Hi! I have two server with group replication.
Percona-Server-server-57-5.7.20-19.1.el7.x86_64
Percona-Server-client-57-5.7.20-19.1.el7.x86_64
Percona-Server-shared-57-5.7.20-19.1.el7.x86_64
Percona-Server-devel-57-5.7.20-19.1.el7.x86_64
Percona-Server-shared-compat-57-5.7.23-23.1.el7.x86_64
kernel: 3.10.0-693.11.1.el7.x86_64
When sysbench run (write load), query is random freezes on “Sending data”. Sometimes more than 2 seconds.
select * from performance_schema.replication_group_member_stats;
CREATE TABLE replication_group_member_stats (
CHANNEL_NAME char(64) NOT NULL,
VIEW_ID char(60) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
MEMBER_ID char(36) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
COUNT_TRANSACTIONS_IN_QUEUE bigint(20) unsigned NOT NULL,
COUNT_TRANSACTIONS_CHECKED bigint(20) unsigned NOT NULL,
COUNT_CONFLICTS_DETECTED bigint(20) unsigned NOT NULL,
COUNT_TRANSACTIONS_ROWS_VALIDATING bigint(20) unsigned NOT NULL,
TRANSACTIONS_COMMITTED_ALL_MEMBERS longtext NOT NULL,
LAST_CONFLICT_FREE_TRANSACTION text NOT NULL
) ENGINE=PERFORMANCE_SCHEMA DEFAULT CHARSET=utf8
Problem in about one of 10 requests.
On test cluster I add to systemd unit CPUQuota=20%(this exacerbated the problem), and periodic run
SET profiling = 1;
select * from performance_schema.replication_group_member_stats\G
SHOW PROFILE SOURCE FOR QUERY 1;
Please explain why this can happen. I have 16 core on server, many core not utilized, but at the moment of the problem, one or two cores rest against 100% utilization for a few seconds.