LargeSelect exists specifically so that you can read through a large
number of records in chunks of size memoryPageLimit.
If you want to read all of the records into memory at once you would not
use LargeSelect. If all the records will not fit into memory then you
use LargeSelect to automatically break up the query result using
offset/limit queries (which is not supported by Torque for all RDBMSs).
If you are presenting data to users you use pageSize to determine the
number of records at a time that you want to pull from those available
(memoryPageLimit). If you are not presenting the data to a user you
should set pageSize to the same value as memoryPageLimit.
LargeSelect is essentially a buffering mechanism:
- memoryPageLimit is a buffer of records from the overall query result -
LargeSelect pulls in more data using offset and limit whenever data
beyond what it has in memory is requested.
- pageSize determines how much of the buffered data you want to access
in each request.
When data within the memoryPageBuffer is requested it can be provided
without executing a further query. If data outside of the
memoryPageBuffer is requested the existing data is discarded and a new
one that contains the requested data is filled by querying for a subset
for the full query results.
So to run through your 185K records you simply keep retrieving them a
page at a time using getNextResults() with a memoryPageLimit well below
185K. If this is not working for you then I would suggest that the
offset/limit processing for the RDBMS you are using may not be fully
implemented, may have a bug, or may be falling back to processing that
actually retrieves all of the records and discards those outside of the
offset/limit (this would totally defeat the purpose of using LargeSelect).
Have you run the LargeSelect tests against your RDBMS? You should do so
and ensure that the offset/limit is fully implemented.
Scott
YannickR wrote:
Thank you Scott for your great explanation of LargeSelect.
I understand that the goal of it is to view a portion of the data at a time
: (PageSize * MemoryPageLimit) records. We already have a program that use
Torque objects for synchronization on a 10 second thread basis and map
objects from one side to the other side. I made some modifications to it so
we can launch it in batch mode, because too much data was accumulated on one
database side (around 185 000 records).
Now I need to read them all with LargeSelect. Like I said, it is not reading
all records as expected, unless I specified PageSize and MemoryPageLimit
values high enough to cover the total number of records. As you suggest, I
would use 435*435 (189 235) in order to read them all with success... This
would surely generate a Heap memory exception. What should I do if i do not
know exactly the volume of data to process ? Do I need to calculate it
dynamically according to total record count ?
With your example, if I set a PageSize of 100, MemoryPageLimit will equal 5
:
(100 * 5 < 250) ? (250 / 100 ) : 5
I will then miss most of the data, it will only read 1000 records on 185
000...
You said that it is possible to configure LargeSelect to pull in just one
page of data at a time.
How can I do that ? That would resolve my issue, since it would stop reading
all records in one shot while instantiating LargeSelect.
Thanks for your help ! It is appreciated.
Regards,
Yannick
seade wrote:
I hadn't been paying close attention to this thread, but it seems that
the a couple of points are somehow being missed:
1. If you have a large amount of data, how much of it is the user
actually going to practically be able to view. It is not such a good
idea to provide the user with a means of browsing through a million
records - they will never do so. You need to provide the ability to
filter the data down to a practical number of records that the user can
then view.
2. If you run a query that pulls in one million records you are more
than likely going to run out of memory. This is in fact the problem
that LargeSelect seeks to address. Instead of pulling in all of the
records, in instead pulls in a subset of these that can then be
presented a page at a time. While you can configure LargeSelect to pull
in just one page of data at a time this may be at odds with the
complexity of the query and the amount of time it takes to execute. To
counter this, LargeSelect provides the ability to cache a configurable
number of pages worth of data - this way the user can at least browse
through a few pages of data without triggering an expensive query for
every hit. It is up to you to determine how much data will be presented
on any given page and how many pages of data to read ahead - make the
values too large and you will still run out of memory.
I am a heavy user of LargeSelect. I use a pageSize of between 10 and
100 (as selected by the user) and a memoryPageLimit of:
(pageSize * 5 < 250) ? (250 / pageSize ) : 5
And everything works nicely.
I have no idea whether or not Torque-84 works, but it is unlikely to be
committed without the addition of test cases and even then it will
require a committer to take the time to ensure it behaves correctly.
From what Greg is saying, for MS SQL Torque-84 should not be required,
but other changes at svn trunk are.
LargeSelect is about presenting data to users - as I said above, a user
is never going to look at 1 million records. You on the other hand are
working on database synchronization, so I assume you are working through
a large number of records (8000 was mentioned somewhere) that are not
actually being presented to users. The first question I would ask is
whether or not you need to instantiate the data as Torque objects - i.e.
could you get by with using native SQL (most likely quicker when dealing
with bulk data like this). That said, there should be no reason why you
cannot use LargeSelect for your purposes - i.e. to limit the number of
records in memory at any given time. To do this I would set pageSize
and memoryPageLimit to the same value, a value that maximises throughput
by balancing the trade-off between memory use and query execution time.
HTH,
Scott
YannickR wrote:
I checked out the current CVS head (without Torque-84 patch) and did some
tests in order to better explain what is happening. It seems that
PageSize *
PageMemoryLimit need to cover the amount of records. For example if you
have
8000 records to read and PageMemoryLimit is set to the default of 5,
PageSize would have a minimum value of 8000/5/2 = 800. If a value lower
than
800 is used, some records won't be read... When you have a huge amount as
185 000 records to read, the limit will be memory : 185 000/5/2 = 18 500
minimum. That means 92 500 records in memory at one time...
To reproduce the situation, LargeSelect unit tests should not use
PageSize
and PageMemoryLimit in order to fill Authors. By doing so, all records
are
covered and the comportment that I just described won't be reproducible.
Anyway, is 9*9 records a "Large" Select test ?
As I already said, when I use Torque-84 patch,
LargeSelect.getNextResultsAvailable() will always return true, so reading
in
an infinite loop ;-(
Could someone clarify, please ?
Greg Monroe wrote:
As a quick aside, it would be much easier to follow your
messages, if your embedded comments where not prefixed
with one or more >'s. Makes it real hard to see what
are new comments and what are old.
That said, I tested the current CVS head (which is 99.9%
final release for 3.3) against MS SQL 2000 just last
week. In order for this to pass all the Limit /
LargeSelect tests in the test project, I committed some
changes to the DBSybase class (which MS SQL extends).
So, try checking out the latest from CVS and using this.
This should work with MS 2005. The support is generic
across all MS SQL versions, so it is "psuedo" support that
requires more data than requested to be read and "trimmed"
down.
-----Original Message-----
From: YannickR [mailto:[EMAIL PROTECTED]
Sent: Friday, October 26, 2007 12:26 PM
To: [email protected]
Subject: Re: LargeSelect example ?
Is the patch working or not ? The status on
https://issues.apache.org/jira/browse/TORQUE-84?page=com.atlassian.jir
a.plugin.system.issuetabpanels:all-tabpanel
seems to be unresolved...
Could someone help me on that one ?
Can I still use LargeSelect with MSSQL 2005 ?
Regards,
Yannick Richard
DukeCE Privacy Statement:
Please be advised that this e-mail and any files transmitted with
it are confidential communication or may otherwise be privileged or
confidential and are intended solely for the individual or entity
to whom they are addressed. If you are not the intended recipient
you may not rely on the contents of this email or any attachments,
and we ask that you please not read, copy or retransmit this
communication, but reply to the sender and destroy the email, its
contents, and all copies thereof immediately. Any unauthorized
dissemination, distribution or copying of this communication is
strictly prohibited.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
