| 1 |
A client holding a row lock can cause node restart |
| 2 |
(CopyFrag) to stall until the row lock is released. |
| 3 |
|
| 4 |
Check that a stalled CopyFrag does not lead to |
| 5 |
TC takeover stall by : |
| 6 |
- Restarting a node |
| 7 |
- Stalling CopyFrag due to a rowlock |
| 8 |
- Causing starting node failure |
| 9 |
- Checking that TC takeover completes |
| 10 |
(As transactions can be committed in |
| 11 |
different epochs) |
| 12 |
create table t1(id int primary key, val int) engine=ndb; |
| 13 |
select count(*) from t1; |
| 14 |
count(*) |
| 15 |
10 |
| 16 |
begin; |
| 17 |
select * from t1 where id = 1 for update; |
| 18 |
id val |
| 19 |
1 1 |
| 20 |
select tc_node_id from ndbinfo.cluster_operations; |
| 21 |
tc_node_id |
| 22 |
1 |
| 23 |
select ndbinfo.cluster_transactions.state, |
| 24 |
ndbinfo.cluster_operations.operation_type, ndbinfo.cluster_operations.state, |
| 25 |
ndbinfo.cluster_locks.mode |
| 26 |
from ndbinfo.cluster_transactions, ndbinfo.cluster_operations, |
| 27 |
ndbinfo.cluster_locks |
| 28 |
where |
| 29 |
ndbinfo.cluster_transactions.transid = ndbinfo.cluster_locks.transid and |
| 30 |
ndbinfo.cluster_locks.transid = ndbinfo.cluster_operations.transid; |
| 31 |
state operation_type state mode |
| 32 |
Started READ-EX Prepared X |
| 33 |
Starting the nowait node |
| 34 |
Eventually it will be waiting for the lock above |
| 35 |
Kill the starting node |
| 36 |
This will result in TC takeover |
| 37 |
Sleeping 10 s for node failure to be detected |
| 38 |
and handling to start [and finish] |
| 39 |
Test succeeded, no GCP stall |
| 40 |
Release lock to allow data node start to succeed |
| 41 |
rollback; |
| 42 |
Wait for all data nodes to start |
| 43 |
Started ok |
| 44 |
drop table t1; |