Support for SAMPLE BY in schema creation, SAMPLE clause, and toStartOfDay datetime function

[asantoni]

Hi all, just quickly hacked in support for SAMPLE BY support when creating models and added a missing datetime function. I also implemented support for the SAMPLE clause.

This adds a new "sample_by" parameter to the constructor of BaseMergeTree, which can be used to enable sampling on your table.

from clickhouse_backend import models
from clickhouse_backend.models.functions.hashes import farmFingerprint64
from clickhouse_backend.models.functions.datetime import toStartOfDay

class DemoLog(models.ClickhouseModel):
    timestamp = models.DateTimeField(default=timezone.now)
    ip = models.GenericIPAddressField(default="::")

    class Meta:
        engine = models.MergeTree(
            primary_key=("timestamp", toStartOfDay("timestamp"), farmFingerprint64("ip")),
            order_by=("timestamp", toStartOfDay("timestamp"), farmFingerprint64("ip")),
            partition_by=toStartOfMonth("timestamp"),
            sample_by=(farmFingerprint64("ip"),),
            index_granularity=8192,
        )

You can query using the SAMPLE clause using the new .sample function like so:

session_count_estimate = DemoLog.objects.filter(timestamp__gte=time_start, timestamp__lte=time_end).sample(0.1).aggregate(session_count=Count('id') * 10

The new sample function takes two parameters:

    def sample(self, sample_fraction, sample_offset=None):

which generates either a SAMPLE k or SAMPLE k OFFSET m clause as per the Clickhouse docs on SAMPLE.

I didn't include unit tests because I'm too lazy to spin up all the docker stuff and I'm in a hurry to get some bare minimum thing here working. I'm hoping this is PR is useful for others and could be useful for the project. Thanks!

[jayvynl]

Nice job.