1

Topic: Cluster decision Postgres-XL and what it should be Load Balancer

.. Postgres-XL - System for creation the multi-master of clusters, I work -
- all nodes always contain the actual data in the synchronous mode...
Good afternoon!
Prompt, what advise Load Balancer for cluster decision Postgres-XL (not to confuse with PostgreSql),
If I correctly understood that in the given picture it is a separate component and the decision OR in Postgres-XL already "All inclusive"
Thanks for any councils and answers.
[img=https://www.postgres-xl.org/wp-content/uploads/2014/04/xl_cluster_architecture1.jpg]

2

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Really nobody had experience with implementation Postgres-XL or hesitate to prompt?!?
I to tell the truth and did not understand till now, whether is there built in Load Balancer and the multiplexer of connections (a pool of connections), but looking at the table under this link , where
Like as all inclusive, BUT I and did not discover, as well as where it is adjusted and ???

3

Re: Cluster decision Postgres-XL and what it should be Load Balancer

4

Re: Cluster decision Postgres-XL and what it should be Load Balancer

5

Re: Cluster decision Postgres-XL and what it should be Load Balancer

As the link cannot be interposed, as a result an empty place though I will write  so https://wiki.postgresql.org/wiki/Replic … on_Pooling

6

Re: Cluster decision Postgres-XL and what it should be Load Balancer

vlad-belkin;
Application should be connected to the coordinator. If your coordinator ceases to cope with the traffic - put the second coordinator and simultaneously receive a riddle on to think, as a loading part to transfer to it, without loading the first coordinator - the second delivered just because one did not consult. As will divide requests about two different coordinators - and is load balancer. Can directly be connected on application to the casual coordinator each time, can scatter through something more simple network connections (haproxy for example), can  though at level of routing L3. Or the administrative decision - here this application walks to 1, and here those two - through the second. The task completely similar to equalization of requests of an application-oriented software on equivalent pieces of iron in a cluster.

7

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Melkij;
Thanks, big, from your answer, I understood that  it is necessary to put indirect HAProxy, the impression that at Postgres-XL built in Load Balancer simply for some reason was added

8

Re: Cluster decision Postgres-XL and what it should be Load Balancer

vlad-belkin wrote:

Melkij;
Thanks, big, from your answer, I understood that  it is necessary to put indirect HAProxy, the impression that at Postgres-XL built in Load Balancer

simply for some reason was added
The coordinator also is load balancer as a matter of fact.
Other question that can not suffice ONE coordinator (but here it already it is necessary to look on was specific to your task)
And 1 pgbouncer can easily not suffice (on my experience on 4 on the server it was necessary to put) and 1 haproxy too can not suffice.
Any load balance has a limit how many it in 1 process or in 1 server can pull.
--
Maxim Boguk
the best support PostgreSQL: dataegret.ru

9

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Maxim Boguk;
Warmly I thank for the answer, already as a result there comes clearing.
And as you think, whether there is a sense to deliver on each DataNodu on the coordinator, that is is is admissible 10 DataNod and on each of them the coordinator is installed and accordingly 10 coordinators turns out, and on the input certainly HAProxy will be (in thoughts for fail safety I think them 2nd to deliver).

10

Re: Cluster decision Postgres-XL and what it should be Load Balancer

If I correctly understood, Postgres-XL works through DRBD which not absolutely the DB tool, and is faster the system tool.
If it so how much it is reliable? In case of ruin will collect it it is problematic enough, it seems to me.

11

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Rinka777 wrote:

if I correctly understood, Postgres-XL works through DRBD which not absolutely the DB tool, and is faster the system tool.
If it so how much it is reliable? In case of ruin will collect it it is problematic enough, it seems to me.

Whence such WELL very strange idea about-XL????
--
Maxim Boguk
the best support PostgreSQL: dataegret.ru

12

Re: Cluster decision Postgres-XL and what it should be Load Balancer

vlad-belkin wrote:

Maxim Boguk;
Warmly I thank for the answer, already as a result there comes clearing.
And as you think, whether there is a sense to deliver on each DataNodu on the coordinator, that is is is admissible 10 DataNod and on each of them the coordinator is installed and accordingly 10 coordinators turns out, and on the input certainly HAProxy will be (in thoughts for fail safety I think them 2nd to deliver).

No... To coordinators it is a lot of storage on-XL it is necessary as they besides equalization much that do.
And  too storage it is necessary.
The bad idea.
--
Maxim Boguk
the best support PostgreSQL: dataegret.ru

13

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Maxim Boguk;
, thanks huge, it agree all of 120 %

14

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Maxim Boguk wrote:

it is passed...
Whence such WELL very strange idea about-XL????
--
Maxim Boguk
the best support PostgreSQL: dataegret.ru

Simply assumed. Trial is necessary

15

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Postgres-XL It MPP ! Under OLTP loading it not . Its business milliard aggregates on a cluster to consider. One request. Instead of one million small crud  to fulfill.

16

Re: Cluster decision Postgres-XL and what it should be Load Balancer

Ivan Durak wrote:

Postgres-XL it MPP ! Under OLTP loading it not . Its business milliard aggregates on a cluster to consider. One request. Instead of one million small crud  to fulfill.

-XL just under OLTP it is ground instead of under reports.
It specially did when 1 server did not suffice on TPS on record OLTP big  in Japan in the beginning.
You with greenplum do not confuse it?
--
Maxim Boguk
the best support PostgreSQL: dataegret.ru

17

Re: Cluster decision Postgres-XL and what it should be Load Balancer

With  I do not confuse, but just the principle is very close to .
Well and all requests through the master too not the super decision for OLTP