diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2018-11-07 12:19:29 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2018-11-07 12:20:17 +0000 |
commit | a64a253794ac64cb40befee54db53bde17dd0d49 (patch) | |
tree | c1024acc5f6e508814b944d99f112259bb28b1be /python.d/web_log.chart.py | |
parent | New upstream version 1.10.0+dfsg (diff) | |
download | netdata-a64a253794ac64cb40befee54db53bde17dd0d49.tar.xz netdata-a64a253794ac64cb40befee54db53bde17dd0d49.zip |
New upstream version 1.11.0+dfsgupstream/1.11.0+dfsg
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'python.d/web_log.chart.py')
-rw-r--r-- | python.d/web_log.chart.py | 1047 |
1 files changed, 0 insertions, 1047 deletions
diff --git a/python.d/web_log.chart.py b/python.d/web_log.chart.py deleted file mode 100644 index be9baba9..00000000 --- a/python.d/web_log.chart.py +++ /dev/null @@ -1,1047 +0,0 @@ -# -*- coding: utf-8 -*- -# Description: web log netdata python.d module -# Author: l2isbad - -import bisect -import re -import os -import sys - -from collections import namedtuple, defaultdict -from copy import deepcopy - -try: - from itertools import filterfalse -except ImportError: - from itertools import ifilter as filter - from itertools import ifilterfalse as filterfalse - -from bases.collection import read_last_line -from bases.FrameworkServices.LogService import LogService - - -ORDER_APACHE_CACHE = ['apache_cache'] - -ORDER_WEB = ['response_statuses', 'response_codes', 'bandwidth', - 'response_time', 'response_time_hist', 'response_time_upstream', 'response_time_upstream_hist', - 'requests_per_url', 'requests_per_user_defined', 'http_method', 'http_version', - 'requests_per_ipproto', 'clients', 'clients_all'] - -ORDER_SQUID = ['squid_response_statuses', 'squid_response_codes', 'squid_detailed_response_codes', - 'squid_method', 'squid_mime_type', 'squid_hier_code', 'squid_transport_methods', - 'squid_transport_errors', 'squid_code', 'squid_handling_opts', 'squid_object_types', - 'squid_cache_events', 'squid_bytes', 'squid_duration', 'squid_clients', 'squid_clients_all'] - -CHARTS_WEB = { - 'response_codes': { - 'options': [None, 'Response Codes', 'requests/s', 'responses', 'web_log.response_codes', 'stacked'], - 'lines': [ - ['2xx', None, 'incremental'], - ['5xx', None, 'incremental'], - ['3xx', None, 'incremental'], - ['4xx', None, 'incremental'], - ['1xx', None, 'incremental'], - ['0xx', 'other', 'incremental'], - ['unmatched', None, 'incremental'] - ]}, - 'bandwidth': { - 'options': [None, 'Bandwidth', 'kilobits/s', 'bandwidth', 'web_log.bandwidth', 'area'], - 'lines': [ - ['resp_length', 'received', 'incremental', 8, 1000], - ['bytes_sent', 'sent', 'incremental', -8, 1000] - ]}, - 'response_time': { - 'options': [None, 'Processing Time', 'milliseconds', 'timings', 'web_log.response_time', 'area'], - 'lines': [ - ['resp_time_min', 'min', 'incremental', 1, 1000], - ['resp_time_max', 'max', 'incremental', 1, 1000], - ['resp_time_avg', 'avg', 'incremental', 1, 1000] - ]}, - 'response_time_hist': { - 'options': [None, 'Processing Time Histogram', 'requests/s', 'timings', 'web_log.response_time_hist', 'line'], - 'lines': [ - ]}, - 'response_time_upstream': { - 'options': [None, 'Processing Time Upstream', 'milliseconds', 'timings', - 'web_log.response_time_upstream', 'area'], - 'lines': [ - ['resp_time_upstream_min', 'min', 'incremental', 1, 1000], - ['resp_time_upstream_max', 'max', 'incremental', 1, 1000], - ['resp_time_upstream_avg', 'avg', 'incremental', 1, 1000] - ]}, - 'response_time_upstream_hist': { - 'options': [None, 'Processing Time Histogram', 'requests/s', 'timings', - 'web_log.response_time_upstream_hist', 'line'], - 'lines': [ - ]}, - 'clients': { - 'options': [None, 'Current Poll Unique Client IPs', 'unique ips', 'clients', 'web_log.clients', 'stacked'], - 'lines': [ - ['unique_cur_ipv4', 'ipv4', 'incremental', 1, 1], - ['unique_cur_ipv6', 'ipv6', 'incremental', 1, 1] - ]}, - 'clients_all': { - 'options': [None, 'All Time Unique Client IPs', 'unique ips', 'clients', 'web_log.clients_all', 'stacked'], - 'lines': [ - ['unique_tot_ipv4', 'ipv4', 'absolute', 1, 1], - ['unique_tot_ipv6', 'ipv6', 'absolute', 1, 1] - ]}, - 'http_method': { - 'options': [None, 'Requests Per HTTP Method', 'requests/s', 'http methods', 'web_log.http_method', 'stacked'], - 'lines': [ - ['GET', 'GET', 'incremental', 1, 1] - ]}, - 'http_version': { - 'options': [None, 'Requests Per HTTP Version', 'requests/s', 'http versions', - 'web_log.http_version', 'stacked'], - 'lines': []}, - 'requests_per_ipproto': { - 'options': [None, 'Requests Per IP Protocol', 'requests/s', 'ip protocols', 'web_log.requests_per_ipproto', - 'stacked'], - 'lines': [ - ['req_ipv4', 'ipv4', 'incremental', 1, 1], - ['req_ipv6', 'ipv6', 'incremental', 1, 1] - ]}, - 'response_statuses': { - 'options': [None, 'Response Statuses', 'requests/s', 'responses', 'web_log.response_statuses', - 'stacked'], - 'lines': [ - ['successful_requests', 'success', 'incremental', 1, 1], - ['server_errors', 'error', 'incremental', 1, 1], - ['redirects', 'redirect', 'incremental', 1, 1], - ['bad_requests', 'bad', 'incremental', 1, 1], - ['other_requests', 'other', 'incremental', 1, 1] - ]}, - 'requests_per_url': { - 'options': [None, 'Requests Per Url', 'requests/s', 'urls', 'web_log.requests_per_url', - 'stacked'], - 'lines': [ - ['url_pattern_other', 'other', 'incremental', 1, 1] - ]}, - 'requests_per_user_defined': { - 'options': [None, 'Requests Per User Defined Pattern', 'requests/s', 'user defined', - 'web_log.requests_per_user_defined', 'stacked'], - 'lines': [ - ['user_pattern_other', 'other', 'incremental', 1, 1] - ]} -} - -CHARTS_APACHE_CACHE = { - 'apache_cache': { - 'options': [None, 'Apache Cached Responses', 'percent cached', 'cached', 'web_log.apache_cache_cache', - 'stacked'], - 'lines': [ - ["hit", 'cache', "percentage-of-absolute-row"], - ["miss", None, "percentage-of-absolute-row"], - ["other", None, "percentage-of-absolute-row"] - ]} -} - -CHARTS_SQUID = { - 'squid_duration': { - 'options': [None, 'Elapsed Time The Transaction Busied The Cache', - 'milliseconds', 'squid_timings', 'web_log.squid_duration', 'area'], - 'lines': [ - ['duration_min', 'min', 'incremental', 1, 1000], - ['duration_max', 'max', 'incremental', 1, 1000], - ['duration_avg', 'avg', 'incremental', 1, 1000] - ]}, - 'squid_bytes': { - 'options': [None, 'Amount Of Data Delivered To The Clients', - 'kilobits/s', 'squid_bandwidth', 'web_log.squid_bytes', 'area'], - 'lines': [ - ['bytes', 'sent', 'incremental', 8, 1000] - ]}, - 'squid_response_statuses': { - 'options': [None, 'Response Statuses', 'responses/s', 'squid_responses', 'web_log.squid_response_statuses', - 'stacked'], - 'lines': [ - ['successful_requests', 'success', 'incremental', 1, 1], - ['server_errors', 'error', 'incremental', 1, 1], - ['redirects', 'redirect', 'incremental', 1, 1], - ['bad_requests', 'bad', 'incremental', 1, 1], - ['other_requests', 'other', 'incremental', 1, 1] - ]}, - 'squid_response_codes': { - 'options': [None, 'Response Codes', 'responses/s', 'squid_responses', - 'web_log.squid_response_codes', 'stacked'], - 'lines': [ - ['2xx', None, 'incremental'], - ['5xx', None, 'incremental'], - ['3xx', None, 'incremental'], - ['4xx', None, 'incremental'], - ['1xx', None, 'incremental'], - ['0xx', None, 'incremental'], - ['other', None, 'incremental'], - ['unmatched', None, 'incremental'] - ]}, - 'squid_code': { - 'options': [None, 'Responses Per Cache Result Of The Request', - 'requests/s', 'squid_squid_cache', 'web_log.squid_code', 'stacked'], - 'lines': [ - ]}, - 'squid_detailed_response_codes': { - 'options': [None, 'Detailed Response Codes', - 'responses/s', 'squid_responses', 'web_log.squid_detailed_response_codes', 'stacked'], - 'lines': [ - ]}, - 'squid_hier_code': { - 'options': [None, 'Responses Per Hierarchy Code', - 'requests/s', 'squid_hierarchy', 'web_log.squid_hier_code', 'stacked'], - 'lines': [ - ]}, - 'squid_method': { - 'options': [None, 'Requests Per Method', - 'requests/s', 'squid_requests', 'web_log.squid_method', 'stacked'], - 'lines': [ - ]}, - 'squid_mime_type': { - 'options': [None, 'Requests Per MIME Type', - 'requests/s', 'squid_requests', 'web_log.squid_mime_type', 'stacked'], - 'lines': [ - ]}, - 'squid_clients': { - 'options': [None, 'Current Poll Unique Client IPs', 'unique ips', 'squid_clients', - 'web_log.squid_clients', 'stacked'], - 'lines': [ - ['unique_ipv4', 'ipv4', 'incremental'], - ['unique_ipv6', 'ipv6', 'incremental'] - ]}, - 'squid_clients_all': { - 'options': [None, 'All Time Unique Client IPs', 'unique ips', 'squid_clients', - 'web_log.squid_clients_all', 'stacked'], - 'lines': [ - ['unique_tot_ipv4', 'ipv4', 'absolute'], - ['unique_tot_ipv6', 'ipv6', 'absolute'] - ]}, - 'squid_transport_methods': { - 'options': [None, 'Transport Methods', 'requests/s', 'squid_squid_transport', - 'web_log.squid_transport_methods', 'stacked'], - 'lines': [ - ]}, - 'squid_transport_errors': { - 'options': [None, 'Transport Errors', 'requests/s', 'squid_squid_transport', - 'web_log.squid_transport_errors', 'stacked'], - 'lines': [ - ]}, - 'squid_handling_opts': { - 'options': [None, 'Handling Opts', 'requests/s', 'squid_squid_cache', - 'web_log.squid_handling_opts', 'stacked'], - 'lines': [ - ]}, - 'squid_object_types': { - 'options': [None, 'Object Types', 'objects/s', 'squid_squid_cache', - 'web_log.squid_object_types', 'stacked'], - 'lines': [ - ]}, - 'squid_cache_events': { - 'options': [None, 'Cache Events', 'events/s', 'squid_squid_cache', - 'web_log.squid_cache_events', 'stacked'], - 'lines': [ - ]} -} - -NAMED_PATTERN = namedtuple('PATTERN', ['description', 'func']) - -DET_RESP_AGGR = ['', '_1xx', '_2xx', '_3xx', '_4xx', '_5xx', '_Other'] - -SQUID_CODES = dict(TCP='squid_transport_methods', UDP='squid_transport_methods', NONE='squid_transport_methods', - CLIENT='squid_handling_opts', IMS='squid_handling_opts', ASYNC='squid_handling_opts', - SWAPFAIL='squid_handling_opts', REFRESH='squid_handling_opts', SHARED='squid_handling_opts', - REPLY='squid_handling_opts', NEGATIVE='squid_object_types', STALE='squid_object_types', - OFFLINE='squid_object_types', INVALID='squid_object_types', FAIL='squid_object_types', - MODIFIED='squid_object_types', UNMODIFIED='squid_object_types', REDIRECT='squid_object_types', - HIT='squid_cache_events', MEM='squid_cache_events', MISS='squid_cache_events', - DENIED='squid_cache_events', NOFETCH='squid_cache_events', TUNNEL='squid_cache_events', - ABORTED='squid_transport_errors', TIMEOUT='squid_transport_errors') - -REQUEST_REGEX = re.compile(r'(?P<method>[A-Z]+) (?P<url>[^ ]+) [A-Z]+/(?P<http_version>\d(?:.\d)?)') - - -class Service(LogService): - def __init__(self, configuration=None, name=None): - """ - :param configuration: - :param name: - """ - LogService.__init__(self, configuration=configuration, name=name) - self.configuration = configuration - self.log_path = self.configuration.get('path') - self.job = None - - def check(self): - """ - :return: bool - - 1. "log_path" is specified in the module configuration file - 2. "log_path" must be readable by netdata user and must exist - 3. "log_path' must not be empty. We need at least 1 line to find appropriate pattern to parse - 4. other checks depends on log "type" - """ - - log_type = self.configuration.get('type', 'web') - log_types = dict(web=Web, apache_cache=ApacheCache, squid=Squid) - - if log_type not in log_types: - self.error("bad log type {log_type}. Supported types: {types}".format(log_type=log_type, - types=log_types.keys())) - return False - - if not self.log_path: - self.error('log path is not specified') - return False - - if not (self._find_recent_log_file() and os.access(self.log_path, os.R_OK)): - self.error('{log_file} not readable or not exist'.format(log_file=self.log_path)) - return False - - if not os.path.getsize(self.log_path): - self.error('{log_file} is empty'.format(log_file=self.log_path)) - return False - - self.job = log_types[log_type](self) - if self.job.check(): - self.order = self.job.order - self.definitions = self.job.definitions - return True - return False - - def _get_data(self): - return self.job.get_data(self._get_raw_data()) - - -class Web: - def __init__(self, service): - self.service = service - self.order = ORDER_WEB[:] - self.definitions = deepcopy(CHARTS_WEB) - self.pre_filter = check_patterns('filter', self.configuration.get('filter')) - self.storage = dict() - self.data = {'bytes_sent': 0, 'resp_length': 0, 'resp_time_min': 0, 'resp_time_max': 0, - 'resp_time_avg': 0, 'resp_time_upstream_min': 0, 'resp_time_upstream_max': 0, - 'resp_time_upstream_avg': 0, 'unique_cur_ipv4': 0, 'unique_cur_ipv6': 0, '2xx': 0, - '5xx': 0, '3xx': 0, '4xx': 0, '1xx': 0, '0xx': 0, 'unmatched': 0, 'req_ipv4': 0, - 'req_ipv6': 0, 'unique_tot_ipv4': 0, 'unique_tot_ipv6': 0, 'successful_requests': 0, - 'redirects': 0, 'bad_requests': 0, 'server_errors': 0, 'other_requests': 0, 'GET': 0} - - def __getattr__(self, item): - return getattr(self.service, item) - - def check(self): - last_line = read_last_line(self.log_path) - if not last_line: - return False - # Custom_log_format or predefined log format. - if self.configuration.get('custom_log_format'): - match_dict, error = self.find_regex_custom(last_line) - else: - match_dict, error = self.find_regex(last_line) - - # "match_dict" is None if there are any problems - if match_dict is None: - self.error(error) - return False - - self.storage['unique_all_time'] = list() - self.storage['url_pattern'] = check_patterns('url_pattern', self.configuration.get('categories')) - self.storage['user_pattern'] = check_patterns('user_pattern', self.configuration.get('user_defined')) - - self.create_web_charts(match_dict) # Create charts - self.info('Collected data: %s' % list(match_dict.keys())) - return True - - def create_web_charts(self, match_dict): - """ - :param match_dict: dict: regex.search.groupdict(). Ex. {'address': '127.0.0.1', 'code': '200', 'method': 'GET'} - :return: - Create/remove additional charts depending on the 'match_dict' keys and configuration file options - """ - if 'resp_time' not in match_dict: - self.order.remove('response_time') - self.order.remove('response_time_hist') - if 'resp_time_upstream' not in match_dict: - self.order.remove('response_time_upstream') - self.order.remove('response_time_upstream_hist') - - # Add 'response_time_hist' and 'response_time_upstream_hist' charts if is specified in the configuration - histogram = self.configuration.get('histogram', None) - if isinstance(histogram, list): - self.storage['bucket_index'] = histogram[:] - self.storage['bucket_index'].append(sys.maxint) - self.storage['buckets'] = [0] * (len(histogram) + 1) - self.storage['upstream_buckets'] = [0] * (len(histogram) + 1) - hist_lines = self.definitions['response_time_hist']['lines'] - upstream_hist_lines = self.definitions['response_time_upstream_hist']['lines'] - for i, le in enumerate(histogram): - hist_key = "response_time_hist_%d" % i - upstream_hist_key = "response_time_upstream_hist_%d" % i - hist_lines.append([hist_key, str(le), 'incremental', 1, 1]) - upstream_hist_lines.append([upstream_hist_key, str(le), 'incremental', 1, 1]) - - hist_lines.append(["response_time_hist_%d" % len(histogram), '+Inf', 'incremental', 1, 1]) - upstream_hist_lines.append(["response_time_upstream_hist_%d" % len(histogram), '+Inf', 'incremental', 1, 1]) - elif histogram is not None: - self.error("expect histogram list, but was {0}".format(type(histogram))) - - if not self.configuration.get('all_time', True): - self.order.remove('clients_all') - - # Add 'detailed_response_codes' chart if specified in the configuration - if self.configuration.get('detailed_response_codes', True): - if self.configuration.get('detailed_response_aggregate', True): - codes = DET_RESP_AGGR[:1] - else: - codes = DET_RESP_AGGR[1:] - - for code in codes: - self.order.append('detailed_response_codes%s' % code) - self.definitions['detailed_response_codes%s' % code] \ - = {'options': [None, 'Detailed Response Codes %s' % code[1:], 'requests/s', 'responses', - 'web_log.detailed_response_codes%s' % code, 'stacked'], - 'lines': []} - - # Add 'requests_per_url' chart if specified in the configuration - if self.storage['url_pattern']: - for elem in self.storage['url_pattern']: - dim = [elem.description, elem.description[12:], 'incremental'] - self.definitions['requests_per_url']['lines'].append(dim) - self.data[elem.description] = 0 - self.data['url_pattern_other'] = 0 - else: - self.order.remove('requests_per_url') - - # Add 'requests_per_user_defined' chart if specified in the configuration - if self.storage['user_pattern'] and 'user_defined' in match_dict: - for elem in self.storage['user_pattern']: - dim = [elem.description, elem.description[13:], 'incremental'] - self.definitions['requests_per_user_defined']['lines'].append(dim) - self.data[elem.description] = 0 - self.data['user_pattern_other'] = 0 - else: - self.order.remove('requests_per_user_defined') - - def get_data(self, raw_data=None): - """ - Parses new log lines - :return: dict OR None - None if _get_raw_data method fails. - In all other cases - dict. - """ - if not raw_data: - return None if raw_data is None else self.data - - filtered_data = filter_data(raw_data=raw_data, pre_filter=self.pre_filter) - - unique_current = set() - timings = defaultdict(lambda: dict(minimum=None, maximum=0, summary=0, count=0)) - - for line in filtered_data: - match = self.storage['regex'].search(line) - if match: - match_dict = match.groupdict() - try: - code = match_dict['code'][0] + 'xx' - self.data[code] += 1 - except KeyError: - self.data['0xx'] += 1 - # detailed response code - if self.configuration.get('detailed_response_codes', True): - self.get_data_per_response_codes_detailed(code=match_dict['code']) - # response statuses - self.get_data_per_statuses(code=match_dict['code']) - # requests per user defined pattern - if self.storage['user_pattern'] and 'user_defined' in match_dict: - self.get_data_per_pattern(row=match_dict['user_defined'], - other='user_pattern_other', - pattern=self.storage['user_pattern']) - # method, url, http version - self.get_data_from_request_field(match_dict=match_dict) - # bandwidth sent - bytes_sent = match_dict['bytes_sent'] if '-' not in match_dict['bytes_sent'] else 0 - self.data['bytes_sent'] += int(bytes_sent) - # request processing time and bandwidth received - if 'resp_length' in match_dict: - resp_length = match_dict['resp_length'] if '-' not in match_dict['resp_length'] else 0 - self.data['resp_length'] += int(resp_length) - if 'resp_time' in match_dict: - resp_time = self.storage['func_resp_time'](float(match_dict['resp_time'])) - get_timings(timings=timings['resp_time'], time=resp_time) - if 'bucket_index' in self.storage: - get_hist(self.storage['bucket_index'], self.storage['buckets'], resp_time / 1000) - if 'resp_time_upstream' in match_dict and match_dict['resp_time_upstream'] != '-': - resp_time_upstream = self.storage['func_resp_time'](float(match_dict['resp_time_upstream'])) - get_timings(timings=timings['resp_time_upstream'], time=resp_time_upstream) - if 'bucket_index' in self.storage: - get_hist(self.storage['bucket_index'], self.storage['upstream_buckets'], resp_time / 1000) - # requests per ip proto - proto = 'ipv6' if ':' in match_dict['address'] else 'ipv4' - self.data['req_' + proto] += 1 - # unique clients ips - if self.configuration.get('all_time', True): - if address_not_in_pool(pool=self.storage['unique_all_time'], - address=match_dict['address'], - pool_size=self.data['unique_tot_ipv4'] + self.data['unique_tot_ipv6']): - self.data['unique_tot_' + proto] += 1 - if match_dict['address'] not in unique_current: - self.data['unique_cur_' + proto] += 1 - unique_current.add(match_dict['address']) - else: - self.data['unmatched'] += 1 - - # timings - for elem in timings: - self.data[elem + '_min'] += timings[elem]['minimum'] - self.data[elem + '_avg'] += timings[elem]['summary'] / timings[elem]['count'] - self.data[elem + '_max'] += timings[elem]['maximum'] - - # histogram - if 'bucket_index' in self.storage: - buckets = self.storage['buckets'] - upstream_buckets = self.storage['upstream_buckets'] - for i in range(0, len(self.storage['bucket_index'])): - hist_key = "response_time_hist_%d" % i - upstream_hist_key = "response_time_upstream_hist_%d" % i - self.data[hist_key] = buckets[i] - self.data[upstream_hist_key] = upstream_buckets[i] - - return self.data - - def find_regex(self, last_line): - """ - :param last_line: str: literally last line from log file - :return: tuple where: - [0]: dict or None: match_dict or None - [1]: str: error description - We need to find appropriate pattern for current log file - All logic is do a regex search through the string for all predefined patterns - until we find something or fail. - """ - # REGEX: 1.IPv4 address 2.HTTP method 3. URL 4. Response code - # 5. Bytes sent 6. Response length 7. Response process time - default = re.compile(r'(?P<address>[\da-f.:]+|localhost)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+|-)') - - apache_ext_insert = re.compile(r'(?P<address>[\da-f.:]+|localhost)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+|-)' - r' (?P<resp_length>\d+|-)' - r' (?P<resp_time>\d+) ') - - apache_ext_append = re.compile(r'(?P<address>[\da-f.:]+|localhost)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+|-)' - r' .*?' - r' (?P<resp_length>\d+|-)' - r' (?P<resp_time>\d+)' - r'(?: |$)') - - nginx_ext_insert = re.compile(r'(?P<address>[\da-f.:]+)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+)' - r' (?P<resp_length>\d+)' - r' (?P<resp_time>\d+\.\d+) ') - - nginx_ext2_insert = re.compile(r'(?P<address>[\da-f.:]+)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+)' - r' (?P<resp_length>\d+)' - r' (?P<resp_time>\d+\.\d+)' - r' (?P<resp_time_upstream>[\d.-]+) ') - - nginx_ext_append = re.compile(r'(?P<address>[\da-f.:]+)' - r' -.*?"(?P<request>[^"]*)"' - r' (?P<code>[1-9]\d{2})' - r' (?P<bytes_sent>\d+)' - r' .*?' - r' (?P<resp_length>\d+)' - r' (?P<resp_time>\d+\.\d+)') - - def func_usec(time): - return time - - def func_sec(time): - return time * 1000000 - - r_regex = [apache_ext_insert, apache_ext_append, - nginx_ext2_insert, nginx_ext_insert, nginx_ext_append, - default] - r_function = [func_usec, func_usec, func_sec, func_sec, func_sec, func_usec] - regex_function = zip(r_regex, r_function) - - match_dict = dict() - for regex, func in regex_function: - match = regex.search(last_line) - if match: - self.storage['regex'] = regex - self.storage['func_resp_time'] = func - match_dict = match.groupdict() - break - - return find_regex_return(match_dict=match_dict or None, - msg='Unknown log format. You need to use "custom_log_format" feature.') - - def find_regex_custom(self, last_line): - """ - :param last_line: str: literally last line from log file - :return: tuple where: - [0]: dict or None: match_dict or None - [1]: str: error description - - We are here only if "custom_log_format" is in logs. We need to make sure: - 1. "custom_log_format" is a dict - 2. "pattern" in "custom_log_format" and pattern is <str> instance - 3. if "time_multiplier" is in "custom_log_format" it must be <int> instance - - If all parameters is ok we need to make sure: - 1. Pattern search is success - 2. Pattern search contains named subgroups (?P<subgroup_name>) (= "match_dict") - - If pattern search is success we need to make sure: - 1. All mandatory keys ['address', 'code', 'bytes_sent', 'method', 'url'] are in "match_dict" - - If this is True we need to make sure: - 1. All mandatory key values from "match_dict" have the correct format - ("code" is integer, "method" is uppercase word, etc) - - If non mandatory keys in "match_dict" we need to make sure: - 1. All non mandatory key values from match_dict ['resp_length', 'resp_time'] have the correct format - ("resp_length" is integer or "-", "resp_time" is integer or float) - - """ - if not hasattr(self.configuration.get('custom_log_format'), 'keys'): - return find_regex_return(msg='Custom log: "custom_log_format" is not a <dict>') - - pattern = self.configuration.get('custom_log_format', dict()).get('pattern') - if not (pattern and isinstance(pattern, str)): - return find_regex_return(msg='Custom log: "pattern" option is not specified or type is not <str>') - - resp_time_func = self.configuration.get('custom_log_format', dict()).get('time_multiplier') or 0 - - if not isinstance(resp_time_func, int): - return find_regex_return(msg='Custom log: "time_multiplier" is not an integer') - - try: - regex = re.compile(pattern) - except re.error as error: - return find_regex_return(msg='Pattern compile error: %s' % str(error)) - - match = regex.search(last_line) - if not match: - return find_regex_return(msg='Custom log: pattern search FAILED') - - match_dict = match.groupdict() or None - if match_dict is None: - return find_regex_return(msg='Custom log: search OK but contains no named subgroups' - ' (you need to use ?P<subgroup_name>)') - mandatory_dict = {'address': r'[\w.:-]+', - 'code': r'[1-9]\d{2}', - 'bytes_sent': r'\d+|-'} - optional_dict = {'resp_length': r'\d+|-', - 'resp_time': r'[\d.]+', - 'resp_time_upstream': r'[\d.-]+', - 'method': r'[A-Z]+', - 'http_version': r'\d(?:.\d)?'} - - mandatory_values = set(mandatory_dict) - set(match_dict) - if mandatory_values: - return find_regex_return(msg='Custom log: search OK but some mandatory keys (%s) are missing' - % list(mandatory_values)) - for key in mandatory_dict: - if not re.search(mandatory_dict[key], match_dict[key]): - return find_regex_return(msg='Custom log: can\'t parse "%s": %s' - % (key, match_dict[key])) - - optional_values = set(optional_dict) & set(match_dict) - for key in optional_values: - if not re.search(optional_dict[key], match_dict[key]): - return find_regex_return(msg='Custom log: can\'t parse "%s": %s' - % (key, match_dict[key])) - - dot_in_time = '.' in match_dict.get('resp_time', '') - if dot_in_time: - self.storage['func_resp_time'] = lambda time: time * (resp_time_func or 1000000) - else: - self.storage['func_resp_time'] = lambda time: time * (resp_time_func or 1) - - self.storage['regex'] = regex - return find_regex_return(match_dict=match_dict) - - def get_data_from_request_field(self, match_dict): - if match_dict.get('request'): - match_dict = REQUEST_REGEX.search(match_dict['request']) - if match_dict: - match_dict = match_dict.groupdict() - else: - return - # requests per url - if match_dict.get('url') and self.storage['url_pattern']: - self.get_data_per_pattern(row=match_dict['url'], - other='url_pattern_other', - pattern=self.storage['url_pattern']) - # requests per http method - if match_dict.get('method'): - if match_dict['method'] not in self.data: - self.charts['http_method'].add_dimension([match_dict['method'], - match_dict['method'], - 'incremental']) - self.data[match_dict['method']] = 0 - self.data[match_dict['method']] += 1 - # requests per http version - if match_dict.get('http_version'): - dim_id = match_dict['http_version'].replace('.', '_') - if dim_id not in self.data: - self.charts['http_version'].add_dimension([dim_id, - match_dict['http_version'], - 'incremental']) - self.data[dim_id] = 0 - self.data[dim_id] += 1 - - def get_data_per_response_codes_detailed(self, code): - """ - :param code: str: CODE from parsed line. Ex.: '202, '499' - :return: - Calls add_new_dimension method If the value is found for the first time - """ - if code not in self.data: - if self.configuration.get('detailed_response_aggregate', True): - self.charts['detailed_response_codes'].add_dimension([code, code, 'incremental']) - self.data[code] = 0 - else: - code_index = int(code[0]) if int(code[0]) < 6 else 6 - chart_key = 'detailed_response_codes' + DET_RESP_AGGR[code_index] - self.charts[chart_key].add_dimension([code, code, 'incremental']) - self.data[code] = 0 - self.data[code] += 1 - - def get_data_per_pattern(self, row, other, pattern): - """ - :param row: str: - :param other: str: - :param pattern: named tuple: (['pattern_description', 'regular expression']) - :return: - Scan through string looking for the first location where patterns produce a match for all user - defined patterns - """ - match = None - for elem in pattern: - if elem.func(row): - self.data[elem.description] += 1 - match = True - break - if not match: - self.data[other] += 1 - - def get_data_per_statuses(self, code): - """ - :param code: str: response status code. Ex.: '202', '499' - :return: - """ - code_class = code[0] - if code_class == '2' or code == '304' or code_class == '1': - self.data['successful_requests'] += 1 - elif code_class == '3': - self.data['redirects'] += 1 - elif code_class == '4': - self.data['bad_requests'] += 1 - elif code_class == '5': - self.data['server_errors'] += 1 - else: - self.data['other_requests'] += 1 - - -class ApacheCache: - def __init__(self, service): - self.service = service - self.order = ORDER_APACHE_CACHE - self.definitions = CHARTS_APACHE_CACHE - - @staticmethod - def check(): - return True - - @staticmethod - def get_data(raw_data=None): - data = dict(hit=0, miss=0, other=0) - if not raw_data: - return None if raw_data is None else data - - for line in raw_data: - if 'cache hit' in line: - data['hit'] += 1 - elif 'cache miss' in line: - data['miss'] += 1 - else: - data['other'] += 1 - return data - - -class Squid: - def __init__(self, service): - self.service = service - self.order = ORDER_SQUID - self.definitions = CHARTS_SQUID - self.pre_filter = check_patterns('filter', self.configuration.get('filter')) - self.storage = dict() - self.data = {'duration_max': 0, 'duration_avg': 0, 'duration_min': 0, 'bytes': 0, - '0xx': 0, '1xx': 0, '2xx': 0, '3xx': 0, '4xx': 0, '5xx': 0, - 'other': 0, 'unmatched': 0, 'unique_ipv4': 0, 'unique_ipv6': 0, - 'unique_tot_ipv4': 0, 'unique_tot_ipv6': 0, 'successful_requests': 0, - 'redirects': 0, 'bad_requests': 0, 'server_errors': 0, 'other_requests': 0 - } - - def __getattr__(self, item): - return getattr(self.service, item) - - def check(self): - last_line = read_last_line(self.log_path) - if not last_line: - return False - self.storage['unique_all_time'] = list() - self.storage['regex'] = re.compile(r'[0-9.]+\s+(?P<duration>[0-9]+)' - r' (?P<client_address>[\da-f.:]+)' - r' (?P<squid_code>[A-Z_]+)/' - r'(?P<http_code>[0-9]+)' - r' (?P<bytes>[0-9]+)' - r' (?P<method>[A-Z_]+)' - r' (?P<url>[^ ]+)' - r' (?P<user>[^ ]+)' - r' (?P<hier_code>[A-Z_]+)/[\da-f.:-]+' - r' (?P<mime_type>[^\n]+)') - - match = self.storage['regex'].search(last_line) - if not match: - self.error('Regex not matches (%s)' % self.storage['regex'].pattern) - return False - self.storage['dynamic'] = { - 'http_code': - {'chart': 'squid_detailed_response_codes', - 'func_dim_id': None, - 'func_dim': None}, - 'hier_code': { - 'chart': 'squid_hier_code', - 'func_dim_id': None, - 'func_dim': lambda v: v.replace('HIER_', '')}, - 'method': { - 'chart': 'squid_method', - 'func_dim_id': None, - 'func_dim': None}, - 'mime_type': { - 'chart': 'squid_mime_type', - 'func_dim_id': lambda v: v.split('/')[0], - 'func_dim': None}} - if not self.configuration.get('all_time', True): - self.order.remove('squid_clients_all') - return True - - def get_data(self, raw_data=None): - if not raw_data: - return None if raw_data is None else self.data - - filtered_data = filter_data(raw_data=raw_data, pre_filter=self.pre_filter) - - unique_ip = set() - timings = defaultdict(lambda: dict(minimum=None, maximum=0, summary=0, count=0)) - - for row in filtered_data: - match = self.storage['regex'].search(row) - if match: - match = match.groupdict() - if match['duration'] != '0': - get_timings(timings=timings['duration'], time=float(match['duration']) * 1000) - try: - self.data[match['http_code'][0] + 'xx'] += 1 - except KeyError: - self.data['other'] += 1 - - self.get_data_per_statuses(match['http_code']) - - self.get_data_per_squid_code(match['squid_code']) - - self.data['bytes'] += int(match['bytes']) - - proto = 'ipv4' if '.' in match['client_address'] else 'ipv6' - # unique clients ips - if self.configuration.get('all_time', True): - if address_not_in_pool(pool=self.storage['unique_all_time'], - address=match['client_address'], - pool_size=self.data['unique_tot_ipv4'] + self.data['unique_tot_ipv6']): - self.data['unique_tot_' + proto] += 1 - - if match['client_address'] not in unique_ip: - self.data['unique_' + proto] += 1 - unique_ip.add(match['client_address']) - - for key, values in self.storage['dynamic'].items(): - if match[key] == '-': - continue - dimension_id = values['func_dim_id'](match[key]) if values['func_dim_id'] else match[key] - if dimension_id not in self.data: - dimension = values['func_dim'](match[key]) if values['func_dim'] else dimension_id - self.charts[values['chart']].add_dimension([dimension_id, - dimension, - 'incremental']) - self.data[dimension_id] = 0 - self.data[dimension_id] += 1 - else: - self.data['unmatched'] += 1 - - for elem in timings: - self.data[elem + '_min'] += timings[elem]['minimum'] - self.data[elem + '_avg'] += timings[elem]['summary'] / timings[elem]['count'] - self.data[elem + '_max'] += timings[elem]['maximum'] - return self.data - - def get_data_per_statuses(self, code): - """ - :param code: str: response status code. Ex.: '202', '499' - :return: - """ - code_class = code[0] - if code_class == '2' or code == '304' or code_class == '1' or code == '000': - self.data['successful_requests'] += 1 - elif code_class == '3': - self.data['redirects'] += 1 - elif code_class == '4': - self.data['bad_requests'] += 1 - elif code_class == '5' or code_class == '6': - self.data['server_errors'] += 1 - else: - self.data['other_requests'] += 1 - - def get_data_per_squid_code(self, code): - """ - :param code: str: squid response code. Ex.: 'TCP_MISS', 'TCP_MISS_ABORTED' - :return: - """ - if code not in self.data: - self.charts['squid_code'].add_dimension([code, code, 'incremental']) - self.data[code] = 0 - self.data[code] += 1 - - for tag in code.split('_'): - try: - chart_key = SQUID_CODES[tag] - except KeyError: - continue - dimension_id = '_'.join(['code_detailed', tag]) - if dimension_id not in self.data: - self.charts[chart_key].add_dimension([dimension_id, tag, 'incremental']) - self.data[dimension_id] = 0 - self.data[dimension_id] += 1 - - -def get_timings(timings, time): - """ - :param timings: - :param time: - :return: - """ - if timings['minimum'] is None: - timings['minimum'] = time - if time > timings['maximum']: - timings['maximum'] = time - elif time < timings['minimum']: - timings['minimum'] = time - timings['summary'] += time - timings['count'] += 1 - -def get_hist(index, buckets, time): - """ - :param index: histogram index (Ex. [10, 50, 100, 150, ...]) - :param buckets: histogram buckets - :param time: time - :return: None - """ - for i in range(len(index)-1, -1, -1): - if time <= index[i]: - buckets[i] += 1 - else: - break - -def address_not_in_pool(pool, address, pool_size): - """ - :param pool: list of ip addresses - :param address: ip address - :param pool_size: current pool size - :return: True if address not in pool. False otherwise. - """ - index = bisect.bisect_left(pool, address) - if index < pool_size: - if pool[index] == address: - return False - bisect.insort_left(pool, address) - return True - bisect.insort_left(pool, address) - return True - - -def find_regex_return(match_dict=None, msg='Generic error message'): - """ - :param match_dict: dict: re.search.groupdict() or None - :param msg: str: error description - :return: tuple: - """ - return match_dict, msg - - -def check_patterns(string, dimension_regex_dict): - """ - :param string: str: - :param dimension_regex_dict: dict: ex. {'dim1': '<pattern1>', 'dim2': '<pattern2>'} - :return: list of named tuples or None: - We need to make sure all patterns are valid regular expressions - """ - if not hasattr(dimension_regex_dict, 'keys'): - return None - - result = list() - - def valid_pattern(pattern): - """ - :param pattern: str - :return: re.compile(pattern) or None - """ - if not isinstance(pattern, str): - return False - try: - return re.compile(pattern) - except re.error: - return False - - def func_search(pattern): - def closure(v): - return pattern.search(v) - - return closure - - for dimension, regex in dimension_regex_dict.items(): - valid = valid_pattern(regex) - if isinstance(dimension, str) and valid_pattern: - func = func_search(valid) - result.append(NAMED_PATTERN(description='_'.join([string, dimension]), - func=func)) - return result or None - - -def filter_data(raw_data, pre_filter): - """ - :param raw_data: - :param pre_filter: - :return: - """ - - if not pre_filter: - return raw_data - filtered = raw_data - for elem in pre_filter: - if elem.description == 'filter_include': - filtered = filter(elem.func, filtered) - elif elem.description == 'filter_exclude': - filtered = filterfalse(elem.func, filtered) - return filtered |