Description:
This could be in documentation category, but this is more a feature request than a documentation bug
According to https://dev.mysql.com/doc/refman/5.7/en/group-replication-replication-group-member-stats.h..., I should get :
mysql> select * from performance_schema.replication_group_member_stats\G;
*************************** 1. row ***************************
CHANNEL_NAME: group_replication_applier
VIEW_ID: 15517946505895476:13
MEMBER_ID: 80c2611b-3f4f-11e9-bc27-0242ac120002
COUNT_TRANSACTIONS_IN_QUEUE: 0
COUNT_TRANSACTIONS_CHECKED: 166245
COUNT_CONFLICTS_DETECTED: 22517
COUNT_TRANSACTIONS_ROWS_VALIDATING: 112532
TRANSACTIONS_COMMITTED_ALL_MEMBERS: 80c2611b-3f4f-11e9-bc27-0242ac120002:1-12,
8bdecced-3f4f-11e9-bc27-0242ac120002:1-118049:1000013-1003006
LAST_CONFLICT_FREE_TRANSACTION: 8bdecced-3f4f-11e9-bc27-0242ac120002:228878
COUNT_TRANSACTIONS_REMOTE_IN_APPLIER_QUEUE: 0
COUNT_TRANSACTIONS_REMOTE_APPLIED: 0
COUNT_TRANSACTIONS_LOCAL_PROPOSED: 0
COUNT_TRANSACTIONS_LOCAL_ROLLBACK: 0
1 row in set (0.00 sec)
Instead, I get (missing last 4 columns) on mysql 5.7.25 :
mysql> select * from performance_schema.replication_group_member_stats\G;
*************************** 1. row ***************************
CHANNEL_NAME: group_replication_applier
VIEW_ID: 15517946505895476:13
MEMBER_ID: 80c2611b-3f4f-11e9-bc27-0242ac120002
COUNT_TRANSACTIONS_IN_QUEUE: 0
COUNT_TRANSACTIONS_CHECKED: 166245
COUNT_CONFLICTS_DETECTED: 22517
COUNT_TRANSACTIONS_ROWS_VALIDATING: 112532
TRANSACTIONS_COMMITTED_ALL_MEMBERS: 80c2611b-3f4f-11e9-bc27-0242ac120002:1-12,
8bdecced-3f4f-11e9-bc27-0242ac120002:1-118049:1000013-1003006
LAST_CONFLICT_FREE_TRANSACTION: 8bdecced-3f4f-11e9-bc27-0242ac120002:228878
1 row in set (0.00 sec)
We have a cluster with 3 nodes (single-primary), one crashed, a new primary was elected.
Since our crashed node is in recovery, we experience a lot of rollbacks (during COMMIT with error 3101) and I suspect that the flow control has been triggered
Problem is, we know flow control happen when a queue exceeds 25k waiting transactions, but we cannot see queue sizes despite what is written is documentation
We cannot see when flow control is activated
We cannot see how late is a node either, because of lacking queue size stats.
Is it a bug ? A documentation failure ?
Anyway, we could use some logs/table to observe flow control behaviour
Is a backport of these 4 missing columns planned ?
How to repeat:
Use dbdeployer to deploy a GR cluster
https://github.com/datacharmer/dbdeployer
dbdeployer deploy replication -t=group 5.7.25
./sandboxes/group_msb_5_7_25/use_all 'select * from performance_schema.replication_group_member_stats\G'
Suggested fix:
- Backport of 8.0.14 performance_schema tables
- Fix documentation
- Flow Control logs feature
