Re: Re: BUG #14680: startup process on standby encounter a deadlock of TwoPhaseStateLock when redo 2PC xlog

On Thu, Jun 1, 2017 at 12:11 AM, wangchuanting <wangchuanting(at)huawei(dot)com> wrote:
> we use Postgres-XC+pg10(4 coordinator + 4 datanode(pg 10 as datanode, and 1
> master 1 standby for each datanode)), and benchmark tpcc, there is some
> cross datanode transactions that use 2pc, during testing, we restart the
> cluster, then one datanode standby can not recovery done and hangup with
> TwoPhaseStateLock deadlock.

(Former Postgres-XC maintainer here)
Are you aware of the fact that this is not going to work? Postgres
protocol has been extended between coordinators and datanodes to be
able to push down transaction ID, snapshot as well as timestamps when
running transactions across nodes. So by using a community Postgres as
a datanode you break the global consistency of the cluster. There are
also a couple of other things proper to datanodes.

> and sorry, we reinstall the cluster, and the log is removed, we will try to
> reproduce, but anyway, the code is not right like Michael said.

On that I agree.
--
Michael