It seems like pg_sleep_until() has issues if used within a transaction, as it uses now() and not clock_timestamp(). Please find attached a patch that solves this issue.
For consistency reasons, I also modified pg_sleep_for() to also use clock_timestamp.
On 01/30/2014 09:48 PM, Robert Haas wrote: > On Thu, Oct 17, 2013 at 9:11 AM, Vik Fearing <vik.fearing@dalibo.com> wrote: >> On 10/17/2013 02:42 PM, Robert Haas wrote: >>> On Thu, Oct 17, 2013 at 8:26 AM, Vik Fearing <vik.fearing@dalibo.com> wrote: >>>> On 10/17/2013 10:03 AM, Fabien COELHO wrote: >>>>> My guess is that it won't be committed if there is a single "but it >>>>> might break one code or surprise one user somewhere in the universe", >>>>> but I wish I'll be proven wrong. IMO, "returned with feedback" on a 1 >>>>> liner is really akin to "rejected". >>>> I have attached here an entirely new patch (new documentation and >>>> everything) that should please everyone. It no longer overloads >>>> pg_sleep(double precision) but instead add two new functions: >>>> >>>> * pg_sleep_for(interval) >>>> * pg_sleep_until(timestamp with time zone) >>>> >>>> Because it's no longer overloading the original pg_sleep, Robert's >>>> ambiguity objection is no more. >>>> >>>> Also, I like how it reads aloud: SELECT pg_sleep_for('5 minutes'); >>>> >>>> If people like this, I'll reject the current patch and add this one to >>>> the next commitfest. >>> I find that naming relatively elegant. However, you've got to >>> schema-qualify every function and operator used in the definitions, or >>> you're creating a search-path security vulnerability. >>> >> Good catch. Updated patch attached. > Committed.
