executors.py

import logging
import random
import time
from enum import Enum
from itertools import chain
from typing import List, Dict

from TM1py import TM1Service, Process

from results import PermutationResult


def swap(order: list, i1, i2) -> List[str]:
    seq = order[:]
    seq[i1], seq[i2] = seq[i2], seq[i1]
    return seq


def swap_random(order: list) -> List[str]:
    idx = range(len(order))
    i1, i2 = random.sample(idx, 2)
    return swap(order, i1, i2)


class ExecutionMode(Enum):
    ORIGINAL_ORDER = 0
    ITERATIONS = 1
    RESULT = 2

    @classmethod
    def _missing_(cls, value):
        for member in cls:
            if member.name.lower() == value.lower():
                return member
        # default
        return cls.ALL


class OptipyzerExecutor:
    def __init__(self, tm1: TM1Service, cube_name: str, view_names: list, displayed_dimension_order: List[str],
                 executions: int, measure_dimension_only_numeric: bool):
        self.tm1 = tm1
        self.cube_name = cube_name
        self.view_names = view_names
        self.dimensions = displayed_dimension_order
        self.executions = executions
        self.measure_dimension_only_numeric = measure_dimension_only_numeric
        self.mode = None

    def _determine_query_permutation_result(self) -> Dict[str, List[float]]:
        query_times_by_view = {}
        for view_name in self.view_names:
            query_times = []
            for _ in range(self.executions):
                self.clear_cube_cache()

                before = time.time()
                self.tm1.cells.create_cellset_from_view(cube_name=self.cube_name, view_name=view_name, private=False)
                query_times.append(time.time() - before)
            query_times_by_view[view_name] = query_times
        return query_times_by_view

    def _evaluate_permutation(self, permutation: List[str], retrieve_ram: bool = False,
                              reset_counter: bool = False) -> PermutationResult:
        ram_percentage_change = self.tm1.cubes.update_storage_dimension_order(self.cube_name, permutation)
        query_times_by_view = self._determine_query_permutation_result()

        ram_usage = None
        if retrieve_ram:
            ram_usage = self._retrieve_ram_usage()

        permutation_result = PermutationResult(self.mode, self.cube_name, self.view_names, permutation,
                                               query_times_by_view, ram_usage, ram_percentage_change, reset_counter)
        logging.info(f"Evaluated order: {permutation} "
                     f"- RAM [GB]: {permutation_result.ram_usage / 1024 ** 3:.2f} "
                     f"- Query time [s]: {permutation_result.median_query_time():.5f}")

        return permutation_result

    def _retrieve_ram_usage(self):
        number_of_iterations = 4
        for i in range(number_of_iterations):
            mdx = """
            SELECT  
            {{ [}}PerfCubes].[{}] }} ON ROWS,
            {{ [}}StatsStatsByCube].[Total Memory Used] }} ON COLUMNS
            FROM [}}StatsByCube]
            WHERE ([}}TimeIntervals].[LATEST])
            """.format(self.cube_name)
            value = list(self.tm1.cells.execute_mdx_values(mdx=mdx))[0]
            if value:
                return value

            logging.info("Failed to retrieve RAM consumption. Waiting 15s before retry")
            if i < number_of_iterations - 1:
                time.sleep(15)

        raise RuntimeError("Performance Monitor must be activated")

    def clear_cube_cache(self):
        process = Process(name="", prolog_procedure=f"DebugUtility(125 ,0 ,0 ,'{self.cube_name}' ,'' ,'');")
        success, status, error_log_file = self.tm1.processes.execute_process_with_return(process)

        if not success:
            raise RuntimeError(f"Failed to clear cache for cube '{self.cube_name}'. Status: '{status}'")


class OriginalOrderExecutor(OptipyzerExecutor):
    def __init__(self, tm1: TM1Service, cube_name: str, view_names: List[str], dimensions: List[str], executions: int,
                 measure_dimension_only_numeric: bool, original_dimension_order: List[str]):
        super().__init__(tm1, cube_name, view_names, dimensions, executions, measure_dimension_only_numeric)
        self.mode = ExecutionMode.ORIGINAL_ORDER
        self.original_dimension_order = original_dimension_order

    def execute(self, reset_counter=True):
        # at initial execution ram must be retrieved
        return [self._evaluate_permutation(
            self.original_dimension_order,
            retrieve_ram=True,
            reset_counter=reset_counter)]


class MainExecutor(OptipyzerExecutor):
    def __init__(self, tm1: TM1Service, cube_name: str, view_names: List[str], dimensions: List[str], executions: int,
                 measure_dimension_only_numeric: bool, fast: bool = False):
        super().__init__(tm1, cube_name, view_names, dimensions, executions,
                         measure_dimension_only_numeric)
        self.mode = ExecutionMode.ITERATIONS
        self.fast = fast

        if len(view_names) > 1:
            logging.warning("BestExecutor mode will use first view and ignore other views: " + str(view_names[1:]))

        self.view_name = view_names[0]

    def execute(self) -> List[PermutationResult]:
        dimensions = self.dimensions[:]
        resulting_order = self.dimensions[:]
        permutation_results = []
        dimension_pool = self.dimensions[:]

        mid = int(len(dimension_pool) / 2)

        if not self.measure_dimension_only_numeric:
            dimension_pool.remove(self.dimensions[-1])
            dimensions.remove(self.dimensions[-1])

        # iteration through positions like: n, 0, n-1, 1, n-2, 2, ...
        for iteration, position in enumerate(chain(*zip(reversed(range(len(dimensions))), range(len(dimensions))))):
            if self.fast and iteration == 2:
                break

            if position == mid:
                break

            results_per_dimension = list()

            for dimension in dimension_pool:
                original_position = resulting_order.index(dimension)

                permutation = list(resulting_order)
                permutation = swap(permutation, position, original_position)
                permutation_result = self._evaluate_permutation(permutation)
                permutation_results.append(permutation_result)
                results_per_dimension.append(permutation_result)

            if position > mid:
                best_order = sorted(
                    results_per_dimension,
                    key=lambda r: r.ram_usage)[0]
            else:
                best_order = sorted(
                    results_per_dimension,
                    key=lambda r: r.median_query_time(self.view_name))[0]

            resulting_order = list(best_order.dimension_order)
            dimension_pool.remove(resulting_order[position])

        return permutation_results