Simple insert into table hanging

AParm

Hi there,

I have a query that joins 2 tables, 1 table has 2 million records and other 1 million records, the query returns 1 record in less than 1 second, both the 2 tables being joined are accessed across a database link, despite the query running in less than a second, if I do a simple insert using the query it just hangs, and I don't understand why. The table is not locked, I have tried dropping the table before the insert and recreating it, to ensure there are no locks.

If I look at the wait events all  I a can see this

I am running 12c on Windows, the database link is to database in another pluggable database in the same container.

SQL*Net message from dblink

AParm

Ok, I think I've worked out why it was hanging, the table had a default value of SYSDATE for a date column, which isn't in the insert select statement, I've removed the default and now it inserts on 0.85 sec.

If I put sysdate in the select statement...

insert table(

..

..

DATE_COLUMN)

select

..

..

SYSDATE

from  TABLE;

it chooses the first plan, and it hangs.

John Thorton

How do I ask a question on the forums?

We can't say what you do wrong since you decided to NOT show us exactly what you do & how Oracle responds.

AndrewSayer

Amarprem wrote:Hi there,I have a query that joins 2 tables, 1 table has 2 million records and other 1 million records, the query returns 1 record in less than 1 second, both the 2 tables being joined are accessed across a database link, despite the query running in less than a second, if I do a simple insert using the query it just hangs, and I don't understand why. The table is not locked, I have tried dropping the table before the insert and recreating it, to ensure there are no locks.If I look at the wait events all I a can see thisI am running 12c on Windows, the database link is to database in another pluggable database in the same container.SQL*Net message from dblink

Does query only return 1 row or was that all that you fetched?

What is being waited on at the remote DB? Is it using the same execution plan to execute the select part of the insert than if it was just doing the select?

You *may* be better off writing pl/sql to do the select statement and insert the result into your target table (if it is just one row, I wouldn't be too concerned with the plsql overhead)

Mike Kutz

Amarprem wrote:Hi there,despite the query running in less than a second..

How did you measure this?

MK

AParm

I'm in sql developer.

AParm

The query as run today returns 1 row, tomorrow it might return more, if the select takes a second to run, why does the insert of one record hanging, what else is it doing?

AndrewSayer

Amarprem wrote:The query as run today returns 1 row, tomorrow it might return more, if the select takes a second to run, why does the insert of one record hanging, what else is it doing?

Consider answering the questions asked of you.

It is doing something different, perhaps your simple query is not as simple as you think, it does operate on remote objects after all.

Oracle can use a different execution plan for an insert statement than the select on it's own. You have the power of finding out if this is the case. You can view the sql that is being remotely executed from your local statement by looking at it's execution plan. You can view the execution plan of the remote sqls by connecting to the remote DB and checking there.

If you don't know how to do what's asked of you then have a google first and if that doesn't help then ask here. Don't just avoid the question.

John Thorton

Amarprem wrote:The query as run today returns 1 row, tomorrow it might return more, if the select takes a second to run, why does the insert of one record hanging, what else is it doing?

It is waiting for more input

Mustafa_KALAYCI

you said table is not locked, what if there is a lock on a row ? did you select v$session for blocking_session and event fields  while your insert statement was hanging ? do that and post the result please.

AParm

I see the execution plans are different, for the first it just has TABLE FULL ACCESS for each of the tables, do I presume it does the join on the remote database, is that right?

The insert statement, has REMOTE for the 2 table, I presume, it fetching all the data for the 2 tables and doing the join locally, is that right? And if that is the case how do I get it to do the select on the remote server?

This is a pluggable database fetching data from another pluggable database, so it's all in the same server process space, why would a remote call be any different from a local call?

AndrewSayer

Amarprem wrote:I see the execution plans are different, for the first it just has TABLE FULL ACCESS for each of the tables, do I presume it does the join on the remote database, is that right?The insert statement, has REMOTE for the 2 table, I presume, it fetching all the data for the 2 tables and doing the join locally, is that right? And if that is the case how do I get it to do the select on the remote server?This is a pluggable database fetching data from another pluggable database, so it's all in the same server process space, why would a remote call be any different from a local call?

Don't just tell us, SHOW us. We can only interpret what we can see, not your own interpretation. Use copy and paste.

If you want it to use the select plan then you could hint for it (usually a driving site hint, you can view the full outline of the select plan in dbms_xplan.display_cursor with the appropriate format parameter). Or you could just do as I suggested my first response and run the select statement in PLSQL and run the return through an insert statement like:

begin

for rec in (select column_a from [email protected], [email protected] where a.col = b.col)

loop

insert into local_Table (column_a) values (rec.column_a);

end loop;

end;

/

John Thorton

How do I ask a question on the forums?

I can't answer your questions since I have no idea what you have, what you do or what you see.

Is COPY & PASTE broken for you?

AParm

These are the execution plans

SELECT STATEMENT (REMOTE)

  - HASH (GROUP BY)                                                         9

    - Access Predicates                                                          9

       A1.EDL_ID = A2.EDL

    - TABLE ACCESS (FULL)           OFFERS             981915

    - TABLE ACCESS (FULL)           LIST                  1786967

   

INSERT STATEMENT

-LOAD TABLE CONVENTIONAL

  - HASH (GROUP BY)                                                         9

    - Filter Predicates                                                             9

        NOT EXISTS(SEELCT 0 FROM A1 LNNVL(LIST.EDL_ID)<>:B1)

    - REMOTE          OFFERS                                      982004

    - REMOTE          LIST                                                       1

How do I get the insert to use the same plan as the select?

Mustafa_KALAYCI

now next step, post your sqls please (both of them). you should really look at the link that john said: How do I ask a question on the forums?

Unknown

I see the execution plans are different, for the first it just has TABLE FULL ACCESS for each of the tables, do I presume it does the join on the remote database, is that right?

Ok - first make sure you understand this concept. A query can only execute on ONE database instance.

So if you have a query that needs data from MORE THAN ONE db (e.g. across a link) then Oracle has to decide what db to execute the query on.

Whichever one it picks it then has to send the data from the other DB(s) to the db executing the query.

So for your simple 'select' query it could be executing it on the remote db but for the 'insert' query it could be executing it on the local db.

Depending on the size of the local and remote tables it can make a HUGE difference which way the data is sent.

1. The last storm blew down, and destroyed, the fence between your property and your neighbor.

2. You decide to build a concrete block wall to replace it and it will take 1000 blocks to build it.

3. Your neighbor has 900 concrete blocks on his property he will let you use if YOU build the wall.

4. You need 100 more and the store delivered them to your property.

5. You can only build the wall from ONE SIDE - yours or your neighbors.

Would you rather:

A. Carry your 100 blocks to your neighbors property?

B. Carry your neighbors 900 blocks to your property?

Post your answer.

We have NO IDEA who has more concrete blocks - you or your neighbor.

Which is why you need to SHOW US:

1. WHAT you do

2. HOW you do it

3. WHAT results you get

AndrewSayer

Amarprem wrote:These are the execution plansSELECT STATEMENT (REMOTE) - HASH (GROUP BY) 9 - Access Predicates 9 A1.EDL_ID = A2.EDL - TABLE ACCESS (FULL) OFFERS 981915 - TABLE ACCESS (FULL) LIST 1786967INSERT STATEMENT -LOAD TABLE CONVENTIONAL - HASH (GROUP BY) 9 - Filter Predicates 9 NOT EXISTS(SEELCT 0 FROM A1 LNNVL(LIST.EDL_ID)<>:B1) - REMOTE OFFERS 982004 - REMOTE LIST 1How do I get the insert to use the same plan as the select?

Your statement doesn't seem that simple if it's doing a group by and a not exists (or is that really a not in?)  on a remote DB.

As I've said before, to get the same plan you can either hint it (by checking the outline and note that it's probably a driving_site hint you are after)

OR you can use PLSQL..

FWIW I would consider a plan exposed using dbms_xplan much easier to read than what you have presented, you can also get it to include the full outline. If you need help doing that then google it or ask - don't just ignore the request.

AParm

insert into OFFERS_CHECK(
EXTRCT,
ISSUE,
A_COUNT,
SITE,
SOURCE
)
select
EXTRCT,
ISSUE,
A_COUNT,
SITE,
SOURCE from
V_OFFERS_CHECK;

CREATE V_OFFERS_CHECK
AS
select
'OFFERS' EXTRCT
'MISSING OFFERS' ISSUE,
count (*) A_Count,
OFFERS.SITE,
OFFERS.SOURCE
from OFFERS
WHERE OFFERS.EDL_ID not in (select EDL_ID from LIST)
and SOURCE = 'PST'
GROUP BY OFFERS.SITE,
OFFERS.SOURCE;

Simple, just check for all offers not in the LIST.

The select on it's own, less than a second. The insert just hangs, I just have to kill it.

Mustafa_KALAYCI

where are the other databases ? OFFERS  and LIST tables, are these on different databases ?

AParm

No they are on the same pluggable database LIVE, the are being accessed for reporting database.

Why should it matter, they are all just pluggable database on the SAME oracle server.

AndrewSayer

Amarprem wrote:No they are on the same pluggable database LIVE, the are being accessed for reporting database.Why should it matter, they are all just pluggable database on the SAME oracle server.

Remote inserts optimize differently.

Remote could be different Pdb or different server completely.

Where in the view/query are you specifying the db link? The view (I'm assuming that's what it is) is not valid DDL so I assume that wasn't copy and paste?

Have you tried the PL/SQL solution yet (I think I've mentioned it 4 times now).

begin

  for rec in (select EXTRCT,

                     ISSUE,

                     A_COUNT,

                     SITE,

                     SOURCE

               from  V_OFFERS_CHECK

              )

  loop

    insert into OFFERS_CHECK(EXTRCT,

                             ISSUE,

                             A_COUNT,

                             SITE,

                             SOURCE

                             )

    values

    (rec.EXTRCT,

     rec.ISSUE,

     rec.A_COUNT,

     rec.SITE,

     rec.SOURCE);

  end loop;

end;

/

AParm

OFFERS and LIST, are synonyms to tables on the 'remote' server.

Thanks for this, I will try it, I will look at dbms_xplan, and take this up again tomorrow.

Unknown

Why should it matter, they are all just pluggable database on the SAME oracle server.

I told you in my reply why it matters.

Each PDB is a separate and distinct database regardless of where they are physically located.

If you follow the example I provided and answer the question I ask then you will understand why it 'matters'.

Mustafa_KALAYCI

@rp0428 gave us a very good example at his first reply. you might be carrying all the rows from the db where tables located to database that you will make your insert. at first I thought that OFFERS and LIST tables are on different databases too but if they are on the same db, then this could be your problem, your insert statement works on the remote db (not at the db that offers and list are in it).

AndrewSayer

Mustafa KALAYCI wrote:rp0428 gave us a very good example at his first reply. you might be carrying all the rows from the db where tables located to database that you will make your insert. at first I thought that OFFERS and LIST tables are on different databases too but if they are on the same db, then this could be your problem, your insert statement works on the remote db (not at the db that offers and list are in it).

The plan tells us that for the insert statement:

All the rows from offers that match the simple predicate (SOURCE = 'PST') are pulled across from the remote DB to the local DB

For each row it then fires another query against the remote DB to check the existence in LIST (FILTER operation)

It then does the group by on the local DB

For the select statement, the entire query was passed to the remote DB which has unnested the not in subquery into an antijoin. The results were then passed to the local DB. This was much better because it was able to unnest the not in subquery rather than execute an existence check for each row.

AParm

How can I get it to use same the plan for insert as for the select?

Why does it have a cardinality of 1 for LIST in the insert plan as opposed to the 1786967 in the select, is this because it is fetching a single row at a time?

AndrewSayer

Amarprem wrote:How can I get it to use same the plan for insert as for the select?Why does it have a cardinality of 1 for LIST in the insert plan as opposed to the 1786967 in the select, is this because it is fetching a single row at a time?

I've recommended two ways, if you're going to just ignore them then why should we bother to help further. I've even allowed you to be lazy enough and just use copy and paste some code.

If you want more help then you will have to do some of the work yourself, if you're struggling to understand what has been recommended then try googling first and then asking for clarification. Repeating the same question will only lead to the same response.