[Bucardo-general] Star topology - master-master async replication with bucardo problem

[Chris Keane]

Hi Chregan,

We use a very similar setup. In our case, we run an independent bucardo
process on each event server which is responsible for keeping just that
event server in sync with the main server. In each sync on each client the
tables on the main server have makedelta set, so when one client updates a
table on the main server, the update then triggers updates on all the other
clients (including disconnected ones which later catch up).

This works really well in theory and in practice works very well up to
about 4 simultaneous clients connected... well, it seems to be a processing
limit on the main server since each client that is concurrently connected
seems to take about 1 CPU for processing, so 4 clients maxes out our 4 CPU
AWS RDS instance. We tested that theory by temporarily increasing it to 32
CPUs, and it did seem to need about 1 CPU per connected client.

Chris.


On Fri, Sep 4, 2015 at 6:44 PM Chregan O'Flynn <chregano at gmail.com> wrote:

> Hey everyone
>
> I am trying to setup Burcardo in the following use case:
> The solution is for music events wherein the internet connect is very
> unreliable. Each event will have a mini server running a clone of the main
> server API+DB and will need to sync the database whenever connected. There
> could be 15-20 events running at any given time, so I would need to be able
> to sync between all of them on a master-master basis but without needing
> any servers (other than the main server) to be online. (so a star topology
> basically)
>
> After some trial and error, I managed to get a sync working between 4
> servers - the main server (always online hosted on a Digital Ocean droplet)
> and three “event” servers.
> The problem was that if I made a change to an event server, the main
> server would update, but none of the other event servers would update.
> Updating the main server would update all event servers though.
> I also tried to create a sync group to include all of the servers (not
> ideal because I don’t know how many event servers I will have), but if one
> went offline, the sync stopped completely.
>
> I have also tried to set makedelta=on for all the DBs, but that doesnt
> seem to help
>
> Any help here would be greatly appreciated!
>
> thanks in advance
>
> take care
> Chregan
> _______________________________________________
> Bucardo-general mailing list
> Bucardo-general at bucardo.org
> https://mail.endcrypt.com/mailman/listinfo/bucardo-general
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.endcrypt.com/pipermail/bucardo-general/attachments/20150905/d04d3732/attachment.html>