Re: Replication - Mailing list pgsql-admin

From Markus Schiltknecht
Subject Re: Replication
Date
Msg-id 459CEC50.6030407@bluegap.ch
Whole thread Raw
In response to Replication  ("anorganic anorganic" <anorganicphc@gmail.com>)
List pgsql-admin
Hi,

[ please keep CCing to the list (reply all), as this certainly isn't a
personal discussion and could help others. ]

anorganic anorganic wrote:
> Hello,
>
> "I don't quite follow what you mean here. I assume you mean rows, not
> columns. Then probably you mean something like what we call data
> partitioning, where each server is a master for a certain subset of the
> data."
>
> I think granurality, some system allow replicate only all db, all table,
> some allow to choose only 4 colums from one table 2 from another and
> replicated only these parts of db. I can add example.
> it's not  partitioning.

Uhm.. I'd still call it partitioning. Although, I'm not exactly sure why
you want to replicate only a partial row. I suspect you will have to
split into two tables and have one table replicated while not
replicating the other.

> "Interesting... where exactly is it to short for you? What else would you
> expect?"
>
> ok may be i'm false but i think this all is not common features of postgre
> and it's necessary use system as slony.
> i want know what i must use, how i can make and detail description of these
> solution. What's happen when connection crash, when server shutdown etc.

Right, but as we only have external solutions, I think it makes sense to
have them document how exactly they work and behave.

> "Hm. If that really is a problem, think again about sync vs. async. In
> async replication, there is a delay by definition. But again, I'm not
> sure what Slony can do to minimize that delay."
>
> yeah may be i am on bad adress but may be somebody here will say me here i
> used this program fort his solution. Or this problem  have no solution on
> postgre.

I fear replication is such a wide field with that many options, you
probably can't get away with a general approach. I especially think so
because I consider your wishes to be quite exotic for two reasons:

  * realtime, as fast as possible, but certainly not sync
    (You are aware that async, no matter how fast, can lead to conflicts,
     aren't you?)
  * replicating only parts of a row

> Async, sync i dont need sync, but i need send change fast as possible. i
> don't want sync. But when i change col on s1 i wnat send change right now,
> no waiting for another 5 changes. It must be quite realtime we have low
> traffic between servers there is not necessary make some optimilaztion as
> sending in packages.

I'd be surprised if Slony couldn't be configured to send changes
immediately (i.e. not waiting for more transactions before sending a
package). But again, the Slony guys may know better.

Regards

Markus

pgsql-admin by date:

Previous
From: Josef Springer
Date:
Subject: I want to import a database from a customer
Next
From: "Shoaib Mir"
Date:
Subject: Re: I want to import a database from a customer