Suggestion about logging only every n-th statement
Back in 2009 I made a suggestion which is not implemented yet but would
still be very valuable and easy to implement in my opinion (not for me
as I am not speaking C):
/messages/by-id/200903161426.56662.vygen@kicktipp.de
We still can't afford to log each and every statement as we have too
many and we don't want to afford more disk performance/space. Therefore
we log only statements slower than 50ms.
But for analyzing usage patterns it would be very nice to have this
combined with a sample_rate for logging.
logging_sample_rate = n
So each n-th statement will get logged regardless of execution time.
What do you think?
regards
Janning
On Wed, 20 Jun 2018 at 12:51, Janning Vygen <vygen@kicktipp.de> wrote:
But for analyzing usage patterns it would be very nice to have this
combined with a sample_rate for logging.logging_sample_rate = n
So each n-th statement will get logged regardless of execution time.
I think you would need to introduce a randomizing element if you wanted to
make it statistically valid. Logging every n'th statement could (depending
on your usage pattern) be catastrophically inaccurate.
Geoff
On 06/20/2018 01:51 PM, Janning Vygen wrote:
Back in 2009 I made a suggestion which is not implemented yet but would
still be very valuable and easy to implement in my opinion (not for me
as I am not speaking C):/messages/by-id/200903161426.56662.vygen@kicktipp.de
We still can't afford to log each and every statement as we have too
many and we don't want to afford more disk performance/space. Therefore
we log only statements slower than 50ms.But for analyzing usage patterns it would be very nice to have this
combined with a sample_rate for logging.logging_sample_rate = n
So each n-th statement will get logged regardless of execution time.
What do you think?
regards
Janning
Hello Janning,
FYI in made this patch which seems do what you want :
/messages/by-id/c30ee535-ee1e-db9f-fa97-146b9f62caed@anayrat.info
I will add an entry in september's commit fest.
Regards,
Am 20.06.2018 um 14:53 schrieb Adrien NAYRAT:
On 06/20/2018 01:51 PM, Janning Vygen wrote:
Back in 2009 I made a suggestion which is not implemented yet but
would still be very valuable and easy to implement in my opinion (not
for me as I am not speaking C):FYI in made this patch which seems do what you want :
/messages/by-id/c30ee535-ee1e-db9f-fa97-146b9f62caed@anayrat.infoI will add an entry in september's commit fest.
excellent. Exactly what I want!
regards
Janning
On 20 June 2018 13:51:25 CEST, Janning Vygen <vygen@kicktipp.de> wrote:
Back in 2009 I made a suggestion which is not implemented yet but would
still be very valuable and easy to implement in my opinion (not for me
as I am not speaking C):/messages/by-id/200903161426.56662.vygen@kicktipp.de
We still can't afford to log each and every statement as we have too
many and we don't want to afford more disk performance/space. Thereforewe log only statements slower than 50ms.
But for analyzing usage patterns it would be very nice to have this
combined with a sample_rate for logging.logging_sample_rate = n
So each n-th statement will get logged regardless of execution time.
What do you think?
regards
Janning
I would suggest using of pg_stat_statements for that.
Regards, Andreas
--
2ndQuadrant - The PostgreSQL Support Company
On 06/20/2018 03:08 PM, Andreas Kretschmer wrote:
I would suggest using of pg_stat_statements for that.
I agree, but sometimes it is not sufficient. For example, if you need query's
parameters.
--
Adrien NAYRAT
On 06/20/2018 03:06 PM, Janning Vygen wrote:
FYI in made this patch which seems do what you want :
/messages/by-id/c30ee535-ee1e-db9f-fa97-146b9f62caed@anayrat.infoI will add an entry in september's commit fest.
excellent. Exactly what I want!
FYI, I added an entry in next commit fest (september is not open) :
https://commitfest.postgresql.org/18/1691/
Feel free to review it ;)
Regards,
--
Adrien NAYRAT