|
| 1 | +import logging |
| 2 | +import os |
| 3 | +import re |
| 4 | +from logging import Logger |
| 5 | +from typing import Dict, List, Literal |
| 6 | + |
| 7 | +from fastapi import FastAPI |
| 8 | + |
| 9 | +# OpenTelemetry |
| 10 | +from opentelemetry import metrics, trace |
| 11 | +from opentelemetry.exporter.otlp.proto.grpc._log_exporter import OTLPLogExporter |
| 12 | +from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter |
| 13 | +from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter |
| 14 | +from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor |
| 15 | +from opentelemetry.instrumentation.httpx import HTTPXClientInstrumentor |
| 16 | +from opentelemetry.instrumentation.requests import RequestsInstrumentor |
| 17 | +from opentelemetry.instrumentation.openai import OpenAIInstrumentor |
| 18 | +from opentelemetry.sdk._logs import LoggerProvider, LoggingHandler |
| 19 | +from opentelemetry.sdk._logs.export import BatchLogRecordProcessor |
| 20 | +from opentelemetry.sdk.metrics import Meter, MeterProvider |
| 21 | +from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader |
| 22 | +from opentelemetry.sdk.resources import Resource |
| 23 | +from opentelemetry.sdk.trace import Span, Tracer, TracerProvider |
| 24 | +from opentelemetry.sdk.trace.export import BatchSpanProcessor |
| 25 | +from opentelemetry.sdk.trace.sampling import ParentBasedTraceIdRatio |
| 26 | + |
| 27 | +SENSITIVE_DATA_SPAN_NAME = "sensitive_data_logged" |
| 28 | +SENSITIVE_DATA_INDICATOR_ATTRIBUTE_NAME = "contains_sensitive_data" |
| 29 | + |
| 30 | +DEVELOPMENT_MODE = Literal["DEVELOPMENT"] |
| 31 | +PRODUCTION_MODE = Literal["PRODUCTION"] |
| 32 | + |
| 33 | +ACTIVE_SERVICE_NAME = "translator_service" |
| 34 | + |
| 35 | +_has_already_init = False |
| 36 | +run_mode: Literal["DEVELOPMENT", "PRODUCTION"] = DEVELOPMENT_MODE |
| 37 | + |
| 38 | +_main_tracer: Tracer = None |
| 39 | +_main_logger: Logger = logging.getLogger() |
| 40 | +_main_meter: Meter = None |
| 41 | +log_level = (os.getenv("OTEL_LOG_LEVEL") or "INFO").upper() |
| 42 | + |
| 43 | +# https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/logging/logging.html |
| 44 | +logging.basicConfig( |
| 45 | + level=log_level, |
| 46 | + format="%(asctime)s - %(name)s[%(process)d] - %(levelname)s - %(message)s", |
| 47 | +) |
| 48 | + |
| 49 | + |
| 50 | +def ensure_initialized(): |
| 51 | + if not _has_already_init: |
| 52 | + raise Exception( |
| 53 | + "Observability module has not been initialized. Please call initialize_observability()." |
| 54 | + ) |
| 55 | + |
| 56 | + |
| 57 | +def is_development() -> bool: |
| 58 | + return run_mode == DEVELOPMENT_MODE |
| 59 | + |
| 60 | + |
| 61 | +def get_tracer(): |
| 62 | + global _main_tracer |
| 63 | + ensure_initialized() |
| 64 | + return _main_tracer |
| 65 | + |
| 66 | + |
| 67 | +def get_logger(name: str | None = None) -> logging.Logger: |
| 68 | + logger = logging.getLogger(name) |
| 69 | + return logger |
| 70 | + |
| 71 | + |
| 72 | +def get_meter(): |
| 73 | + global _main_meter |
| 74 | + ensure_initialized() |
| 75 | + return _main_meter |
| 76 | + |
| 77 | + |
| 78 | +def initialize_observability( |
| 79 | + mode: Literal["DEVELOPMENT", "PRODUCTION"], service_name: str = "ai.translator", environment: str = "Unspecified" |
| 80 | +): |
| 81 | + """Initializes the observability once for the lifetime of the application/process""" |
| 82 | + global \ |
| 83 | + _has_already_init, \ |
| 84 | + run_mode, \ |
| 85 | + _main_tracer, \ |
| 86 | + _main_logger, \ |
| 87 | + _main_meter, \ |
| 88 | + ACTIVE_SERVICE_NAME |
| 89 | + |
| 90 | + if _has_already_init: |
| 91 | + _main_logger.warning("Attempt made to initialize observability more than once") |
| 92 | + return |
| 93 | + |
| 94 | + _has_already_init = True |
| 95 | + |
| 96 | + run_mode = mode |
| 97 | + _main_logger.info(f"Initializing the observability with mode: {mode}") |
| 98 | + # See this for all the config options using environment variables: https://opentelemetry.io/docs/specs/otel/protocol/exporter/ |
| 99 | + opentelemetry_exporter_otlp_endpoint = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT") |
| 100 | + |
| 101 | + if opentelemetry_exporter_otlp_endpoint: |
| 102 | + _main_logger.info("🚀 Configuring OTLP telemetry") |
| 103 | + service_name = os.getenv( |
| 104 | + "OTEL_SERVICE_NAME", service_name |
| 105 | + ) # https://opentelemetry.io/docs/languages/sdk-configuration/general/#otel_service_name |
| 106 | + sample_ratio = float( |
| 107 | + os.getenv("OTEL_TRACES_SAMPLER_ARG", "1.0") |
| 108 | + ) # https://opentelemetry-python.readthedocs.io/en/latest/sdk/trace.sampling.html |
| 109 | + |
| 110 | + # setup the instrumentors |
| 111 | + resource = Resource.create( |
| 112 | + attributes={ |
| 113 | + "service.name": service_name, # https://opentelemetry.io/docs/specs/semconv/resource/#service |
| 114 | + "service.namespace": "ai.translator", |
| 115 | + "deployment.environment.name": environment, # https://opentelemetry.io/docs/specs/semconv/resource/deployment-environment/ |
| 116 | + "process.pid": str( |
| 117 | + os.getpid() |
| 118 | + ), # https://opentelemetry.io/docs/specs/semconv/attributes-registry/process/ |
| 119 | + } |
| 120 | + ) |
| 121 | + |
| 122 | + ACTIVE_SERVICE_NAME = service_name |
| 123 | + |
| 124 | + # tracing |
| 125 | + trace.set_tracer_provider( |
| 126 | + TracerProvider( |
| 127 | + resource=resource, sampler=ParentBasedTraceIdRatio(sample_ratio) |
| 128 | + ) |
| 129 | + ) |
| 130 | + span_processor = BatchSpanProcessor(OTLPSpanExporter()) |
| 131 | + trace.get_tracer_provider().add_span_processor(span_processor) |
| 132 | + _main_tracer = trace.get_tracer_provider().get_tracer(service_name) |
| 133 | + |
| 134 | + # metrics |
| 135 | + metric_reader = PeriodicExportingMetricReader(OTLPMetricExporter()) |
| 136 | + meter_provider = MeterProvider( |
| 137 | + resource=resource, metric_readers=[metric_reader] |
| 138 | + ) |
| 139 | + metrics.set_meter_provider(meter_provider) |
| 140 | + _main_meter = metrics.get_meter(service_name) |
| 141 | + |
| 142 | + # logging |
| 143 | + logger_provider = LoggerProvider(resource=resource) |
| 144 | + batch_log_record_processor = BatchLogRecordProcessor(OTLPLogExporter()) |
| 145 | + logger_provider.add_log_record_processor(batch_log_record_processor) |
| 146 | + |
| 147 | + handler = LoggingHandler(level=log_level, logger_provider=logger_provider) |
| 148 | + # Attach OTLP handler to root logger |
| 149 | + logging.getLogger().addHandler(handler) |
| 150 | + else: |
| 151 | + _main_logger.info( |
| 152 | + "🚀 OTLP telemetry exporter not configured (set OTEL_EXPORTER_OTLP_ENDPOINT)" |
| 153 | + ) |
| 154 | + _main_tracer = trace.get_tracer("default") |
| 155 | + _main_meter = metrics.get_meter("default") |
| 156 | + |
| 157 | + _main_logger = get_logger() |
| 158 | + _main_logger.info("Observability initialization complete") |
| 159 | + |
| 160 | + |
| 161 | +def mark_span_as_sensitive(span: Span): |
| 162 | + span.set_attribute(SENSITIVE_DATA_INDICATOR_ATTRIBUTE_NAME, "true") |
| 163 | + |
| 164 | + |
| 165 | +def add_sensitive_event(span: Span, event: str, attributes: dict[str, str]): |
| 166 | + if not attributes: |
| 167 | + attributes = {} |
| 168 | + |
| 169 | + attributes[SENSITIVE_DATA_INDICATOR_ATTRIBUTE_NAME] = "true" |
| 170 | + span.add_event(name=event, attributes=attributes) |
| 171 | + |
| 172 | + |
| 173 | +def log_sensitive_data( |
| 174 | + message: str, |
| 175 | + attributes: str | Dict | int | List = None, |
| 176 | + print_to_console: bool = False, |
| 177 | + span_name: str | None = None, |
| 178 | +) -> None: |
| 179 | + if is_development() and print_to_console: |
| 180 | + _main_logger.info(f"{message} - attributes={attributes}") |
| 181 | + |
| 182 | + if not span_name: |
| 183 | + span_name = SENSITIVE_DATA_SPAN_NAME |
| 184 | + |
| 185 | + with get_tracer().start_as_current_span(span_name) as span: |
| 186 | + if not attributes: |
| 187 | + attributes = {} |
| 188 | + if isinstance(attributes, dict): |
| 189 | + span.set_attributes({k: str(v) for k, v in attributes.items()}) |
| 190 | + if attributes: |
| 191 | + span.set_attribute("event.attributes", str(attributes)) |
| 192 | + |
| 193 | + span.set_attribute("message", message) |
| 194 | + span.set_attribute(SENSITIVE_DATA_INDICATOR_ATTRIBUTE_NAME, "true") |
| 195 | + |
| 196 | + |
| 197 | +def convert_to_metric_name(input_string: str) -> str: |
| 198 | + """ |
| 199 | + Converts a string into a metric name compatible with OpenTelemetry. |
| 200 | + # https://opentelemetry.io/docs/specs/otel/metrics/api/#instrument-name-syntax |
| 201 | +
|
| 202 | + Args: |
| 203 | + input_string (str): The input string to be converted. |
| 204 | +
|
| 205 | + Returns: |
| 206 | + str: The converted metric name. |
| 207 | + """ |
| 208 | + |
| 209 | + # Remove leading and trailing whitespace |
| 210 | + input_string = input_string.strip() |
| 211 | + # Add leading alpha character |
| 212 | + if not re.match(r"^[a-zA-Z]", input_string): |
| 213 | + input_string = "A" + input_string |
| 214 | + # Replace spaces with underscores |
| 215 | + input_string = input_string.replace(" ", "_") |
| 216 | + # Remove special characters and non-alphanumeric characters |
| 217 | + input_string = re.sub(r"[^a-zA-Z0-9_]", "", input_string) |
| 218 | + # Limit the length to 100 characters |
| 219 | + input_string = input_string[:100] |
| 220 | + |
| 221 | + return input_string |
| 222 | + |
| 223 | + |
| 224 | +def instrument_application(app: FastAPI): |
| 225 | + _main_logger.info("Setting up OpenTelemetry instrumentation...") |
| 226 | + RequestsInstrumentor().instrument() |
| 227 | + HTTPXClientInstrumentor().instrument() |
| 228 | + OpenAIInstrumentor().instrument() |
| 229 | + FastAPIInstrumentor.instrument_app( |
| 230 | + app, |
| 231 | + http_capture_headers_server_request=[".*"] |
| 232 | + ) |
0 commit comments