How to beat Percona XtraDB Cluster?

Question

Percona

Hello. The situation is as follows: there is a cluster of 3 machines on Percona XtraDB Cluster 5.6. Initially, it was configured and tested, the data was synchronized normally, the working databases were flooded and released for production. At first everything was fine, but by the morning on the first node in the processlist there were a bunch of requests under max connection in the wsrep in pre-commit stage during long transaction status.
I turned off all the servers, turned them on in order, everything seemed to work, but then it turned out that the 1st server worked in some kind of fantasy world of its own, and the other two in its own. Info from 1st server:

mysql> show global status like '%wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name            | Value                                |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id    | 3                                    |
| wsrep_cluster_size       | 3                                    |
| wsrep_cluster_state_uuid | 2da7c277-596b-11e6-9df7-da34104b031c |
| wsrep_cluster_status     | Primary                              |
+--------------------------+--------------------------------------+

show global status like '%wsrep_ready%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| wsrep_ready   | ON    |
+---------------+-------+

Second:

mysql> show global status like '%wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name            | Value                                |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id    | 2                                    |
| wsrep_cluster_size       | 2                                    |
| wsrep_cluster_state_uuid | 2da7c277-596b-11e6-9df7-da34104b031c |
| wsrep_cluster_status     | Primary                              |
+--------------------------+--------------------------------------+

show global status like '%wsrep_ready%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| wsrep_ready   | OFF   |
+---------------+-------+

The third:

show global status like '%wsrep_cluster%';
+--------------------------+--------------------------------------+
| Variable_name            | Value                                |
+--------------------------+--------------------------------------+
| wsrep_cluster_conf_id    | 3                                    |
| wsrep_cluster_size       | 3                                    |
| wsrep_cluster_state_uuid | 2da7c277-596b-11e6-9df7-da34104b031c |
| wsrep_cluster_status     | Primary                              |
+--------------------------+--------------------------------------+

show global status like '%wsrep_ready%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| wsrep_ready   | OFF   |
+---------------+-------+

Naturally, on the second and third, when trying any sql query, it issues:

ERROR 1047 (08S01): WSREP has not yet prepared node for application use

I tried to turn off all servers, then on the first one in wsrep_cluster_address specify only gcomm:// and run through /etc/init.d/mysql bootstrap-pxc and then turn on the other two servers, but it did not help. At the same time, according to the logs, everything seems to be in order: paste2.org/Dt31XZ8N
Please tell me how to solve this problem?