Batch insert in CTAS/MatView code

Hi,

On Wed, Mar 06, 2019 at 10:06:27PM +0800, Paul Guo wrote:

The copy code has used batch insert with function heap_multi_insert() to
speed up. It seems that Create Table As or Materialized View could leverage
that code also to boost the performance also. Attached is a patch to
implement that. That was done by Taylor (cc-ed) and me.

Please note that we are currently in the last commit fest of Postgres
12, and it is too late to propose new features. Please feel free to
add an entry to the commit fest happening afterwards.
--
Michael

On 06/03/2019 22:06, Paul Guo wrote:

The patch also modifies heap_multi_insert() a bit to do a bit further
code-level optimization by using static memory, instead of using memory
context and dynamic allocation.

If toasting is required, heap_prepare_insert() creates a palloc'd tuple.
That is still leaked to the current memory context.

Leaking into the current memory context is not a bad thing, because
resetting a memory context is faster than doing a lot of pfree() calls.
The callers just need to be prepared for that, and use a short-lived
memory context.

By the way, while looking at the code, I noticed that there are 9 local
arrays with large length in toast_insert_or_update() which seems to be a
risk of stack overflow. Maybe we should put it as static or global.

Hmm. We currently reserve 512 kB between the kernel's limit, and the
limit we check in check_stack_depth(). See STACK_DEPTH_SLOP. Those
arrays add up to 52800 bytes on a 64-bit maching, if I did my math
right. So there's still a lot of headroom. I agree that it nevertheless
seems a bit excessive, though.

With the patch,

Time: 4728.142 ms (00:04.728)
Time: 14203.983 ms (00:14.204)
Time: 1008.669 ms (00:01.009)

Baseline,
Time: 11096.146 ms (00:11.096)
Time: 13106.741 ms (00:13.107)
Time: 1100.174 ms (00:01.100)

Nice speedup!

While for toast and large column size there is < 10% decrease but for
small column size the improvement is super good. Actually if I hardcode
the batch count as 4 all test cases are better but the improvement for
small column size is smaller than that with current patch. Pretty much
the number 4 is quite case specific so I can not hardcode that in the
patch. Of course we could further tune that but the current value seems
to be a good trade-off?

Have you done any profiling, on why the multi-insert is slower with
large tuples? In principle, I don't see why it should be slower.

- Heikki

Sorry for the late reply.

To Michael. Thank you. I know this commitfest is ongoing and I'm not
targeting for this.

On Thu, Mar 7, 2019 at 4:54 PM Heikki Linnakangas <hlinnaka@iki.fi> wrote:

On 06/03/2019 22:06, Paul Guo wrote:

The patch also modifies heap_multi_insert() a bit to do a bit further
code-level optimization by using static memory, instead of using memory
context and dynamic allocation.

If toasting is required, heap_prepare_insert() creates a palloc'd tuple.
That is still leaked to the current memory context.

Thanks. I checked the code for that but apparently, I missed that one. I'll
see what proper context can be used for CTAS. For copy code maybe just
revert my change.

Leaking into the current memory context is not a bad thing, because
resetting a memory context is faster than doing a lot of pfree() calls.
The callers just need to be prepared for that, and use a short-lived
memory context.

By the way, while looking at the code, I noticed that there are 9 local
arrays with large length in toast_insert_or_update() which seems to be a
risk of stack overflow. Maybe we should put it as static or global.

Hmm. We currently reserve 512 kB between the kernel's limit, and the
limit we check in check_stack_depth(). See STACK_DEPTH_SLOP. Those
arrays add up to 52800 bytes on a 64-bit maching, if I did my math
right. So there's still a lot of headroom. I agree that it nevertheless
seems a bit excessive, though.

I was worried about some recursive calling of it, but probably there should
be no worry for toast_insert_or_update().

With the patch,

Time: 4728.142 ms (00:04.728)
Time: 14203.983 ms (00:14.204)
Time: 1008.669 ms (00:01.009)

Baseline,
Time: 11096.146 ms (00:11.096)
Time: 13106.741 ms (00:13.107)
Time: 1100.174 ms (00:01.100)

Nice speedup!

While for toast and large column size there is < 10% decrease but for
small column size the improvement is super good. Actually if I hardcode
the batch count as 4 all test cases are better but the improvement for
small column size is smaller than that with current patch. Pretty much
the number 4 is quite case specific so I can not hardcode that in the
patch. Of course we could further tune that but the current value seems
to be a good trade-off?

Have you done any profiling, on why the multi-insert is slower with
large tuples? In principle, I don't see why it should be slower.

Thanks for the suggestion. I'll explore a bit more on this.

Show quoted text

- Heikki

On Wed, Mar 06, 2019 at 10:06:27PM +0800, Paul Guo wrote:

Hello, Postgres hackers,

The copy code has used batch insert with function heap_multi_insert() to
speed up. It seems that Create Table As or Materialized View could leverage
that code also to boost the performance also. Attached is a patch to
implement that.

This is great!

Is this optimization doable for multi-row INSERTs, either with tuples
spelled out in the body of the query or in constructs like INSERT ...
SELECT ...?

Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate

On Mon, Mar 11, 2019 at 2:58 AM David Fetter <david@fetter.org> wrote:

On Wed, Mar 06, 2019 at 10:06:27PM +0800, Paul Guo wrote:

Hello, Postgres hackers,

The copy code has used batch insert with function heap_multi_insert() to
speed up. It seems that Create Table As or Materialized View could

leverage

that code also to boost the performance also. Attached is a patch to
implement that.

This is great!

Is this optimization doable for multi-row INSERTs, either with tuples
spelled out in the body of the query or in constructs like INSERT ...
SELECT ...?

Yes. That's "batch insert" in the ModifyTable nodes which I mentioned in
the first email.
By the way, batch is a usual optimization mechanism for iteration kind
model (like postgres executor),
so batch should benefit many executor nodes in theory also.

Show quoted text

Best,
David.
--
David Fetter <david(at)fetter(dot)org>
https://urldefense.proofpoint.com/v2/url?u=http-3A__fetter.org_&amp;d=DwIBAg&amp;c=lnl9vOaLMzsy2niBC8-h_K-7QJuNJEsFrzdndhuJ3Sw&amp;r=Usi0ex6Ch92MsB5QQDgYFw&amp;m=wgGDTDFzZV7nnMm0NFt-yGKmm_KZk18RXKP9HL8h6UE&amp;s=tnaoLdajjR0Ew-93XUliHW1FUspVl09pIFd9aXxvqc8&amp;e=
Phone: +1 415 235 3778

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate

Hi all,

I've been working other things until recently I restarted the work,
profiling & refactoring the code.
It's been a long time since the last patch was proposed. The new patch has
now been firstly refactored due to
4da597edf1bae0cf0453b5ed6fc4347b6334dfe1 (Make TupleTableSlots extensible,
finish split of existing slot type).

Now that TupleTableSlot, instead of HeapTuple is one argument of
intorel_receive() so we can not get the
tuple length directly. This patch now gets the tuple length if we know all
columns are with fixed widths, else
we calculate an avg. tuple length using the first MAX_MULTI_INSERT_SAMPLES
(defined as 1000) tuples
and use for the total length of tuples in a batch.

I noticed that to do batch insert, we might need additional memory copy
sometimes comparing with "single insert"
(that should be the reason that we previously saw a bit regressions) so a
good solution seems to fall back
to "single insert" if the tuple length is larger than a threshold. I set
this as 2000 after quick testing.

To make test stable and strict, I run checkpoint before each ctas, the test
script looks like this:

checkpoint;
\timing
create table tt as select a,b,c from t11;
\timing
drop table tt;

Also previously I just tested the BufferHeapTupleTableSlot (i.e. create
table tt as select * from t11),
this time I test VirtualTupleTableSlot (i.e. create table tt as select
a,b,c from t11) additionally.
It seems that VirtualTupleTableSlot is very common in real cases.

I tested four kinds of tables, see below SQLs.

-- tuples with small size.
create table t11 (a int, b int, c int, d int);
insert into t11 select s,s,s,s from generate_series(1, 10000000) s;
analyze t11;

-- tuples that are untoasted and tuple size is 1984 bytes.
create table t12 (a name, b name, c name, d name, e name, f name, g name, h
name, i name, j name, k name, l name, m name, n name, o name, p name, q
name, r name, s name, t name, u name, v name, w name, x name, y name, z
name, a1 name, a2 name, a3 name, a4 name, a5 name);
insert into t12 select 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j',
'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y',
'z', 'a', 'b', 'c', 'd', 'e' from generate_series(1, 500000);
analyze t12;

-- tuples that are untoasted and tuple size is 2112 bytes.
create table t13 (a name, b name, c name, d name, e name, f name, g name, h
name, i name, j name, k name, l name, m name, n name, o name, p name, q
name, r name, s name, t name, u name, v name, w name, x name, y name, z
name, a1 name, a2 name, a3 name, a4 name, a5 name, a6 name, a7 name);
insert into t13 select 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j',
'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y',
'z', 'a', 'b', 'c', 'd', 'e', 'f', 'g' from generate_series(1, 500000);
analyze t13;

-- tuples that are toastable and tuple compressed size is 1084.
create table t14 (a text, b text, c text, d text, e text, f text, g text, h
text, i text, j text, k text, l text, m text, n text, o text, p text, q
text, r text, s text, t text, u text, v text, w text, x text, y text, z
text);
insert into t14 select i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i,
i, i, i, i, i, i, i, i, i from (select repeat('123456789', 10000) from
generate_series(1,5000)) i;
analyze t14;

I also tested two scenarios for each testing.

One is to clean up all kernel caches (page & inode & dentry on Linux) using
the command below and then run the test,
sync; echo 3 > /proc/sys/vm/drop_caches
After running all tests all relation files will be in kernel cache (my test
system memory is large enough to accommodate all relation files),
then I run the tests again. I run like this because in real scenario the
result of the test should be among the two results. Also I rerun
each test and finally I calculate the average results as the experiment
results. Below are some results:

scenario1: All related kernel caches are cleaned up (note the first two
columns are time with second).

baseline patch diff% SQL

10.1 5.57 44.85% create table tt as select * from t11;

10.7 5.52 48.41% create table tt as select a,b,c from t11;

9.57 10.2 -6.58% create table tt as select * from t12;

9.64 8.63 10.48% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4 from t12;

14.2 14.46 -1.83% create table tt as select * from t13;

11.88 12.05 -1.43% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4,a5,a6 from
t13;

3.17 3.25 -2.52% create table tt as select * from t14;

2.93 3.12 -6.48% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y from t14;

scenario2: all related kernel caches are populated after previous testing.

baseline patch diff% SQL

9.6 4.97 48.23% create table tt as select * from t11;

10.41 5.32 48.90% create table tt as select a,b,c from t11;

9.12 9.52 -4.38% create table tt as select * from t12;

9.66 8.6 10.97% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4 from t12;

13.56 13.6 -0.30% create table tt as select * from t13;

11.36 11.7 -2.99% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4,a5,a6 from
t13;

3.08 3.13 -1.62% create table tt as select * from t14;

2.95 3.03 -2.71% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y from t14;

From above we can get some tentative conclusions:

1. t11: For short-size tables, batch insert improves much (40%+).

2. t12: For BufferHeapTupleTableSlot, the patch slows down 4.x%-6.x%, but
for VirtualTupleTableSlot it improves 10.x%.
If we look at execTuples.c, it looks like this is quite relevant to
additional memory copy. It seems that VirtualTupleTableSlot is
more common than the BufferHeapTupleTableSlot so probably the current code
should be fine for most real cases. Or it's possible
to determine multi-insert also according to the input slot tuple but this
seems to be ugly in code. Or continue to lower the threshold a bit
so that "create table tt as select * from t12;" also improves although this
hurts the VirtualTupleTableSlot case.

3. for t13, new code still uses single insert so the difference should be
small. I just want to see the regression when even we use "single insert".

4. For toast case t14, the degradation is small, not a big deal.

By the way, did we try or think about allow better prefetch (on Linux) for
seqscan. i.e. POSIX_FADV_SEQUENTIAL in posix_fadvise() to enlarge the
kernel readahead window. Suppose this should help if seq tuple handling is
faster than default kernel readahead setting.

v2 patch is attached.

On Thu, Mar 7, 2019 at 4:54 PM Heikki Linnakangas <hlinnaka@iki.fi> wrote:

Show quoted text

On 06/03/2019 22:06, Paul Guo wrote:

The patch also modifies heap_multi_insert() a bit to do a bit further
code-level optimization by using static memory, instead of using memory
context and dynamic allocation.

If toasting is required, heap_prepare_insert() creates a palloc'd tuple.
That is still leaked to the current memory context.

Leaking into the current memory context is not a bad thing, because
resetting a memory context is faster than doing a lot of pfree() calls.
The callers just need to be prepared for that, and use a short-lived
memory context.

By the way, while looking at the code, I noticed that there are 9 local
arrays with large length in toast_insert_or_update() which seems to be a
risk of stack overflow. Maybe we should put it as static or global.

Hmm. We currently reserve 512 kB between the kernel's limit, and the
limit we check in check_stack_depth(). See STACK_DEPTH_SLOP. Those
arrays add up to 52800 bytes on a 64-bit maching, if I did my math
right. So there's still a lot of headroom. I agree that it nevertheless
seems a bit excessive, though.

With the patch,

Time: 4728.142 ms (00:04.728)
Time: 14203.983 ms (00:14.204)
Time: 1008.669 ms (00:01.009)

Baseline,
Time: 11096.146 ms (00:11.096)
Time: 13106.741 ms (00:13.107)
Time: 1100.174 ms (00:01.100)

Nice speedup!

While for toast and large column size there is < 10% decrease but for
small column size the improvement is super good. Actually if I hardcode
the batch count as 4 all test cases are better but the improvement for
small column size is smaller than that with current patch. Pretty much
the number 4 is quite case specific so I can not hardcode that in the
patch. Of course we could further tune that but the current value seems
to be a good trade-off?

Have you done any profiling, on why the multi-insert is slower with
large tuples? In principle, I don't see why it should be slower.

- Heikki

On Mon, Jun 17, 2019 at 8:53 PM Paul Guo <pguo@pivotal.io> wrote:

Hi all,

I've been working other things until recently I restarted the work,
profiling & refactoring the code.
It's been a long time since the last patch was proposed. The new patch has
now been firstly refactored due to
4da597edf1bae0cf0453b5ed6fc4347b6334dfe1 (Make TupleTableSlots extensible,
finish split of existing slot type).

Now that TupleTableSlot, instead of HeapTuple is one argument of
intorel_receive() so we can not get the
tuple length directly. This patch now gets the tuple length if we know all
columns are with fixed widths, else
we calculate an avg. tuple length using the first MAX_MULTI_INSERT_SAMPLES
(defined as 1000) tuples
and use for the total length of tuples in a batch.

I noticed that to do batch insert, we might need additional memory copy
sometimes comparing with "single insert"
(that should be the reason that we previously saw a bit regressions) so a
good solution seems to fall back
to "single insert" if the tuple length is larger than a threshold. I set
this as 2000 after quick testing.

To make test stable and strict, I run checkpoint before each ctas, the
test script looks like this:

checkpoint;
\timing
create table tt as select a,b,c from t11;
\timing
drop table tt;

Also previously I just tested the BufferHeapTupleTableSlot (i.e. create
table tt as select * from t11),
this time I test VirtualTupleTableSlot (i.e. create table tt as select
a,b,c from t11) additionally.
It seems that VirtualTupleTableSlot is very common in real cases.

I tested four kinds of tables, see below SQLs.

-- tuples with small size.
create table t11 (a int, b int, c int, d int);
insert into t11 select s,s,s,s from generate_series(1, 10000000) s;
analyze t11;

-- tuples that are untoasted and tuple size is 1984 bytes.
create table t12 (a name, b name, c name, d name, e name, f name, g name,
h name, i name, j name, k name, l name, m name, n name, o name, p name, q
name, r name, s name, t name, u name, v name, w name, x name, y name, z
name, a1 name, a2 name, a3 name, a4 name, a5 name);
insert into t12 select 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j',
'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y',
'z', 'a', 'b', 'c', 'd', 'e' from generate_series(1, 500000);
analyze t12;

-- tuples that are untoasted and tuple size is 2112 bytes.
create table t13 (a name, b name, c name, d name, e name, f name, g name,
h name, i name, j name, k name, l name, m name, n name, o name, p name, q
name, r name, s name, t name, u name, v name, w name, x name, y name, z
name, a1 name, a2 name, a3 name, a4 name, a5 name, a6 name, a7 name);
insert into t13 select 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j',
'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y',
'z', 'a', 'b', 'c', 'd', 'e', 'f', 'g' from generate_series(1, 500000);
analyze t13;

-- tuples that are toastable and tuple compressed size is 1084.
create table t14 (a text, b text, c text, d text, e text, f text, g text,
h text, i text, j text, k text, l text, m text, n text, o text, p text, q
text, r text, s text, t text, u text, v text, w text, x text, y text, z
text);
insert into t14 select i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i, i,
i, i, i, i, i, i, i, i, i from (select repeat('123456789', 10000) from
generate_series(1,5000)) i;
analyze t14;

I also tested two scenarios for each testing.

One is to clean up all kernel caches (page & inode & dentry on Linux)
using the command below and then run the test,
sync; echo 3 > /proc/sys/vm/drop_caches
After running all tests all relation files will be in kernel cache (my
test system memory is large enough to accommodate all relation files),
then I run the tests again. I run like this because in real scenario the
result of the test should be among the two results. Also I rerun
each test and finally I calculate the average results as the experiment
results. Below are some results:

scenario1: All related kernel caches are cleaned up (note the first two
columns are time with second).

baseline patch diff% SQL

10.1 5.57 44.85% create table tt as select * from t11;

10.7 5.52 48.41% create table tt as select a,b,c from t11;

9.57 10.2 -6.58% create table tt as select * from t12;

9.64 8.63 10.48% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4 from t12;

14.2 14.46 -1.83% create table tt as select * from t13;

11.88 12.05 -1.43% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4,a5,a6 from
t13;

3.17 3.25 -2.52% create table tt as select * from t14;

2.93 3.12 -6.48% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y from t14;

scenario2: all related kernel caches are populated after previous testing.

baseline patch diff% SQL

9.6 4.97 48.23% create table tt as select * from t11;

10.41 5.32 48.90% create table tt as select a,b,c from t11;

9.12 9.52 -4.38% create table tt as select * from t12;

9.66 8.6 10.97% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4 from t12;

13.56 13.6 -0.30% create table tt as select * from t13;

11.36 11.7 -2.99% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z,a1,a2,a3,a4,a5,a6 from
t13;

3.08 3.13 -1.62% create table tt as select * from t14;

2.95 3.03 -2.71% create table tt as select
a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y from t14;

From above we can get some tentative conclusions:

1. t11: For short-size tables, batch insert improves much (40%+).

2. t12: For BufferHeapTupleTableSlot, the patch slows down 4.x%-6.x%, but
for VirtualTupleTableSlot it improves 10.x%.
If we look at execTuples.c, it looks like this is quite relevant to
additional memory copy. It seems that VirtualTupleTableSlot is
more common than the BufferHeapTupleTableSlot so probably the current code
should be fine for most real cases. Or it's possible
to determine multi-insert also according to the input slot tuple but this
seems to be ugly in code. Or continue to lower the threshold a bit
so that "create table tt as select * from t12;" also improves although
this hurts the VirtualTupleTableSlot case.

To alleviate this. I tuned MAX_TUP_LEN_FOR_MULTI_INSERT a bit and set it
from 2000 to 1600. With a table with 24 name-typed columns (total size
1536), I tried both
case1: create table tt as select * from t12;
case2: create table tt as
select a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w from t12;

This patch increases the performance for both. Note, of course, this change
(MAX_TUP_LEN_FOR_MULTI_INSERT) does not affect the test results of previous
t11, t13, t14 in theory since the code path is not affected.

kernel caches cleaned up:

baseline(s) patch(s) diff%
case1: 7.65 7.30 4.6%
case2: 7.75 6.80 12.2%

relation files are in cache:

case1: 7.09 6.66 6.1%
case2: 7.49 6.83 8.8%

We do not need to find a larger threshold that just makes the case1
improvement near to zero since on other test environments the threshold
might be a bit different so it should be set as a rough value, and it seems
that 1600 should benefit most cases.

I attached the v3 patch which just has the MAX_TUP_LEN_FOR_MULTI_INSERT
change.

Thanks.

Show quoted text

3. for t13, new code still uses single insert so the difference should be
small. I just want to see the regression when even we use "single insert".

4. For toast case t14, the degradation is small, not a big deal.

By the way, did we try or think about allow better prefetch (on Linux) for
seqscan. i.e. POSIX_FADV_SEQUENTIAL in posix_fadvise() to enlarge the
kernel readahead window. Suppose this should help if seq tuple handling is
faster than default kernel readahead setting.

v2 patch is attached.

On Thu, Mar 7, 2019 at 4:54 PM Heikki Linnakangas <hlinnaka@iki.fi> wrote:

On 06/03/2019 22:06, Paul Guo wrote:

The patch also modifies heap_multi_insert() a bit to do a bit further
code-level optimization by using static memory, instead of using memory
context and dynamic allocation.

If toasting is required, heap_prepare_insert() creates a palloc'd tuple.
That is still leaked to the current memory context.

Leaking into the current memory context is not a bad thing, because
resetting a memory context is faster than doing a lot of pfree() calls.
The callers just need to be prepared for that, and use a short-lived
memory context.

By the way, while looking at the code, I noticed that there are 9 local
arrays with large length in toast_insert_or_update() which seems to be

a

risk of stack overflow. Maybe we should put it as static or global.

Hmm. We currently reserve 512 kB between the kernel's limit, and the
limit we check in check_stack_depth(). See STACK_DEPTH_SLOP. Those
arrays add up to 52800 bytes on a 64-bit maching, if I did my math
right. So there's still a lot of headroom. I agree that it nevertheless
seems a bit excessive, though.

With the patch,

Time: 4728.142 ms (00:04.728)
Time: 14203.983 ms (00:14.204)
Time: 1008.669 ms (00:01.009)

Baseline,
Time: 11096.146 ms (00:11.096)
Time: 13106.741 ms (00:13.107)
Time: 1100.174 ms (00:01.100)

Nice speedup!

While for toast and large column size there is < 10% decrease but for
small column size the improvement is super good. Actually if I hardcode
the batch count as 4 all test cases are better but the improvement for
small column size is smaller than that with current patch. Pretty much
the number 4 is quite case specific so I can not hardcode that in the
patch. Of course we could further tune that but the current value seems
to be a good trade-off?

Have you done any profiling, on why the multi-insert is slower with
large tuples? In principle, I don't see why it should be slower.

- Heikki

On 17/06/2019 15:53, Paul Guo wrote:

I noticed that to do batch insert, we might need additional memory copy
sometimes comparing with "single insert"
(that should be the reason that we previously saw a bit regressions) so a
good solution seems to fall back
to "single insert" if the tuple length is larger than a threshold. I set
this as 2000 after quick testing.

Where does the additional memory copy come from? Can we avoid doing it
in the multi-insert case?

- Heikki

On Fri, Aug 2, 2019 at 2:55 AM Heikki Linnakangas <hlinnaka@iki.fi> wrote:

On 17/06/2019 15:53, Paul Guo wrote:

I noticed that to do batch insert, we might need additional memory copy
sometimes comparing with "single insert"
(that should be the reason that we previously saw a bit regressions) so a
good solution seems to fall back
to "single insert" if the tuple length is larger than a threshold. I set
this as 2000 after quick testing.

Where does the additional memory copy come from? Can we avoid doing it
in the multi-insert case?

Hi Heikki,

Sorry for the late reply. I took some time on looking at & debugging the
code of TupleTableSlotOps
of various TupleTableSlot types carefully, especially the
BufferHeapTupleTableSlot case on which
we seemed to see regression if no threshold is set, also debugging &
testing more of the CTAS case.
I found my previous word "additional memory copy" (mainly tuple content
copy against single insert)
is wrong based on the latest code (probably is wrong also with previous
code). So in theory
we should not worry about additional tuple copy overhead now, and then I
tried the patch without setting
multi-insert threshold as attached.

To make test results more stable, this time I run a simple ' select
count(*) from tbl' before each CTAS to
warm up the shared buffer, run checkpoint before each CTAS, disable
autovacuum by setting
'autovacuum = off', set larger shared buffers (but < 25% of total memory
which is recommended
by PG doc) so that CTAS all hits shared buffer read if there exists warm
buffers (double-checked via
explain(analyze, buffers)). These seem to be reasonable for performance
testing. Each kind of CTAS
testing is run three times (Note before each run we do warm up and
checkpoint as mentioned).

I mainly tested the t12 (normal table with tuple size ~ 2k) case since for
others our patch either
performs better or similarly.

Patch: 1st_run 2nd_run 3rd_run

t12_BufferHeapTuple 7883.400 7549.966 8090.080
t12_Virtual 8041.637 8191.317 8182.404

Baseline: 1st_run 2nd_run 3rd_run

t12_BufferHeapTuple: 8264.290 7508.410 7681.702
t12_Virtual 8167.792 7970.537 8106.874

I actually roughly tested other tables we mentioned also (t11 and t14) -
the test results and conclusions are same.
t12_BufferHeapTuple means: create table tt as select * from t12;
t12_Virtual means: create table tt as select *partial columns* from t12;

So it looks like for t12 the results between our code and baseline are
similar so not setting
threshoud seem to be good though it looks like t12_BufferHeapTuple test
results varies a
lot (at most 0.5 seconds) for both our patch and baseline vs the virtual
case which is quite stable.

This actually confused me a bit given we've cached the source table in
shared buffers. I suspected checkpoint affects,
so I disabled checkpoint by setting max_wal_size = 3000 during CTAS, the
BufferHeapTuple case (see below)
still varies some. I'm not sure what's the reason but this does not seem to
a be blocker for the patch.
Patch: 1st_run 2nd_run 3rd_run
t12_BufferHeapTuple 7717.304 7413.259 7452.773
t12_Virtual 7445.742 7483.148 7593.583

Baseline: 1st_run 2nd_run 3rd_run
t12_BufferHeapTuple 8186.302 7736.541 7759.056
t12_Virtual 8004.880 8096.712 7961.483

On Mon, Sep 9, 2019 at 4:02 PM Paul Guo <pguo@pivotal.io> wrote:

So in theory
we should not worry about additional tuple copy overhead now, and then I

tried the patch without setting

multi-insert threshold as attached.

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API only.
However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide whether
to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision solely
based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple call
altogether?

The multi insert copy code deals with index tuples also, which I don't see
in the patch. Don't we need to consider populating indexes?

Asim

On 2019-Sep-25, Asim R P wrote:

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API only.
However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide whether
to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision solely
based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple call
altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple? That would make this easy. (I'm not
sure however what to do about TOAST considerations -- is it size in
memory that we're worried about?)

Also:

+ myState->mi_slots_size >= 65535)

This magic number should be placed in a define next to the other one,
but I'm not sure that heapam.h is a good location, since surely this
applies to matviews in other table AMs too.

--
�lvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Asim Thanks for the review.

On Wed, Sep 25, 2019 at 6:39 PM Asim R P <apraveen@pivotal.io> wrote:

On Mon, Sep 9, 2019 at 4:02 PM Paul Guo <pguo@pivotal.io> wrote:

So in theory
we should not worry about additional tuple copy overhead now, and then I

tried the patch without setting

multi-insert threshold as attached.

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API only.
However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide whether
to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision solely
based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple call
altogether?

For heapam, ExecFetchSlotHeapTuple() will be called again in
heap_multi_insert() to prepare the final multi-insert. if we check
ExecFetchSlotHeapTuple(), we could find that calling it multiple time just
involves very very few overhead for the BufferHeapTuple case. Note for
virtual tuple case the 2nd ExecFetchSlotHeapTuple() call still copies slot
contents, but we've called ExecCopySlot(batchslot, slot); to copy to a
BufferHeap case so no worries for the virtual tuple case (as a source).

Previously (long ago) I probably understood the code incorrectly so had the
concern also. I used sampling to do that (for variable-length tuple), but
now apparently we do not need that.

The multi insert copy code deals with index tuples also, which I don't see
in the patch. Don't we need to consider populating indexes?

create table as/create mat view DDL does not involve index creation for the
table/matview. The code seems to be able to used in RefreshMatView also,
for that we need to consider if we use multi-insert in that code.

On Thu, Sep 26, 2019 at 9:43 PM Alvaro Herrera <alvherre@2ndquadrant.com>
wrote:

On 2019-Sep-25, Asim R P wrote:

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API only.
However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide whether
to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision solely
based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple

call

altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple? That would make this easy. (I'm not
sure however what to do about TOAST considerations -- is it size in
memory that we're worried about?)

Also:

+ myState->mi_slots_size >= 65535)

This magic number should be placed in a define next to the other one,
but I'm not sure that heapam.h is a good location, since surely this
applies to matviews in other table AMs too.

yes defining 65535 seems better. Let's fix this one later when having

more feedback. Thanks.

On Thu, Sep 26, 2019 at 7:13 PM Alvaro Herrera <alvherre@2ndquadrant.com>
wrote:

On 2019-Sep-25, Asim R P wrote:

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API only.
However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide

whether

to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision

solely

based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple

call

altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple? That would make this easy. (I'm not
sure however what to do about TOAST considerations -- is it size in
memory that we're worried about?)

That will help. For slots containing heap tuples, heap_compute_data_size()
is what we need. Approximate size is better than nothing.
In case of CTAS, we are dealing with slots returned by a scan node.
Wouldn't TOAST datums be already expanded in those slots?

Asim

Hi,

On 2019-09-26 10:43:27 -0300, Alvaro Herrera wrote:

On 2019-Sep-25, Asim R P wrote:

I reviewed your patch today. It looks good overall. My concern is that
the ExecFetchSlotHeapTuple call does not seem appropriate. In a generic
place such as createas.c, we should be using generic tableam API
only.

Indeed.

However, I can also see that there is no better alternative. We need to
compute the size of accumulated tuples so far, in order to decide whether
to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision solely
based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple call
altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple?

Hm, I'm not convinced that it's worth adding that as a dedicated
operation. It's not that clear what it'd exactly mean anyway - what
would it measure? As referenced in the slot? As if it were stored on
disk? etc?

I wonder if the right answer wouldn't be to just measure the size of a
memory context containing the batch slots, or something like that.

That would make this easy. (I'm not sure however what to do about
TOAST considerations -- is it size in memory that we're worried
about?)

The in-memory size is probably fine, because in all likelihood the
toasted cols are just going to point to on-disk datums, no?

Also:

+ myState->mi_slots_size >= 65535)

This magic number should be placed in a define next to the other one,
but I'm not sure that heapam.h is a good location, since surely this
applies to matviews in other table AMs too.

Right. I think it'd be better to move this into an AM independent place.

Greetings,

Andres Freund

Hi,

On 2019-09-09 18:31:54 +0800, Paul Guo wrote:

diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.c
index e9544822bf..8a844b3b5f 100644
--- a/src/backend/access/heap/heapam.c
+++ b/src/backend/access/heap/heapam.c
@@ -2106,7 +2106,6 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
CommandId cid, int options, BulkInsertState bistate)
{
TransactionId xid = GetCurrentTransactionId();
-	HeapTuple  *heaptuples;
int			i;
int			ndone;
PGAlignedBlock scratch;
@@ -2115,6 +2114,10 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
Size		saveFreeSpace;
bool		need_tuple_data = RelationIsLogicallyLogged(relation);
bool		need_cids = RelationIsAccessibleInLogicalDecoding(relation);
+	/* Declare it as static to let this memory be not on stack. */
+	static HeapTuple	heaptuples[MAX_MULTI_INSERT_TUPLES];
+
+	Assert(ntuples <= MAX_MULTI_INSERT_TUPLES);

/* currently not needed (thus unsupported) for heap_multi_insert() */
AssertArg(!(options & HEAP_INSERT_NO_LOGICAL));
@@ -2124,7 +2127,6 @@ heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
HEAP_DEFAULT_FILLFACTOR);

/* Toast and set header data in all the slots */
- heaptuples = palloc(ntuples * sizeof(HeapTuple));
for (i = 0; i < ntuples; i++)
{
HeapTuple tuple;

I don't think this is a good idea. We shouldn't unnecessarily allocate
8KB on the stack. Is there any actual evidence this is a performance
benefit? To me this just seems like it'll reduce the flexibility of the
API, without any benefit. I'll also note that you've apparently not
updated tableam.h to document this new restriction.

Greetings,

Andres Freund

On Sat, Sep 28, 2019 at 5:49 AM Andres Freund <andres@anarazel.de> wrote:

Hi,

On 2019-09-09 18:31:54 +0800, Paul Guo wrote:

diff --git a/src/backend/access/heap/heapam.c

b/src/backend/access/heap/heapam.c

index e9544822bf..8a844b3b5f 100644
--- a/src/backend/access/heap/heapam.c
+++ b/src/backend/access/heap/heapam.c
@@ -2106,7 +2106,6 @@ heap_multi_insert(Relation relation,

TupleTableSlot **slots, int ntuples,

CommandId cid, int options,

BulkInsertState bistate)

{
TransactionId xid = GetCurrentTransactionId();
- HeapTuple *heaptuples;
int i;
int ndone;
PGAlignedBlock scratch;
@@ -2115,6 +2114,10 @@ heap_multi_insert(Relation relation,

TupleTableSlot **slots, int ntuples,

Size saveFreeSpace;
bool need_tuple_data =

RelationIsLogicallyLogged(relation);

bool need_cids =

RelationIsAccessibleInLogicalDecoding(relation);

+     /* Declare it as static to let this memory be not on stack. */
+     static HeapTuple        heaptuples[MAX_MULTI_INSERT_TUPLES];
+
+     Assert(ntuples <= MAX_MULTI_INSERT_TUPLES);

/* currently not needed (thus unsupported) for heap_multi_insert()

*/

AssertArg(!(options & HEAP_INSERT_NO_LOGICAL));
@@ -2124,7 +2127,6 @@ heap_multi_insert(Relation relation,

TupleTableSlot **slots, int ntuples,

HEAP_DEFAULT_FILLFACTOR);

/* Toast and set header data in all the slots */
- heaptuples = palloc(ntuples * sizeof(HeapTuple));
for (i = 0; i < ntuples; i++)
{
HeapTuple tuple;

I don't think this is a good idea. We shouldn't unnecessarily allocate
8KB on the stack. Is there any actual evidence this is a performance
benefit? To me this just seems like it'll reduce the flexibility of the

Previous heaptuples is palloc-ed in each batch, which should be slower than
pre-allocated & reusing memory in theory.

API, without any benefit. I'll also note that you've apparently not

updated tableam.h to document this new restriction.

Yes it should be moved from heapam.h to that file along with the 65535
definition.

However, I can also see that there is no better alternative. We need

to

compute the size of accumulated tuples so far, in order to decide

whether

to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision

solely

based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple

call

altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple?

Hm, I'm not convinced that it's worth adding that as a dedicated
operation. It's not that clear what it'd exactly mean anyway - what
would it measure? As referenced in the slot? As if it were stored on
disk? etc?

I wonder if the right answer wouldn't be to just measure the size of a
memory context containing the batch slots, or something like that.

Probably a better way is to move those logic (append slot to slots, judge
when to flush, flush, clean up slots) into table_multi_insert()? Generally
the final implementation of table_multi_insert() should be able to know
the sizes easily. One concern is that currently just COPY in the repo uses
multi insert, so not sure if other callers in the future want their own
logic (or set up a flag to allow customization but seems a bit
over-designed?).

Hi,

On 2019-09-30 12:12:31 +0800, Paul Guo wrote:

However, I can also see that there is no better alternative. We need

to

compute the size of accumulated tuples so far, in order to decide

whether

to stop accumulating tuples. There is no convenient way to obtain the
length of the tuple, given a slot. How about making that decision

solely

based on number of tuples, so that we can avoid ExecFetchSlotHeapTuple

call

altogether?

... maybe we should add a new operation to slots, that returns the
(approximate?) size of a tuple?

Hm, I'm not convinced that it's worth adding that as a dedicated
operation. It's not that clear what it'd exactly mean anyway - what
would it measure? As referenced in the slot? As if it were stored on
disk? etc?

I wonder if the right answer wouldn't be to just measure the size of a
memory context containing the batch slots, or something like that.

Probably a better way is to move those logic (append slot to slots, judge
when to flush, flush, clean up slots) into table_multi_insert()?

That does not strike me as a good idea. The upper layer is going to need
to manage some resources (e.g. it's the only bit that knows about how to
manage lifetime of the incoming data), and by exposing it to each AM
we're going to duplicate the necessary code too.

Generally the final implementation of table_multi_insert() should be
able to know the sizes easily. One concern is that currently just COPY
in the repo uses multi insert, so not sure if other callers in the
future want their own logic (or set up a flag to allow customization
but seems a bit over-designed?).

And that is also a concern, it seems unlikely that we'll get the
interface good.

Greetings,

Andres Freund