Merge pull request #75 from Blockchain-Technology-Lab/concurrency

Concurrency

Author: dimkarakostas

requirements.txt

@@ -7,3 +7,4 @@ matplotlib~=3.4.3
 pandas~=1.3.4
 python-dateutil~=2.8.2
 pytest-mock~=3.12.0
+psutil~=6.0.0

tests/test_analyze.py

@@ -1,5 +1,5 @@
-from tokenomics_decentralization.analyze import analyze_snapshot, analyze, get_entries
-from unittest.mock import call
+from tokenomics_decentralization.analyze import analyze_snapshot, analyze, get_entries, analyze_ledger_snapshot
+from unittest.mock import call, Mock
 import pathlib
 
 
@@ -129,20 +129,25 @@ def test_get_entries(mocker):
 
 
 def test_analyze(mocker):
+    get_concurrency_mock = mocker.patch('tokenomics_decentralization.helper.get_concurrency_per_ledger')
+    get_concurrency_mock.return_value = {'bitcoin': 2, 'ethereum': 2}
+
+    write_csv_output_mock = mocker.patch('tokenomics_decentralization.helper.write_csv_output')
+
+    analyze(['bitcoin'], ['2010-01-01'])
+    assert len(write_csv_output_mock.call_args_list) == 1
+
+
+def test_analyze_ledger_snapshot(mocker):
     get_input_directories_mock = mocker.patch('tokenomics_decentralization.helper.get_input_directories')
     get_input_directories_mock.return_value = [pathlib.Path('/').resolve()]
 
     is_file_mock = mocker.patch('os.path.isfile')
     is_file_mock.side_effect = {
         pathlib.Path('/bitcoin_2010-01-01_raw_data.csv').resolve(): True,
-        pathlib.Path('/bitcoin_2011-01-01_raw_data.csv').resolve(): False,
         pathlib.Path('/ethereum_2010-01-01_raw_data.csv').resolve(): False,
-        pathlib.Path('/ethereum_2011-01-01_raw_data.csv').resolve(): True,
     }.get
 
-    get_db_connector_mock = mocker.patch('tokenomics_decentralization.db_helper.get_connector')
-    get_db_connector_mock.return_value = 'connector'
-
     get_entries_mock = mocker.patch('tokenomics_decentralization.analyze.get_entries')
     entries = [1, 2]
     get_entries_mock.return_value = entries
@@ -153,36 +158,26 @@ def test_analyze(mocker):
     get_output_row_mock = mocker.patch('tokenomics_decentralization.helper.get_output_row')
     get_output_row_mock.return_value = 'row'
 
-    write_csv_output_mock = mocker.patch('tokenomics_decentralization.helper.write_csv_output')
+    sema = Mock()
 
-    get_input_dirs_calls = []
     get_entries_calls = []
     analyze_snapshot_calls = []
     get_row_calls = []
-    write_output_calls = []
+    sema_release_calls = []
 
-    analyze(['bitcoin'], ['2010-01-01'])
-    get_input_dirs_calls.append(call())
-    assert get_input_directories_mock.call_args_list == get_input_dirs_calls
+    analyze_ledger_snapshot('bitcoin', '2010-01-01', [], sema)
     get_entries_calls.append(call('bitcoin', '2010-01-01', pathlib.Path('/bitcoin_2010-01-01_raw_data.csv').resolve()))
     assert get_entries_mock.call_args_list == get_entries_calls
     analyze_snapshot_calls.append(call(entries))
     assert analyze_snapshot_mock.call_args_list == analyze_snapshot_calls
     get_row_calls.append(call('bitcoin', '2010-01-01', {'hhi': 1}))
     assert get_output_row_mock.call_args_list == get_row_calls
-    write_output_calls.append(call(['row']))
-    assert write_csv_output_mock.call_args_list == write_output_calls
+    sema_release_calls.append(call())
+    assert sema.release.call_args_list == sema_release_calls
 
-    analyze(['bitcoin', 'ethereum'], ['2010-01-01', '2011-01-01'])
-    get_input_dirs_calls += 4 * [call()]
-    assert get_input_directories_mock.call_args_list == get_input_dirs_calls
-    get_entries_calls.append(call('bitcoin', '2010-01-01', pathlib.Path('/bitcoin_2010-01-01_raw_data.csv').resolve()))
-    get_entries_calls.append(call('ethereum', '2011-01-01', pathlib.Path('/ethereum_2011-01-01_raw_data.csv').resolve()))
+    analyze_ledger_snapshot('ethereum', '2010-01-01', [], sema)
     assert get_entries_mock.call_args_list == get_entries_calls
-    analyze_snapshot_calls += 2 * [call(entries)]
     assert analyze_snapshot_mock.call_args_list == analyze_snapshot_calls
-    get_row_calls.append(call('bitcoin', '2010-01-01', {'hhi': 1}))
-    get_row_calls.append(call('ethereum', '2011-01-01', {'hhi': 1}))
     assert get_output_row_mock.call_args_list == get_row_calls
-    write_output_calls.append(call(['row', 'row']))
-    assert write_csv_output_mock.call_args_list == write_output_calls
+    sema_release_calls.append(call())  # Test that semaphore release is called even if file does not exist
+    assert sema.release.call_args_list == sema_release_calls

tests/test_helper.py

@@ -1,7 +1,7 @@
 import tokenomics_decentralization.helper as hlp
+from collections import namedtuple
 import pathlib
 import os
-import argparse
 import datetime
 import pytest
 
@@ -12,9 +12,9 @@ def test_valid_date():
 
     for d in ['2022/1/01', '2022/01/1', '2022/1', '2022/01', '2022.1.01', '2022.01.1', '2022.1', '2022.01', 'blah',
               '2022-', '2022-1-1', '2022-1-01', '2022-1', '2022-01-1', '2022-02-29']:
-        with pytest.raises(argparse.ArgumentTypeError) as e_info:
+        with pytest.raises(ValueError) as e_info:
             hlp.valid_date(d)
-        assert e_info.type == argparse.ArgumentTypeError
+        assert e_info.type == ValueError
 
 
 def test_get_date_beginning():
@@ -486,3 +486,28 @@ def test_get_clusters(mocker):
     assert clusters['entity1'] == clusters['entity3']
     assert clusters['entity4'] == clusters['entity5']
     assert 'entity7' not in clusters.keys()
+
+
+def test_get_concurrency_per_ledger(mocker):
+    psutil_memory_mock = mocker.patch('psutil.virtual_memory')
+    psutil_memory_mock.return_value = namedtuple('VM', 'total')(10*10**9)
+
+    get_input_directories_mock = mocker.patch('tokenomics_decentralization.helper.get_input_directories')
+    get_input_directories_mock.return_value = [pathlib.Path('/').resolve()]
+
+    get_ledgers_mock = mocker.patch('tokenomics_decentralization.helper.get_ledgers')
+    get_ledgers_mock.return_value = ['bitcoin', 'ethereum']
+
+    os_walk_mock = mocker.patch('os.walk')
+    os_walk_mock.return_value = [('/', 'foo', ['bitcoin_2010-01-01_raw_data.csv'])]
+
+    os_stat_mock = mocker.patch('os.stat')
+    os_stat_mock.return_value = namedtuple('ST', 'st_size')(10*10**8)
+
+    concurrency = hlp.get_concurrency_per_ledger()
+    assert concurrency == {'bitcoin': 3, 'ethereum': 1}
+
+    os_stat_mock.return_value = namedtuple('ST', 'st_size')(5*10**9)
+
+    with pytest.raises(ValueError):
+        hlp.get_concurrency_per_ledger()

tokenomics_decentralization/analyze.py

@@ -1,4 +1,5 @@
 import csv
+import multiprocessing
 import os.path
 import tokenomics_decentralization.helper as hlp
 import tokenomics_decentralization.db_helper as db_hlp
@@ -113,35 +114,53 @@ def get_entries(ledger, date, filename):
     return entries
 
 
+def analyze_ledger_snapshot(ledger, date, output_rows, sema):
+    """
+    Executes the analysis of a given ledgers and snapshot date.
+    :param ledger: a ledger name
+    :param date: a string in YYYY-MM-DD format
+    :param output_rows: a list of strings in the form of csv output rows
+    :param sema: a multiprocessing semaphore
+    """
+    input_filename = None
+    input_paths = [input_dir / f'{ledger}_{date}_raw_data.csv' for input_dir in hlp.get_input_directories()]
+    for filename in input_paths:
+        if os.path.isfile(filename):
+            input_filename = filename
+            break
+    if input_filename:
+        logging.info(f'[*] {ledger} - {date}')
+
+        entries = get_entries(ledger, date, filename)
+        metrics_values = analyze_snapshot(entries)
+        del entries
+
+        row = hlp.get_output_row(ledger, date, metrics_values)
+        output_rows.append(row)
+
+    sema.release()  # Release the semaphore s.t. the loop in analyze() can continue
+
+
 def analyze(ledgers, snapshot_dates):
     """
     Executes the analysis of the given ledgers for the snapshot dates and writes the output
     to csv files.
     :param ledgers: a list of ledger names
     :param snapshot_dates: a list of strings in YYYY-MM-DD format
     """
-    output_rows = []
+    manager = multiprocessing.Manager()
+    output_rows = manager.list()  # output_rows is a shared list across all parallel processes
+
+    concurrency = hlp.get_concurrency_per_ledger()
     for ledger in ledgers:
-        logging.info(f'[*] {ledger} - Analyzing')
+        sema = multiprocessing.Semaphore(concurrency[ledger])
+        jobs = []
         for date in snapshot_dates:
-            logging.info(f'[*] {ledger} - {date}')
-
-            input_filename = None
-            input_paths = [input_dir / f'{ledger}_{date}_raw_data.csv' for input_dir in hlp.get_input_directories()]
-            for filename in input_paths:
-                if os.path.isfile(filename):
-                    input_filename = filename
-                    break
-            if not input_filename:
-                logging.error(f'{ledger} input data for {date} do not exist')
-                continue
-
-            entries = get_entries(ledger, date, filename)
-            metrics_values = analyze_snapshot(entries)
-            del entries
-
-            output_rows.append(hlp.get_output_row(ledger, date, metrics_values))
-            for metric, value in metrics_values.items():
-                logging.info(f'{metric}: {value}')
-
-    hlp.write_csv_output(output_rows)
+            sema.acquire()  # Loop blocks here while the active processes are as many as the semaphore's limit
+            p = multiprocessing.Process(target=analyze_ledger_snapshot, args=(ledger, date, output_rows, sema))
+            jobs.append(p)
+            p.start()
+        for proc in jobs:
+            proc.join()
+
+    hlp.write_csv_output(sorted(output_rows, key=lambda x: (x[0], x[1])))  # Csv rows ordered by ledger and date

tokenomics_decentralization/helper.py

@@ -6,7 +6,7 @@
 import os
 import datetime
 import calendar
-import argparse
+import psutil
 import json
 from collections import defaultdict
 import logging
@@ -33,8 +33,8 @@ def valid_date(date_string):
     try:
         get_date_beginning(date_string)
     except ValueError:
-        raise argparse.ArgumentTypeError("Please use the format YYYY-MM-DD for the timeframe argument "
-                                         "(day and month can be omitted).")
+        raise ValueError("Please use the format YYYY-MM-DD for the timeframe argument "
+                         "(day and month can be omitted).")
     return date_string
 
 
@@ -607,3 +607,41 @@ def get_clusters(ledger):
                 cluster_mapping[item[0]] = cluster_name
 
     return cluster_mapping
+
+
+def get_concurrency_per_ledger():
+    """
+    Computes the maximum number of parallel processes that can run per ledger,
+    based on the system's available memory.
+    :returns: a dictionary where the keys are ledger names and values are integers
+    """
+    system_memory_total = psutil.virtual_memory().total  # Get the system's total memory
+    system_memory_total -= 10**9  # Leave 1GB of memory to be used by other processes
+
+    concurrency = {}
+    too_large_ledgers = set()
+    input_dirs = get_input_directories()
+    for ledger in get_ledgers():
+        # Find the size of the largest input file per ledger
+        max_file_size = 0
+        for input_dir in input_dirs:
+            for folder, _, files in os.walk(input_dir):
+                for file in files:
+                    if file.startswith(ledger):
+                        max_file_size = max(max_file_size, os.stat(os.path.join(folder, file)).st_size)
+        # Compute the max number of processes that can open the largest ledger file
+        # and run in parallel without exhausting the system's memory.
+        if max_file_size > 0:
+            # When loaded in (a dict in) memory, each file consumes approx. 2.5x space compared to storage.
+            concurrency[ledger] = int(system_memory_total / (2.5 * max_file_size))
+            # Find if some ledger files are too large to fit in the system's available memory.
+            if concurrency[ledger] == 0:
+                too_large_ledgers.add(ledger)
+        else:
+            concurrency[ledger] = 1
+
+    if too_large_ledgers:
+        raise ValueError('The max input files of the following ledgers are too'
+                         'large to load in memory' + ','.join(too_large_ledgers))
+
+    return concurrency