postgres-xc-general Mailing List for Postgres-XC

On Mon, May 30, 2011 at 7:34 PM, Lionel Frachon <lio...@gm...>wrote:

> Hi again,
>
> I turned off connection pooling on coordinator (dunno why it sayed
> on), raised the shared_buffers of coordinator, allowed 1000
> connections and the error disappeared.
>
I am not really sure I get the meaning of this, but how did you turn off
pooler on coordinator.
Did you use the parameter persistent_connections?
Connection pooling from coordinator is an automatic feature and you have to
use it if you want to connect from a remote coordinator to backend XC nodes.

You also have to know that it is important to set a limit of connections on
datanodes equal to the sum of max connections on all coordinators.
For example, if your cluster is using 2 coordinator with 20 max connections
each, you may have a maximum of 40 connections to datanodes.
This uses a lot of shared buffer on a node, but typically this maximum
number of connections is never reached thanks to the connection pooling.

Please node also that number of Coordinator <-> Coordinator connections may
also increase if DDL are used from several coordinators.

However, all data is still going on one node (and whatever I could
> choose as primary datanode), with 40 warehouses... any specific syntax
> to load balance warehouses over nodes ?
>

CREATE TABLE foo (column_key type, other_column int) DISTRIBUTE BY
HASH(column_key);
-- 
Michael Paquier
http://michael.otacoo.com

Hi again,

I turned off connection pooling on coordinator (dunno why it sayed
on), raised the shared_buffers of coordinator, allowed 1000
connections and the error disappeared.
However, all data is still going on one node (and whatever I could
choose as primary datanode), with 40 warehouses... any specific syntax
to load balance warehouses over nodes ?

Lionel F.



2011/5/30 Lionel Frachon <lio...@gm...>:
> Hello,
>
> tried again with your compilations options, the datanodes connect
> correctly to the gtm, and the coordinator aswell, but on loadData.sh,
> only one node is loaded, whereas the 2 other ones complains
>
> "STATEMENT:  COMMIT PREPARED 'T709'
> ERROR:  prepared transaction with identifier "T711" does not exist
> STATEMENT:  COMMIT PREPARED 'T711'
> ERROR:  prepared transaction with identifier "T713" does not exist
> STATEMENT:  COMMIT PREPARED 'T713'"
>
> and the loader itself complains aswell
>
> ERROR: Could not commit prepared transaction implicitely
>  Elasped Time(ms): 0.86        Writing record 215000 of 500000
>
> Looks like a lack of comm between them ?
>
> Lionel F.
>
>
>
> 2011/5/30 Lionel Frachon <lio...@gm...>:
>> Ok, testing your compile flags.
>>
>> On install, I'm adding an /etc/ld.so.conf.d/pgxc.conf file pointing to
>> /usr/local/pgsql/lib, then doing an lddconfig to ensure the lib is
>> enabled system-wide. Will add LD_LIBRARY_PATH in user env just to be
>> sure (but I think it already works properly, as if I miss it, the
>> server does not start...)
>>
>> Lionel F.
>>
>>
>>
>> 2011/5/30 Michael Paquier <mic...@gm...>:
>>> Compilation looks to be correct, I myself use this one:
>>> ./configure CFLAGS="-DPGXC -O2" --enable-depend --enable-debug
>>> --disable-rpath --enable-cassert
>>> but even if you define the flag before configure it works correctly.
>>>
>>> On Mon, May 30, 2011 at 4:52 PM, Lionel Frachon <lio...@gm...>
>>> wrote:
>>>>
>>>> Hi Michael,
>>>>
>>>> thanks for your tests and involvment. The error on my side may be a
>>>> compilation, installation or environment problem (at this point, I
>>>> have no clues) , here are the CFLAGS I use :
>>>>
>>>> CFLAGS="-O2"
>>>> and configure :
>>>> "./configure --enable-debug --disable-rpath --enable-depend"
>>>>
>>>> I'm then packing with rpmbuild everything in /usr/local/pgsql. Is that
>>>> a good method (apart from compiling directly on host)?
>>>>
>>>> Are there any env variable (like /etc/security/limits.conf tweaking,
>>>> semaphores or whatever) I should be aware of ?
>>>
>>> LD_LIBRARY_PATH is an environment variable you should set to point to the
>>> correct XC libraries.
>>> It is the only thing that may mess up your settings I think.
>>> --
>>> Michael Paquier
>>> http://michael.otacoo.com
>>>
>>
>

Hello,

tried again with your compilations options, the datanodes connect
correctly to the gtm, and the coordinator aswell, but on loadData.sh,
only one node is loaded, whereas the 2 other ones complains

"STATEMENT:  COMMIT PREPARED 'T709'
ERROR:  prepared transaction with identifier "T711" does not exist
STATEMENT:  COMMIT PREPARED 'T711'
ERROR:  prepared transaction with identifier "T713" does not exist
STATEMENT:  COMMIT PREPARED 'T713'"

and the loader itself complains aswell

ERROR: Could not commit prepared transaction implicitely
  Elasped Time(ms): 0.86        Writing record 215000 of 500000

Looks like a lack of comm between them ?

Lionel F.



2011/5/30 Lionel Frachon <lio...@gm...>:
> Ok, testing your compile flags.
>
> On install, I'm adding an /etc/ld.so.conf.d/pgxc.conf file pointing to
> /usr/local/pgsql/lib, then doing an lddconfig to ensure the lib is
> enabled system-wide. Will add LD_LIBRARY_PATH in user env just to be
> sure (but I think it already works properly, as if I miss it, the
> server does not start...)
>
> Lionel F.
>
>
>
> 2011/5/30 Michael Paquier <mic...@gm...>:
>> Compilation looks to be correct, I myself use this one:
>> ./configure CFLAGS="-DPGXC -O2" --enable-depend --enable-debug
>> --disable-rpath --enable-cassert
>> but even if you define the flag before configure it works correctly.
>>
>> On Mon, May 30, 2011 at 4:52 PM, Lionel Frachon <lio...@gm...>
>> wrote:
>>>
>>> Hi Michael,
>>>
>>> thanks for your tests and involvment. The error on my side may be a
>>> compilation, installation or environment problem (at this point, I
>>> have no clues) , here are the CFLAGS I use :
>>>
>>> CFLAGS="-O2"
>>> and configure :
>>> "./configure --enable-debug --disable-rpath --enable-depend"
>>>
>>> I'm then packing with rpmbuild everything in /usr/local/pgsql. Is that
>>> a good method (apart from compiling directly on host)?
>>>
>>> Are there any env variable (like /etc/security/limits.conf tweaking,
>>> semaphores or whatever) I should be aware of ?
>>
>> LD_LIBRARY_PATH is an environment variable you should set to point to the
>> correct XC libraries.
>> It is the only thing that may mess up your settings I think.
>> --
>> Michael Paquier
>> http://michael.otacoo.com
>>
>

Ok, testing your compile flags.

On install, I'm adding an /etc/ld.so.conf.d/pgxc.conf file pointing to
/usr/local/pgsql/lib, then doing an lddconfig to ensure the lib is
enabled system-wide. Will add LD_LIBRARY_PATH in user env just to be
sure (but I think it already works properly, as if I miss it, the
server does not start...)

Lionel F.



2011/5/30 Michael Paquier <mic...@gm...>:
> Compilation looks to be correct, I myself use this one:
> ./configure CFLAGS="-DPGXC -O2" --enable-depend --enable-debug
> --disable-rpath --enable-cassert
> but even if you define the flag before configure it works correctly.
>
> On Mon, May 30, 2011 at 4:52 PM, Lionel Frachon <lio...@gm...>
> wrote:
>>
>> Hi Michael,
>>
>> thanks for your tests and involvment. The error on my side may be a
>> compilation, installation or environment problem (at this point, I
>> have no clues) , here are the CFLAGS I use :
>>
>> CFLAGS="-O2"
>> and configure :
>> "./configure --enable-debug --disable-rpath --enable-depend"
>>
>> I'm then packing with rpmbuild everything in /usr/local/pgsql. Is that
>> a good method (apart from compiling directly on host)?
>>
>> Are there any env variable (like /etc/security/limits.conf tweaking,
>> semaphores or whatever) I should be aware of ?
>
> LD_LIBRARY_PATH is an environment variable you should set to point to the
> correct XC libraries.
> It is the only thing that may mess up your settings I think.
> --
> Michael Paquier
> http://michael.otacoo.com
>

Compilation looks to be correct, I myself use this one:
./configure CFLAGS="-DPGXC -O2" --enable-depend --enable-debug
--disable-rpath --enable-cassert
but even if you define the flag before configure it works correctly.

On Mon, May 30, 2011 at 4:52 PM, Lionel Frachon <lio...@gm...>wrote:

> Hi Michael,
>
> thanks for your tests and involvment. The error on my side may be a
> compilation, installation or environment problem (at this point, I
> have no clues) , here are the CFLAGS I use :
>
> CFLAGS="-O2"
> and configure :
> "./configure --enable-debug --disable-rpath --enable-depend"
>
> I'm then packing with rpmbuild everything in /usr/local/pgsql. Is that
> a good method (apart from compiling directly on host)?
>
> Are there any env variable (like /etc/security/limits.conf tweaking,
> semaphores or whatever) I should be aware of ?
>
LD_LIBRARY_PATH is an environment variable you should set to point to the
correct XC libraries.
It is the only thing that may mess up your settings I think.
-- 
Michael Paquier
http://michael.otacoo.com

Hi Michael,

thanks for your tests and involvment. The error on my side may be a
compilation, installation or environment problem (at this point, I
have no clues) , here are the CFLAGS I use :

CFLAGS="-O2"
and configure :
"./configure --enable-debug --disable-rpath --enable-depend"

I'm then packing with rpmbuild everything in /usr/local/pgsql. Is that
a good method (apart from compiling directly on host)?

Are there any env variable (like /etc/security/limits.conf tweaking,
semaphores or whatever) I should be aware of ?

Thanks.

Kind regards

Lionel F.



2011/5/30 Michael Paquier <mic...@gm...>:
> Hi,
>
> I tested the benchmark with latest head code, and it looks to work on my
> side without crashing.
> Respecting what you said in your previous email and what is written in
> README.
> #Create database
> createdb test
> #Create tables
> ./runSQL.sh postgres.properties sqlTableCreates
> #Data generation
> ./loadData.sh postgres.properties numWarehouses 1
> #Index creation
> ./runSQL.sh postgres.properties sqlIndexCreates
> #To launch test terminal
> ./runBenchmark.sh postgres.properties
>
> With the following properties
> driver=org.postgresql.Driver
> conn=jdbc:postgresql://localhost:5432/test
> user=******
> password=*****
>
> I found in the logs some errors such as:
> LOG:  execute <unnamed>: SELECT c_discount, c_last, c_credit, w_tax  FROM
> customer, warehouse WHERE w_id = $1 AND w_id = c_w_id AND c_d_id = $2 AND
> c_id = $3
> DETAIL:  parameters: $1 = '1', $2 = '10', $3 = '1549'
> ERROR:  bind message supplies 3 parameters, but prepared statement ""
> requires 0
>
> But this is expected as XC does not yet manage multi-prepared queries and
> BenchmarkSQL looks to use it.

Yes, I've seen it and saw your warning; as you said, it should be
however sufficient to get some correct results regarding
perfs/scalability.

> Btw, this is enough to get some results.
>
> Also, I am wondering about this error you got:
> LOG:  could not create IPv6 socket: Address family not supported by protocol
> I am using here only IPv4 protocol.


> --
> Michael Paquier
> http://michael.otacoo.com
>

There is also something I noticed with the way data is distributed among
nodes.
Now all the table are using the warehouse ID (this looks exactly like TPC-C
benchmark test) as a distribution key.
If you let the benchmark running like that by default with only 1 warehouse,
you will finish with all your data located on the same node, resulting in
bad performance.

If you are looking for performance tests, you may need to use far more
warehouses (don't know, 20~100) or a different distribution key, but
regarding all the tables warehouse ID is definitely the best distribution
key. So, I would recommend to use a high number of warehouses and distribute
all the tables with warehouse ID as distribution key.
item table does not use warehouse ID, so it is better to replicate this
table among nodes.

On Mon, May 30, 2011 at 10:50 AM, Michael Paquier <mic...@gm...
> wrote:

> Hi,
>
> I tested the benchmark with latest head code, and it looks to work on my
> side without crashing.
> Respecting what you said in your previous email and what is written in
> README.
> #Create database
> createdb test
> #Create tables
> ./runSQL.sh postgres.properties sqlTableCreates
> #Data generation
> ./loadData.sh postgres.properties numWarehouses 1
> #Index creation
> ./runSQL.sh postgres.properties sqlIndexCreates
> #To launch test terminal
> ./runBenchmark.sh postgres.properties
>
> With the following properties
> driver=org.postgresql.Driver
> conn=jdbc:postgresql://localhost:5432/test
>
> user=******
> password=*****
>
> I found in the logs some errors such as:
> LOG:  execute <unnamed>: SELECT c_discount, c_last, c_credit, w_tax  FROM
> customer, warehouse WHERE w_id = $1 AND w_id = c_w_id AND c_d_id = $2 AND
> c_id = $3
> DETAIL:  parameters: $1 = '1', $2 = '10', $3 = '1549'
> ERROR:  bind message supplies 3 parameters, but prepared statement ""
> requires 0
>
> But this is expected as XC does not yet manage multi-prepared queries and
> BenchmarkSQL looks to use it.
> Btw, this is enough to get some results.
>
> Also, I am wondering about this error you got:
>
> LOG:  could not create IPv6 socket: Address family not supported by
> protocol
> I am using here only IPv4 protocol.
> --
> Michael Paquier
> http://michael.otacoo.com
>


Regards,
-- 
Michael Paquier
http://michael.otacoo.com

Hi,

I tested the benchmark with latest head code, and it looks to work on my
side without crashing.
Respecting what you said in your previous email and what is written in
README.
#Create database
createdb test
#Create tables
./runSQL.sh postgres.properties sqlTableCreates
#Data generation
./loadData.sh postgres.properties numWarehouses 1
#Index creation
./runSQL.sh postgres.properties sqlIndexCreates
#To launch test terminal
./runBenchmark.sh postgres.properties

With the following properties
driver=org.postgresql.Driver
conn=jdbc:postgresql://localhost:5432/test

user=******
password=*****

I found in the logs some errors such as:
LOG:  execute <unnamed>: SELECT c_discount, c_last, c_credit, w_tax  FROM
customer, warehouse WHERE w_id = $1 AND w_id = c_w_id AND c_d_id = $2 AND
c_id = $3
DETAIL:  parameters: $1 = '1', $2 = '10', $3 = '1549'
ERROR:  bind message supplies 3 parameters, but prepared statement ""
requires 0

But this is expected as XC does not yet manage multi-prepared queries and
BenchmarkSQL looks to use it.
Btw, this is enough to get some results.

Also, I am wondering about this error you got:
LOG:  could not create IPv6 socket: Address family not supported by protocol
I am using here only IPv4 protocol.
-- 
Michael Paquier
http://michael.otacoo.com