |
| 11 |
--echo - Causing starting node failure |
| 12 |
--echo - Checking that TC takeover completes |
| 13 |
--echo (As transactions can be committed in |
| 14 |
--echo different epochs) |
| 15 |
|
| 16 |
create table t1(id int primary key, val int) engine=ndb; |
| 17 |
|
| 18 |
--disable_query_log |
| 19 |
let $i = 0; |
| 20 |
while($i < 10) |
| 21 |
{ |
| 22 |
eval insert into t1 values ($i, $i); |
| 23 |
inc $i; |
| 24 |
} |
| 25 |
--enable_query_log |
| 26 |
|
| 27 |
select count(*) from t1; |
| 28 |
|
| 29 |
# Restart one data node with -n (nowait) |
| 30 |
# and ensure it stopped |
| 31 |
--exec $NDB_MGM -e "2 restart -n" >> $NDB_TOOLS_OUTPUT |
| 32 |
--exec $NDB_WAITER --nowait-nodes=1 --not-started > $NDB_TOOLS_OUTPUT |
| 33 |
|
| 34 |
connect (conn1,localhost,root,,); |
| 35 |
--connection conn1 |
| 36 |
|
| 37 |
# Begin a transaction, and let it take an exclusive lock on a row |
| 38 |
begin; |
| 39 |
select * from t1 where id = 1 for update; |
| 40 |
|
| 41 |
--connection default |
| 42 |
|
| 43 |
# See TC node. For a 2 data node cluster, it will be on the alive node |
| 44 |
select tc_node_id from ndbinfo.cluster_operations; |
| 45 |
|
| 46 |
# See the transaction state and the lock |
| 47 |
select ndbinfo.cluster_transactions.state, |
| 48 |
ndbinfo.cluster_operations.operation_type, ndbinfo.cluster_operations.state, |
| 49 |
ndbinfo.cluster_locks.mode |
| 50 |
from ndbinfo.cluster_transactions, ndbinfo.cluster_operations, |
| 51 |
ndbinfo.cluster_locks |
| 52 |
where |
| 53 |
ndbinfo.cluster_transactions.transid = ndbinfo.cluster_locks.transid and |
| 54 |
ndbinfo.cluster_locks.transid = ndbinfo.cluster_operations.transid; |
| 55 |
|
| 56 |
--echo Starting the nowait node |
| 57 |
--echo Eventually it will be waiting for the lock above |
| 58 |
--exec $NDB_MGM -e "2 start" >> $NDB_TOOLS_OUTPUT |
| 59 |
|
| 60 |
let $wait=30; |
| 61 |
while ($wait) |
| 62 |
{ |
| 63 |
# TODO : Could theoretically wait for 2 row locks claims on |
| 64 |
# the same rowid to wait for 'stalled CopyFrag' |
| 65 |
--sleep 1 |
| 66 |
#select * from ndbinfo.cluster_locks; |
| 67 |
dec $wait; |
| 68 |
} |
| 69 |
|
| 70 |
--echo Kill the starting node |
| 71 |
--echo This will result in TC takeover |
| 72 |
--exec $NDB_MGM -e "2 ERROR 9999" >> $NDB_TOOLS_OUTPUT |
| 73 |
|
| 74 |
--echo Sleeping 10 s for node failure to be detected |
| 75 |
--echo and handling to start [and finish] |
| 76 |
let $wait=10; |
| 77 |
while ($wait) |
| 78 |
{ |
| 79 |
--sleep 1 |
| 80 |
# select * from ndbinfo.cluster_locks; |
| 81 |
dec $wait; |
| 82 |
} |
| 83 |
|
| 84 |
# Check that we can commit changes into different epochs |
| 85 |
# - implies no GCP stall |
| 86 |
# If TC takeover were stalled then GCP would stall and |
| 87 |
# commits would all have the same epoch. |
| 88 |
# |
| 89 |
|
| 90 |
--disable_query_log |
| 91 |
--disable_result_log |
| 92 |
|
| 93 |
let $first_epoch = 0; |
| 94 |
let $i = 0; |
| 95 |
while($i < 10) |
| 96 |
{ |
| 97 |
eval insert into t1 values ($i+10, $i+10); |
| 98 |
inc $i; |
| 99 |
|
| 100 |
select Variable_value into @epoch |
| 101 |
from performance_schema.global_status |
| 102 |
where Variable_name like 'Ndb_last_commit_epoch_session'; |
| 103 |
|
| 104 |
if ($first_epoch != 0) |
| 105 |
{ |
| 106 |
let $latest_epoch=query_get_value(select @epoch as e, e, 1); |
| 107 |
if ($latest_epoch != $first_epoch) |
| 108 |
{ |
| 109 |
# Break out |
| 110 |
let $i = 10; |
| 111 |
} |
| 112 |
} |
| 113 |
if ($first_epoch == 0) |
| 114 |
{ |
| 115 |
# Initial value |
| 116 |
let $first_epoch=query_get_value(select @epoch as e, e, 1); |
| 117 |
} |
| 118 |
|
| 119 |
--sleep 1 |
| 120 |
} |
| 121 |
|
| 122 |
--enable_result_log |
| 123 |
--enable_query_log |
| 124 |
|
| 125 |
if ($first_epoch != $latest_epoch) |
| 126 |
{ |
| 127 |
--echo Test succeeded, no GCP stall |
| 128 |
} |
| 129 |
|
| 130 |
--echo Release lock to allow data node start to succeed |
| 131 |
--connection conn1 |
| 132 |
rollback; |
| 133 |
|
| 134 |
--connection default |
| 135 |
|
| 136 |
--echo Wait for all data nodes to start |
| 137 |
--exec $NDB_WAITER > $NDB_TOOLS_OUTPUT |
| 138 |
|
| 139 |
--echo Started ok |
| 140 |
|
| 141 |
drop table t1; |
| 142 |
|
| 143 |
--remove_file $NDB_TOOLS_OUTPUT |
| 144 |
|