I am not sure how parallel replication impacts SBM calculation but the computation is obviously wrong when using parallel threads.
SBM should be (imho) calculated by comparing the timestamp of the currently executed event in the binary log to the current datetime. I understand that there might be issues when executing events in parallel however that should work for most cases.
simple reproducer:
|
<pre>
|
#stop perfectly working slave
|
STOP SLAVE;
|
SELECT SLEEP(10);
|
START SLAVE;
|
SHOW PROCESSLIST\G
|
MariaDB(db-02)[(none)]> SHOW PROCESSLIST\G
|
*************************** 1. row ***************************
|
Id: 3
|
User: system user
|
Host:
|
db: NULL
|
Command: Connect
|
Time: 10
|
State: Waiting for work from SQL thread
|
Info: NULL
|
Progress: 0.000
|
*************************** 2. row ***************************
|
Id: 4
|
User: system user
|
Host:
|
db: NULL
|
Command: Connect
|
Time: 10
|
State: Waiting for work from SQL thread
|
Info: NULL
|
Progress: 0.000
|
*************************** 3. row ***************************
|
SHOW SLAVE STATUS\G
|
|
Seconds_Behind_Master: 0
|
SBM first reports 0 then will increase monotonically from this point until it reaches the actual delay value (whatever is reported in "Time" for a worker thread in the processlist, for example).
Bug
Major
