On 10/27/25 16:14, Tomas Vondra wrote:
> On 10/21/25 16:43, Tomas Vondra wrote:
>> ...
>>
>> The results seem fairly stable, and the overall trend is clear. It'd be
>> great if there were no regressions, but considering how narrow is this
>> microbenchmark (and considering the benefits for practica
On 10/21/25 16:43, Tomas Vondra wrote:
> ...
>
> The results seem fairly stable, and the overall trend is clear. It'd be
> great if there were no regressions, but considering how narrow is this
> microbenchmark (and considering the benefits for practical COPY runs),
> I'd say it's probably OK.
>
On 10/21/25 09:10, Heikki Linnakangas wrote:
> On 18/10/2025 01:49, Tomas Vondra wrote:
>> On 10/17/25 12:32, Tomas Vondra wrote:
>>>
>>>
>>> On 10/17/25 10:31, Heikki Linnakangas wrote:
> typedef struct ResourceElem
> {
> Datum item;
> + uint32 count;
On 18/10/2025 01:49, Tomas Vondra wrote:
On 10/17/25 12:32, Tomas Vondra wrote:
On 10/17/25 10:31, Heikki Linnakangas wrote:
typedef struct ResourceElem
{
Datum item;
+ uint32 count; /* number of occurrences */
const ResourceOwnerDesc *kind;
On 10/16/25 21:28, Tom Lane wrote:
> Tomas Vondra writes:
>> On 10/16/25 20:12, Tom Lane wrote:
>>> Can you find evidence of this change being helpful for anything
>>> except this specific scenario in COPY?
>
>> I went through the ResourceOwnerRemember() calls, looking for other
>> cases that mig
Tomas Vondra writes:
> On 10/16/25 21:28, Tom Lane wrote:
>> I was thinking of adding some temporary instrumentation, like
>> just elog'ing whenever the count goes above 1, and seeing where
>> you get hits during the regression tests. I'm prepared to believe
>> this is worth doing, but it'd be ni
On 17/10/2025 06:13, Chao Li wrote:
```
@@ -250,12 +257,21 @@ ResourceOwnerAddToHash(ResourceOwner owner, Datum value,
const ResourceOwnerDesc
idx = hash_resource_elem(value, kind) & mask;
for (;;)
{
+ /* found an exact match - just increment the counter */
Tomas Vondra writes:
> The reason is pretty simple - ResourceOwner tracks the resources in a
> very simple hash table, with O(n^2) behavior with duplicates. This
> happens with COPY, because COPY creates an array of a 1000 tuple slots,
> and each slot references the same tuple descriptor. And the
> On Oct 17, 2025, at 01:46, Tomas Vondra wrote:
>
> --
> Tomas Vondra
Nice patch! I eyeball reviewed the patch, only got a few small comments:
1
```
@@ -250,12 +257,21 @@ ResourceOwnerAddToHash(ResourceOwner owner, Datum value,
const ResourceOwnerDesc
idx = hash_resource_elem(valu
On 10/17/25 10:31, Heikki Linnakangas wrote:
> On 17/10/2025 06:13, Chao Li wrote:
>> ```
>> @@ -250,12 +257,21 @@ ResourceOwnerAddToHash(ResourceOwner owner,
>> Datum value, const ResourceOwnerDesc
>> idx = hash_resource_elem(value, kind) & mask;
>> for (;;)
>> {
>> + /
Hi,
While reviewing and testing a nearby patch (using COPY for batching in
postgres_fdw), I noticed some of the COPY queries are spending a
substantial amount of time in ResourceOwnerAddToHash(). The exact figure
depends on amount of data in the COPY, but it was often close to 50%
(according to pe
On 10/17/25 12:32, Tomas Vondra wrote:
>
>
> On 10/17/25 10:31, Heikki Linnakangas wrote:
>> On 17/10/2025 06:13, Chao Li wrote:
>>> ```
>>> @@ -250,12 +257,21 @@ ResourceOwnerAddToHash(ResourceOwner owner,
>>> Datum value, const ResourceOwnerDesc
>>> idx = hash_resource_elem(value, kind) &
On 10/16/25 20:12, Tom Lane wrote:
> Tomas Vondra writes:
>> The reason is pretty simple - ResourceOwner tracks the resources in a
>> very simple hash table, with O(n^2) behavior with duplicates. This
>> happens with COPY, because COPY creates an array of a 1000 tuple slots,
>> and each slot refer
Tomas Vondra writes:
> On 10/16/25 20:12, Tom Lane wrote:
>> Can you find evidence of this change being helpful for anything
>> except this specific scenario in COPY?
> I went through the ResourceOwnerRemember() calls, looking for other
> cases that might create a lot of duplicates, similar to th
On 10/17/25 00:17, Tom Lane wrote:
> Tomas Vondra writes:
>> On 10/16/25 21:28, Tom Lane wrote:
>>> I was thinking of adding some temporary instrumentation, like
>>> just elog'ing whenever the count goes above 1, and seeing where
>>> you get hits during the regression tests. I'm prepared to belie
15 matches
Mail list logo
