Time Traveling Key-Value Store, Again

2019-08-05

Introduction

In my last post I talked about a time traveling key-value store implementation in some functional Scala. After thinking about the problem more—probably a little too much—I decided to try my hand at another implementation, this time in Python again. Rather than going the purely functional route, though, I wanted to focus on how we represent the data internally. Since we essentially have tabular data with three columns (timestamp, key, and value), it made some sense to try and work with a table via the sqlite3 module in the Python standard library.

Implementation

First, some imports:

from pickle import dumps, loads
from sqlite3 import Row, connect
from time import perf_counter
from typing import Any, List, Optional

We use the pickle module to marshal (almost) arbitrary keys and values to binary blobs, time to give us a global clock as in my in-the-room implementation from the interview, and typing to label our types.

Since I wanted to wrap all this up with a stable release of nix, I needed to use Python 3.6; this meant that time.perf_counter_ns() wasn’t available, so we make our own out of time.perf_counter():

def perf_counter_ns() -> int:
    return int(perf_counter() * 1e9)

If one feels really strongly about Python 3.7, one could use the unstable nixpkgs channel.

Next we create a TTKV class:

class TTKV:

    def __init__(self) -> None:
        self.conn = connect(":memory:")
        self.conn.row_factory = Row
        self.cursor = self.conn.cursor()
        self.cursor.execute(
            "create table ttkv (timestamp integer, key blob, value blob)"
        )
        self.cursor.execute("create index timestamp_index on ttkv(timestamp)")
        self.cursor.execute("create index key_index on ttkv(key)")
        self.conn.commit()

Using this in-memory sqlite db, put() is an INSERT statement:

    def put(self, key: Any, value: Any) -> None:
        self.cursor.execute(
            "insert into ttkv values (?, ?, ?)",
            (perf_counter_ns(), dumps(key), dumps(value)),
        )

While a SELECT statement takes care of get():

    def get(self, key: Any, timestamp: Optional[int] = None) -> Any:
        select = "select * from ttkv where key = ?"
        order = "order by timestamp desc"
        if timestamp is None:
            self.cursor.execute(f"{select} {order}", (dumps(key),))
        elif isinstance(timestamp, int):
            self.cursor.execute(
                f"{select} and timestamp <= ? {order}", (dumps(key), timestamp)
            )
        else:
            raise ValueError(timestamp)
        result = self.cursor.fetchone()
        if result is None:
            raise KeyError(key)
        return loads(result["value"])

To be more Pythonic, get() throws errors if given an unknown key or a non-integral timestamp rather than going the Scala version’s Option route.

Finally, the times() helper returns relevant timestamps like in the Scala version:

    def times(self, key: Any = None) -> List[int]:
        select = "select timestamp from ttkv"
        order = "order by timestamp desc"
        if key is None:
            self.cursor.execute(f"{select} {order}")
        else:
            self.cursor.execute(f"{select} where key = ? {order}", (dumps(key),))
        return [row["timestamp"] for row in self.cursor.fetchall()]

Testing

My ttkv_spec.py looks similar to the previous TTKVSpec.scala, using the wonderful hypothesis library in place of scalacheck for property testing. I also lean on pytest for running the tests and for handling (deliberate) exceptions. I’ll copy just three tests here; interested readers should check the source on GitHub for the rest.

@given(STR)
def test_initially_empty(a: str) -> None:
    with raises(KeyError):
        TTKV().get(a)

@given(distinct_keys(), INT, INT)
def test_two_gets_different_keys(ab: Tuple[str, str], x: int, y: int) -> None:
    a, b = ab
    ttkv = TTKV()
    ttkv.put(a, x)
    ttkv.put(b, y)
    assert ttkv.get(a) == x
    assert ttkv.get(b) == y

@given(STR, INT, integers(min_value=0, max_value=SQLITE_MAX_INT))
def test_get_before_time(a: str, x: int, t: int) -> None:
    ttkv = TTKV()
    ttkv.put(a, x)
    t0 = ttkv.times(a)[0]
    with raises(KeyError):
        TTKV().get(a, t0 - t)

Thoughts

I’m of two minds here. Like the in-the-interview version I mentioned in my last post, this code relies heavily on side-effects, using its global nanosecond clock for timestamp generation. I liked how the functional Scala version explicitly isolated the effectful put() computation. However, I like how easily we map the TTKV API onto the sqlite table; the ORDER BY clauses obviate the need to sort things, e.g.

It also seems that if the spec for TTKV changes at all (as software specs are wont to do), we could easily modify this approach. It also suggests using multiple data backends, wherein we might define an abstract API and connect it to different specific data structures depending on our needs.

As a happy accident of using the pickle module, this TTKV implementation can use more types of things as keys than even the base dict class (e.g. lists). There are restrictions on what can be pickled, of course, and this implementation doesn’t handle any PicklingErrors that might occur if a use tries to save particular exotic keys or values.

I’m a huge fan of hypothesis, pytest, and typing along with mypy; I’d go so far as to say that I’d be reticent to develop serious software in Python without them. One (perhaps) interesting thing that happened in testing: though the sqlite3 module’s documentation claims that one can make arbitrary Python integers into native sqlite integer values, hypothesis quickly exposed a flaw in that claim, running into OverflowErrors when trying to save integral values too large to fit into 64 bits.

Appendix

Here’s the flake.nix file to specify the W O R L D:

{
  description = "Python TTKV";

  inputs = {
    flake-utils.url = "github:numtide/flake-utils";
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-22.11";
  };

  outputs = {
    self,
    flake-utils,
    nixpkgs,
  }:
    flake-utils.lib.eachDefaultSystem (system: let
      pkgs = import nixpkgs {inherit system;};
      python = pkgs.python311.withPackages (p: [p.hypothesis p.pytest]);
    in {
      packages.default = pkgs.writeShellApplication {
        name = "ttkv";
        text = ''
          pytest ttkv_spec.py
        '';
      };
    });
}

To run the tests yourself, poke around in the code, etc. feel free to copy the whole ttkv_py directory from my programming workbench on GitHub.