import base64 import hashlib import heapq import itertools import os import re import string from itertools import zip_longest from time import sleep from typing import Any, Callable, Iterator, List, Optional, Sequence, Set, Tuple, TypeVar from django.conf import settings T = TypeVar('T') def statsd_key(val: Any, clean_periods: bool=False) -> str: if not isinstance(val, str): val = str(val) if ':' in val: val = val.split(':')[0] val = val.replace('-', "_") if clean_periods: val = val.replace('.', '_') return val class StatsDWrapper: """Transparently either submit metrics to statsd or do nothing without erroring out""" # Backported support for gauge deltas # as our statsd server supports them but supporting # pystatsd is not released yet def _our_gauge(self, stat: str, value: float, rate: float=1, delta: bool=False) -> None: """Set a gauge value.""" from django_statsd.clients import statsd if delta: value_str = f'{value:+g}|g' else: value_str = f'{value:g}|g' statsd._send(stat, value_str, rate) def __getattr__(self, name: str) -> Any: # Hand off to statsd if we have it enabled # otherwise do nothing if name in ['timer', 'timing', 'incr', 'decr', 'gauge']: if settings.STATSD_HOST != '': from django_statsd.clients import statsd if name == 'gauge': return self._our_gauge else: return getattr(statsd, name) else: return lambda *args, **kwargs: None raise AttributeError statsd = StatsDWrapper() # Runs the callback with slices of all_list of a given batch_size def run_in_batches(all_list: Sequence[T], batch_size: int, callback: Callable[[Sequence[T]], None], sleep_time: int=0, logger: Optional[Callable[[str], None]]=None) -> None: if len(all_list) == 0: return limit = (len(all_list) // batch_size) + 1 for i in range(limit): start = i*batch_size end = (i+1) * batch_size if end >= len(all_list): end = len(all_list) batch = all_list[start:end] if logger: logger(f"Executing {end-start} in batch {i+1} of {limit}") callback(batch) if i != limit - 1: sleep(sleep_time) def make_safe_digest(string: str, hash_func: Callable[[bytes], Any]=hashlib.sha1) -> str: """ return a hex digest of `string`. """ # hashlib.sha1, md5, etc. expect bytes, so non-ASCII strings must # be encoded. return hash_func(string.encode('utf-8')).hexdigest() def log_statsd_event(name: str) -> None: """ Sends a single event to statsd with the desired name and the current timestamp This can be used to provide vertical lines in generated graphs, for example when doing a prod deploy, bankruptcy request, or other one-off events Note that to draw this event as a vertical line in graphite you can use the drawAsInfinite() command """ event_name = f"events.{name}" statsd.incr(event_name) def generate_random_token(length: int) -> str: return str(base64.b16encode(os.urandom(length // 2)).decode('utf-8').lower()) def generate_api_key() -> str: choices = string.ascii_letters + string.digits altchars = ''.join([choices[ord(os.urandom(1)) % 62] for _ in range(2)]).encode("utf-8") api_key = base64.b64encode(os.urandom(24), altchars=altchars).decode("utf-8") return api_key def has_api_key_format(key: str) -> bool: return bool(re.fullmatch(r"([A-Za-z0-9]){32}", key)) def query_chunker(queries: List[Any], id_collector: Optional[Set[int]]=None, chunk_size: int=1000, db_chunk_size: Optional[int]=None) -> Iterator[Any]: ''' This merges one or more Django ascending-id queries into a generator that returns chunks of chunk_size row objects during each yield, preserving id order across all results.. Queries should satisfy these conditions: - They should be Django filters. - They should return Django objects with "id" attributes. - They should be disjoint. The generator also populates id_collector, which we use internally to enforce unique ids, but which the caller can pass in to us if they want the side effect of collecting all ids. ''' if db_chunk_size is None: db_chunk_size = chunk_size // len(queries) assert db_chunk_size >= 2 assert chunk_size >= 2 if id_collector is not None: assert(len(id_collector) == 0) else: id_collector = set() def chunkify(q: Any, i: int) -> Iterator[Tuple[int, int, Any]]: q = q.order_by('id') min_id = -1 while True: assert db_chunk_size is not None # Hint for mypy, but also workaround for mypy bug #3442. rows = list(q.filter(id__gt=min_id)[0:db_chunk_size]) if len(rows) == 0: break for row in rows: yield (row.id, i, row) min_id = rows[-1].id iterators = [chunkify(q, i) for i, q in enumerate(queries)] merged_query = heapq.merge(*iterators) while True: tup_chunk = list(itertools.islice(merged_query, 0, chunk_size)) if len(tup_chunk) == 0: break # Do duplicate-id management here. tup_ids = {tup[0] for tup in tup_chunk} assert len(tup_ids) == len(tup_chunk) assert len(tup_ids.intersection(id_collector)) == 0 id_collector.update(tup_ids) yield [row for row_id, i, row in tup_chunk] def process_list_in_batches(lst: List[Any], chunk_size: int, process_batch: Callable[[List[Any]], None]) -> None: offset = 0 while True: items = lst[offset:offset+chunk_size] if not items: break process_batch(items) offset += chunk_size def split_by(array: List[Any], group_size: int, filler: Any) -> List[List[Any]]: """ Group elements into list of size `group_size` and fill empty cells with `filler`. Recipe from https://docs.python.org/3/library/itertools.html """ args = [iter(array)] * group_size return list(map(list, zip_longest(*args, fillvalue=filler)))