> When MASTER_GTID_WAIT is called, it would seem that the domain-server-seq
> transactions should be queued on the next available parallel slave thread so
> it has the potential to be processed before the master_gtid_wait timout
> seems more likely.
Sorry, I don't understand how this would work.
Transactions are always queued in binlog order, as they are read. Normally,
all transactions will always be queued, unless the slave is lacking severely
behind the master (controlled by --slave-parallel-max-queued). And all slave
threads are always available for queueing.
Generally, in current parallel replication it is expected that the amount of
parallelism available is limited, so there does not seem to be much
opportunity for selectively prioritising certain transactions in the
scheduling.
If I've misunderstood, please reopen the bug with a clarification of what was
intended.
Thanks,
Minor
> When MASTER_GTID_WAIT is called, it would seem that the domain-server-seq
> transactions should be queued on the next available parallel slave thread so
> it has the potential to be processed before the master_gtid_wait timout
> seems more likely.
Sorry, I don't understand how this would work.
Transactions are always queued in binlog order, as they are read. Normally,
all transactions will always be queued, unless the slave is lacking severely
behind the master (controlled by --slave-parallel-max-queued). And all slave
threads are always available for queueing.
Generally, in current parallel replication it is expected that the amount of
parallelism available is limited, so there does not seem to be much
opportunity for selectively prioritising certain transactions in the
scheduling.
If I've misunderstood, please reopen the bug with a clarification of what was
intended.
Thanks,