You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@activemq.apache.org by blue_nekton <je...@gmail.com> on 2019/10/14 20:35:59 UTC
AMQ Artemis High Availability - Use shared store only for lock
Hi,
For small environment to use High availability, it is often difficult to
setup 3-pair cluster to avoid split-brain.
It would be great if artemis master and slave use data replication and use
shared store only for lock and providing HA . So anytime, master/slave
verifies and gets lock before take control of being master again. it will be
like hybrid of replication and shared store. Do we have any feature or
planning for this kind of feature?
if that is not possible, if Artemis can allow the custom locking method to
get lock (instead of just network pinger) like a custom rest call, it will
be easy for work-around for split-brain.
Thanks
--
Sent from: http://activemq.2283324.n4.nabble.com/ActiveMQ-User-f2341805.html
Re: AMQ Artemis High Availability - Use shared store only for lock
Posted by blue_nekton <je...@gmail.com>.
Hi Justin,
Yes.I agree that if we have shared store, we don't need to replication. But
it would good option to have jdbc (shared store) for lock only to avid the
split-brain issue as configuring 3 pair is not suitable for small
environment and 3 pairs add lot of complexity.
Thanks for the detail regarding the raft.
Thanks,
jesu
--
Sent from: http://activemq.2283324.n4.nabble.com/ActiveMQ-User-f2341805.html
Re: AMQ Artemis High Availability - Use shared store only for lock
Posted by Justin Bertram <jb...@apache.org>.
> ...many people use the replication to avoid a spof with shared store and
also perf with NVMe...
Fair enough.
> I dont see why we would discourage anyone in implementing that.
For what it's worth, I didn't mean to discourage anybody from implementing
anything. I was mainly trying to answer his question about whether or not
there were any plans to implement such a scheme. Anybody is certainly
welcome to send a PR for this functionality and further discussion about
the specifics can continue there.
Justin
On Tue, Oct 15, 2019 at 2:34 AM <mi...@me.com.invalid> wrote:
> Justin i disagree many people use the replication to avoid a spof with
> shared store and also perf with NVMe , simply having a shared store just
> for the filelock sounds entirely sensible to me. I dont see why we would
> discourage anyone in implementing that.
> Get Outlook for Android
>
>
>
> On Mon, Oct 14, 2019 at 9:45 PM +0100, "Justin Bertram" <
> jbertram@apache.org> wrote:
>
>
>
>
>
>
>
>
>
>
> I'm not aware of any plans to implement what you suggest here. It seems to
> me that if you want to mitigate the risk of split brain and you have access
> to shared storage then you should just use a shared store configuration
> instead of replication. The main benefit of replication that you don't need
> shared storage.
>
> I think it's more likely that raft [1] integration would be implemented.
>
>
> Justin
>
> [1] https://raft.github.io/
>
> On Mon, Oct 14, 2019 at 3:36 PM blue_nekton wrote:
>
> > Hi,
> >
> > For small environment to use High availability, it is often difficult to
> > setup 3-pair cluster to avoid split-brain.
> > It would be great if artemis master and slave use data replication and
> use
> > shared store only for lock and providing HA . So anytime, master/slave
> > verifies and gets lock before take control of being master again. it will
> > be
> > like hybrid of replication and shared store. Do we have any feature or
> > planning for this kind of feature?
> >
> > if that is not possible, if Artemis can allow the custom locking method
> to
> > get lock (instead of just network pinger) like a custom rest call, it
> will
> > be easy for work-around for split-brain.
> >
> > Thanks
> >
> >
> >
> > --
> > Sent from:
> > http://activemq.2283324.n4.nabble.com/ActiveMQ-User-f2341805.html
> >
>
>
>
>
>
>
Re: AMQ Artemis High Availability - Use shared store only for lock
Posted by mi...@me.com.INVALID.
Justin i disagree many people use the replication to avoid a spof with shared store and also perf with NVMe , simply having a shared store just for the filelock sounds entirely sensible to me. I dont see why we would discourage anyone in implementing that.
Get Outlook for Android
On Mon, Oct 14, 2019 at 9:45 PM +0100, "Justin Bertram" <jb...@apache.org> wrote:
I'm not aware of any plans to implement what you suggest here. It seems to
me that if you want to mitigate the risk of split brain and you have access
to shared storage then you should just use a shared store configuration
instead of replication. The main benefit of replication that you don't need
shared storage.
I think it's more likely that raft [1] integration would be implemented.
Justin
[1] https://raft.github.io/
On Mon, Oct 14, 2019 at 3:36 PM blue_nekton wrote:
> Hi,
>
> For small environment to use High availability, it is often difficult to
> setup 3-pair cluster to avoid split-brain.
> It would be great if artemis master and slave use data replication and use
> shared store only for lock and providing HA . So anytime, master/slave
> verifies and gets lock before take control of being master again. it will
> be
> like hybrid of replication and shared store. Do we have any feature or
> planning for this kind of feature?
>
> if that is not possible, if Artemis can allow the custom locking method to
> get lock (instead of just network pinger) like a custom rest call, it will
> be easy for work-around for split-brain.
>
> Thanks
>
>
>
> --
> Sent from:
> http://activemq.2283324.n4.nabble.com/ActiveMQ-User-f2341805.html
>
Re: AMQ Artemis High Availability - Use shared store only for lock
Posted by Justin Bertram <jb...@apache.org>.
I'm not aware of any plans to implement what you suggest here. It seems to
me that if you want to mitigate the risk of split brain and you have access
to shared storage then you should just use a shared store configuration
instead of replication. The main benefit of replication that you don't need
shared storage.
I think it's more likely that raft [1] integration would be implemented.
Justin
[1] https://raft.github.io/
On Mon, Oct 14, 2019 at 3:36 PM blue_nekton <je...@gmail.com> wrote:
> Hi,
>
> For small environment to use High availability, it is often difficult to
> setup 3-pair cluster to avoid split-brain.
> It would be great if artemis master and slave use data replication and use
> shared store only for lock and providing HA . So anytime, master/slave
> verifies and gets lock before take control of being master again. it will
> be
> like hybrid of replication and shared store. Do we have any feature or
> planning for this kind of feature?
>
> if that is not possible, if Artemis can allow the custom locking method to
> get lock (instead of just network pinger) like a custom rest call, it will
> be easy for work-around for split-brain.
>
> Thanks
>
>
>
> --
> Sent from:
> http://activemq.2283324.n4.nabble.com/ActiveMQ-User-f2341805.html
>