Re: plpythonu memory leak

On Fri, Jan 14, 2011 at 19:14, Daniel Popowich <danielpopowich@gmail.com> wrote:
[ snip ]

> CREATE FUNCTION pygaps(start_ts timestamp without time zone, end_ts timestamp without time zone, gap_length interval)
RETURNSSETOF timerange 
>    LANGUAGE plpythonu
>    AS $$
>
>    # because pg passes date/time to python as strings I'm using pg to
>    # recompute values as seconds so I have numbers to do math
>
>    gap = plpy.execute("select extract(epoch from '%s'::interval) as sec"
>                       % gap_length)[0]['sec']
>
>    results = plpy.execute("""select ts, extract(epoch from ts) as epoch
>                              from timeseries
>                              where ts between '%s' and '%s'"""
>                           % (start_ts, end_ts))
>    if results.nrows() < 2:
>        return
>
>    # prime the well by setting prev(ious) to the first tic and
>    # iterate starting with the second...
>    prev = results[0]
>    for curr in results[1:]:

FYI if I don't use a slice copy here I can't get it to leak. ( find my
test case at the end ) I don't know enough about python to know if
thats a pl/python issue or python doing what its told--  having never
really wrote any python myself.

---------------
-- leaks big time
CREATE  or replace FUNCTION pygaps_leak() RETURNS void
   LANGUAGE plpythonu
   AS $$
results = plpy.execute("""select generate_series(0, 1000000)""")
prev = results[0]
for curr in results[1:]:
  prev = curr
return

-- does not leak
CREATE  or replace FUNCTION pygaps_no_leak() RETURNS void
   LANGUAGE plpythonu
   AS $$
results = plpy.execute("""select generate_series(0, 1000000)""")
prev = results[0]
for curr in range(1, len(results)):
  prev = curr
return