一.什麼是流控（FC）？如何工作？

節點接收寫集並把它們按照全域性順序組織起來，節點將接收到的未應用和提交的事務儲存在接收佇列中，
當這個接收佇列達到一定的大小，將觸發限流；此時節點將暫停複製，節點會先處理接收佇列中的任務。
當接收佇列減小到一個可管理的值後，複製將恢復。

它普遍存在於galera集群系統。

二.流控是發生了什麼，會有哪些全域性值可以觀察到流控？

mysql>  show global status like 'wsrep_flow%';
+----------------------------------+----------------+
| Variable_name                    |
 
 Value          |
+----------------------------------+----------------+
| wsrep_flow_control_paused_ns     | 0              |
| wsrep_flow_control_paused        | 0.000000       |
| wsrep_flow_control_sent          | 0              |
| wsrep_flow_control_recv          | 0              |
| wsrep_flow_control_interval      |
 
 [ 1024, 1024 ] |
| wsrep_flow_control_interval_low  | 1024           |
| wsrep_flow_control_interval_high | 1024           |
| wsrep_flow_control_status        | OFF            |
+----------------------------------+----------------+

wsrep_flow_control_paused_ns

限流發生時，複製同步暫停的時間（各節點都可能出現，不適合作為監控項）

wsrep_flow_control_paused

該狀態值發生變化,含義為從上一次SHOW GLOBAL STATUS命令開始，
限流佔全體同步資料時間的百分比（初始值0.0），理想情況下應該趨近於0.0；
當它比較大的時候（超過0.6），我們需要採取一些手段（新增新節點、刪除慢節點，調高wsrep_slave_threads值）
來改善限流情況

wsrep_flow_control_sent

本地節點發送給叢集的限流事件資訊數量，可以用來當做監控項，來確認哪個節點導致了限流產生

wsrep_flow_control_recv

本地節點收到的叢集限流事件資訊數量

wsrep_flow_control_interval

wsrep_flow_control_interval_low

wsrep_flow_control_interval_high

wsrep_flow_control_status

 

如何進行限流調優？

1. wsrep_slave_threads
The number of threads to use for applying slave write sets.
用於設定讀節點執行寫集的執行緒個數

mysql> show global variables like 'wsrep_slave_threads';
+---------------------+-------+
| Variable_name       | Value |
+---------------------+-------+
| wsrep_slave_threads | 24    |
+---------------------+-------+

預設值1是遠遠不夠的，我們需要根據另外一個狀態值進行調整
公司安裝的PXC叢集，預設該引數值為16（也是不夠的）

2. wsrep_cert_deps_distance
可以並行執行的最高與最低佇列值之間的平均距離
代表可以同時並行執行多少個寫集的操作

mysql> show global status like 'wsrep_cert_deps_distance';
+--------------------------+-----------+
| Variable_name            | Value     |
+--------------------------+-----------+
| wsrep_cert_deps_distance | 95.288623 |
+--------------------------+-----------+

 

我們可以將wsrep_slave_threads的值按照wsrep_cert_deps_distance的值設定

注意：剛做完SST的時候，這個狀態值會非常高，然後緩慢下降，此時該值不具備參考性

mysql> show global status like 'wsrep_cert_deps_distance';
+--------------------------+-------------+
| Variable_name            | Value       |
+--------------------------+-------------+
| wsrep_cert_deps_distance | 6015.952973 |
+--------------------------+-------------+
1 row in set (0.00 sec)
 
mysql> show global status like 'wsrep_cert_deps_distance';
+--------------------------+-------------+
| Variable_name            | Value       |
+--------------------------+-------------+
| wsrep_cert_deps_distance | 5840.756210 |
+--------------------------+-------------+
1 row in set (0.00 sec)
 
mysql> show global status like 'wsrep_cert_deps_distance';
+--------------------------+-------------+
| Variable_name            | Value       |
+--------------------------+-------------+
| wsrep_cert_deps_distance | 5421.252076 |
+--------------------------+-------------+
1 row in set (0.00 sec)

 

其他引數、狀態值

1. wsrep_local_recv_queue_%

mysql> show global status like 'wsrep_local_recv_queue_avg';
+----------------------------+----------+
| Variable_name              | Value    |
+----------------------------+----------+
| wsrep_local_recv_queue_avg | 0.110581 |
+----------------------------+----------+
1 row in set (0.00 sec)

 

When the node returns a value higher than 0.0 it means that the node cannot apply write-sets as fast as it receives them,
which can lead to replication throttling.
簡單地說：這個值高於0.0，說明發生同步延遲，將會引起限流

mysql> show global status like 'wsrep_local_recv_queue_m%';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| wsrep_local_recv_queue_max | 3788  |
| wsrep_local_recv_queue_min | 0     |
+----------------------------+-------+
2 rows in set (0.00 sec)

 

In addition to this status variable, you can also use wsrep_local_recv_queue_max and wsrep_local_recv_queue_min
to see the maximum and minimum sizes the node recorded for the local received queue.