Question

According to (my reading of) the official dox here:

https://docs.djangoproject.com/en/dev/ref/models/querysets/#when-querysets-are-evaluated

a Django QuerySet should be cached when you evaluate it. But that doesn’t seem to be the case. In the example that follows, TrackingImport is a model with a very large table behind it. (Output slightly edited for brevity.)

recs = TrackingImport.objects.filter(...stuff...)

In [102]: time(recs[0])
Wall time: 1.84 s

In [103]: time(recs[0])
Wall time: 1.84 s

Calling len() seems to work as advertised:

In [104]: len(recs)
Out[104]: 1823

In [105]: time(recs[0])
Wall time: 0.00 s

I don’t get why dereferencing the array didn’t cache the QuerySet results. It had to evaluate it, right? So what am I missing?

Answer 1

You can go through the source code(django.db.model.query), then you’ll be clear, here’s django 1.3.4’s query.py,

def __getitem__(self, k):
    """
    Retrieves an item or slice from the set of results.
    """
    if not isinstance(k, (slice, int, long)):
        raise TypeError
    assert ((not isinstance(k, slice) and (k >= 0))
            or (isinstance(k, slice) and (k.start is None or k.start >= 0)
                and (k.stop is None or k.stop >= 0))), \
            "Negative indexing is not supported."

    if self._result_cache is not None:
        if self._iter is not None:
            # The result cache has only been partially populated, so we may
            # need to fill it out a bit more.
            if isinstance(k, slice):
                if k.stop is not None:
                    # Some people insist on passing in strings here.
                    bound = int(k.stop)
                else:
                    bound = None
            else:
                bound = k + 1
            if len(self._result_cache) < bound:
                self._fill_cache(bound - len(self._result_cache))
        return self._result_cache[k]

    if isinstance(k, slice):
        qs = self._clone()
        if k.start is not None:
            start = int(k.start)
        else:
            start = None
        if k.stop is not None:
            stop = int(k.stop)
        else:
            stop = None
        qs.query.set_limits(start, stop)
        return k.step and list(qs)[::k.step] or qs
    try:
        qs = self._clone()
        qs.query.set_limits(k, k + 1)
        return list(qs)[0]
    except self.model.DoesNotExist, e:
        raise IndexError(e.args)

When you not iterate through the query set, the _result_cache is None, then when you invoke resc[0], it will just skip to following lines,

try:
   qs = self._clone()
   qs.query.set_limits(k, k + 1)
   return list(qs)[0]
except self.model.DoesNotExist, e:
   raise IndexError(e.args)

You’ll find that, in this case, the _result_cache is not being set. That’s why the duration of multiple resc[0] costs same time.

After you invoke len(resc), you can find source code,

def __len__(self):
    # Since __len__ is called quite frequently (for example, as part of
    # list(qs), we make some effort here to be as efficient as possible
    # whilst not messing up any existing iterators against the QuerySet.
    if self._result_cache is None:
        if self._iter:
            self._result_cache = list(self._iter)
        else:
            self._result_cache = list(self.iterator())
    elif self._iter:
        self._result_cache.extend(self._iter)
    return len(self._result_cache)

You can see the _result_cache has values, then you invoke recs[0], it will just use the cache,

 if self._result_cache is not None:
         ....
     return self._result_cache[k]

The souce code never lies, so it’s better to read the souce code when you don’t find your answer in documents.