This thread has been stale since January with no movement at all during the
March CF, and according to the CFBot it stopped building at all ~ 14 weeks ago.
I'm marking this returned with feedback, it can be resubmitted for a future CF
if someone decides to pick it up.
--
Daniel Gustafsson
On Tue, 17 Jan 2023 at 14:52, Tomas Vondra
wrote:
>
> On 1/17/23 19:46, Andres Freund wrote:
>
> > I think a "hybrid" explain mode might be worth thinking about. Use the
> > "current" sampling method for the first execution of a node, and for the
> > first
> > few milliseconds of a query (or
On 1/17/23 19:46, Andres Freund wrote:
> Hi,
>
> On 2023-01-17 19:00:02 +0100, Tomas Vondra wrote:
>> On 1/17/23 18:02, Andres Freund wrote:
>>> On 2023-01-17 15:52:07 +0100, Tomas Vondra wrote:
That also does not have issues with timestamp "rounding" - considering
e.g. sample rate
Hi,
On 2023-01-17 19:00:02 +0100, Tomas Vondra wrote:
> On 1/17/23 18:02, Andres Freund wrote:
> > On 2023-01-17 15:52:07 +0100, Tomas Vondra wrote:
> >> That also does not have issues with timestamp "rounding" - considering
> >> e.g. sample rate 1000Hz, that's 1ms between samples. And it's quite
On 1/17/23 18:02, Andres Freund wrote:
> Hi,
>
> On 2023-01-17 15:52:07 +0100, Tomas Vondra wrote:
>> I don't understand why we would even use timestamps, in this case? AFAIK
>> "sampling profilers" simply increment a counter for the executing node,
>> and then approximate the time as
Hi,
On 2023-01-17 15:52:07 +0100, Tomas Vondra wrote:
> I don't understand why we would even use timestamps, in this case? AFAIK
> "sampling profilers" simply increment a counter for the executing node,
> and then approximate the time as proportional to the count.
The timer interrupt distances
On 1/15/23 21:22, Andres Freund wrote:
> Hi,
>
> On 2023-01-13 09:11:06 +0100, David Geier wrote:
>> Mostly I'm wondering if the sampling based approach gains us enough to be
>> worth it, once the patch to use RDTSC hopefully landed (see [1]).
>
> Well, I'm not sure we have a path forward on
On Fri, Jan 6, 2023 at 1:19 AM Jelte Fennema wrote:
> Nice addition! And the code looks pretty straight forward.
>
Thanks for reviewing!
The current patch triggers warnings:
> https://cirrus-ci.com/task/6016013976731648 Looks like you need to add
> void as the argument.
>
Fixed in v2
Hi,
On 2023-01-13 09:11:06 +0100, David Geier wrote:
> Mostly I'm wondering if the sampling based approach gains us enough to be
> worth it, once the patch to use RDTSC hopefully landed (see [1]).
Well, I'm not sure we have a path forward on it. There's portability and
accuracy concerns. But
Nice idea.
On 1/6/23 10:19, Jelte Fennema wrote:
Do you have some performance comparison between TIMING ON and TIMING
SAMPLING?
+1 to see some numbers compared to TIMING ON.
Mostly I'm wondering if the sampling based approach gains us enough to
be worth it, once the patch to use RDTSC
Nice addition! And the code looks pretty straight forward.
The current patch triggers warnings:
https://cirrus-ci.com/task/6016013976731648 Looks like you need to add void
as the argument.
Do you have some performance comparison between TIMING ON and TIMING
SAMPLING?
In InstrStartSampling
Hi,
Since EXPLAIN ANALYZE with TIMING ON still carries noticeable overhead on
modern hardware (despite time sources being faster), I'd like to propose a
new setting EXPLAIN ANALYZE, called "TIMING SAMPLING", as compared to
TIMING ON.
This new timing mode uses a timer on a fixed recurring
12 matches
Mail list logo
