|
| 1 | +import os |
| 2 | +import logging |
| 3 | +import grpc |
| 4 | +import signal |
| 5 | +from concurrent import futures |
| 6 | +from typing import Dict, Any, List, Optional |
| 7 | + |
| 8 | +from azure.monitor.opentelemetry.exporter import AzureMonitorTraceExporter |
| 9 | +from opentelemetry import trace |
| 10 | +from opentelemetry.trace.status import Status, StatusCode |
| 11 | +from opentelemetry.sdk.resources import Resource |
| 12 | +from opentelemetry.sdk.trace import TracerProvider |
| 13 | +from opentelemetry.trace import SpanContext, TraceFlags, TraceState |
| 14 | +from opentelemetry.proto.collector.trace.v1 import ( |
| 15 | + trace_service_pb2, |
| 16 | + trace_service_pb2_grpc, |
| 17 | +) |
| 18 | +from opentelemetry.proto.trace.v1.trace_pb2 import Span as ProtoSpan |
| 19 | +from opentelemetry.proto.common.v1.common_pb2 import KeyValue |
| 20 | + |
| 21 | +DEFAULT_GRPC_PORT = 4317 |
| 22 | +SHUTDOWN_GRACE_PERIOD_SEC = 5 |
| 23 | + |
| 24 | +AZURE_CONN_STR = "InstrumentationKey=c0b360fa-422d-40e5-b8a9-d642578f9fce;IngestionEndpoint=https://eastus-8.in.applicationinsights.azure.com/;ApplicationId=087d527c-b60e-4346-a679-f6abf367d0f0" |
| 25 | + |
| 26 | + |
| 27 | +logging.basicConfig(level=logging.INFO) |
| 28 | +logger = logging.getLogger("image-customizer-telemetry") |
| 29 | + |
| 30 | + |
| 31 | +class SpanData: |
| 32 | + """SpanData class for Azure Monitor export.""" |
| 33 | + |
| 34 | + def __init__( |
| 35 | + self, proto_span: ProtoSpan, resource_attrs: Dict[str, Any], inst_scope: Any |
| 36 | + ) -> None: |
| 37 | + try: |
| 38 | + self.name = proto_span.name |
| 39 | + self.start_time = proto_span.start_time_unix_nano |
| 40 | + self.end_time = proto_span.end_time_unix_nano |
| 41 | + self.kind = proto_span.kind |
| 42 | + |
| 43 | + self.attributes = self._set_attributes( |
| 44 | + proto_span.attributes, resource_attrs |
| 45 | + ) |
| 46 | + self.status = self._extract_status(proto_span) |
| 47 | + self.events = proto_span.events |
| 48 | + self.links = proto_span.links |
| 49 | + self.context = self._create_span_context( |
| 50 | + proto_span.trace_id, proto_span.span_id |
| 51 | + ) |
| 52 | + self.parent = self._create_span_context( |
| 53 | + proto_span.trace_id, proto_span.parent_span_id |
| 54 | + ) |
| 55 | + self.resource = Resource.create(resource_attrs) |
| 56 | + self.instrumentation_scope = inst_scope |
| 57 | + |
| 58 | + except Exception as e: |
| 59 | + logger.error(f"Failed to initialize SpanData: {e}") |
| 60 | + raise |
| 61 | + |
| 62 | + def _set_attributes( |
| 63 | + self, proto_attributes: List[KeyValue], resource_attrs: Dict[str, Any] |
| 64 | + ) -> Dict[str, Any]: |
| 65 | + attributes = dict(resource_attrs) |
| 66 | + |
| 67 | + span_attrs = extract_attributes_from_proto(proto_attributes) |
| 68 | + attributes.update(span_attrs) |
| 69 | + |
| 70 | + return attributes |
| 71 | + |
| 72 | + def _extract_status(self, proto_span: ProtoSpan) -> Status: |
| 73 | + if proto_span.HasField("status"): |
| 74 | + return Status( |
| 75 | + status_code=StatusCode(proto_span.status.code), |
| 76 | + description=proto_span.status.message or None, |
| 77 | + ) |
| 78 | + return Status(StatusCode.UNSET) |
| 79 | + |
| 80 | + def _create_span_context(self, trace_id, span_id) -> SpanContext: |
| 81 | + return SpanContext( |
| 82 | + trace_id=int.from_bytes(trace_id, "big"), |
| 83 | + span_id=int.from_bytes(span_id, "big"), |
| 84 | + is_remote=True, |
| 85 | + trace_flags=TraceFlags(0), |
| 86 | + trace_state=TraceState(), |
| 87 | + ) |
| 88 | + |
| 89 | + |
| 90 | +class TraceServiceHandler(trace_service_pb2_grpc.TraceServiceServicer): |
| 91 | + """OTLP trace service handler that forwards traces to Azure Monitor.""" |
| 92 | + |
| 93 | + def __init__(self) -> None: |
| 94 | + """Initialize the trace service handler.""" |
| 95 | + self.exporter = self._initialize_telemetry() |
| 96 | + |
| 97 | + def _initialize_telemetry(self) -> AzureMonitorTraceExporter: |
| 98 | + """Initialize OpenTelemetry and Azure Monitor exporter.""" |
| 99 | + provider = TracerProvider(resource=Resource.create({})) |
| 100 | + trace.set_tracer_provider(provider) |
| 101 | + |
| 102 | + return AzureMonitorTraceExporter(connection_string=AZURE_CONN_STR) |
| 103 | + |
| 104 | + def Export(self, request, context) -> trace_service_pb2.ExportTraceServiceResponse: |
| 105 | + """Export traces to Azure Monitor.""" |
| 106 | + try: |
| 107 | + spans = self._process_trace_request(request) |
| 108 | + |
| 109 | + if spans: |
| 110 | + result = self.exporter.export(spans) |
| 111 | + logger.info( |
| 112 | + "Successfully exported %d spans to Azure Monitor (result: %s)", |
| 113 | + len(spans), |
| 114 | + result, |
| 115 | + ) |
| 116 | + return trace_service_pb2.ExportTraceServiceResponse() |
| 117 | + |
| 118 | + except Exception as e: |
| 119 | + logger.error("Error processing spans: %s", e, exc_info=True) |
| 120 | + context.set_code(grpc.StatusCode.INTERNAL) |
| 121 | + context.set_details(f"Failed to process spans: {str(e)}") |
| 122 | + return trace_service_pb2.ExportTraceServiceResponse() |
| 123 | + |
| 124 | + def _process_trace_request(self, request) -> List[SpanData]: |
| 125 | + """Process trace request and convert to SpanData objects.""" |
| 126 | + spans = [] |
| 127 | + |
| 128 | + for rs in request.resource_spans: |
| 129 | + resource_attrs = extract_attributes_from_proto(rs.resource.attributes) |
| 130 | + |
| 131 | + for ss in rs.scope_spans: |
| 132 | + for proto_span in ss.spans: |
| 133 | + try: |
| 134 | + span_data = SpanData(proto_span, resource_attrs, ss.scope) |
| 135 | + spans.append(span_data) |
| 136 | + except Exception as e: |
| 137 | + logger.warning(f"Failed to process span {proto_span.name}: {e}") |
| 138 | + |
| 139 | + return spans |
| 140 | + |
| 141 | + |
| 142 | +# Utility functions for protobuf attribute extraction |
| 143 | +def extract_attribute_value(value_proto: Any) -> Optional[Any]: |
| 144 | + """Extract value from protobuf AnyValue.""" |
| 145 | + value_case = value_proto.WhichOneof("value") |
| 146 | + value_mapping = { |
| 147 | + "string_value": value_proto.string_value, |
| 148 | + "int_value": value_proto.int_value, |
| 149 | + "double_value": value_proto.double_value, |
| 150 | + "bool_value": value_proto.bool_value, |
| 151 | + } |
| 152 | + return value_mapping.get(value_case) |
| 153 | + |
| 154 | + |
| 155 | +def extract_attributes_from_proto(proto_attributes: List[KeyValue]) -> Dict[str, Any]: |
| 156 | + """Extract attributes from protobuf KeyValue pairs.""" |
| 157 | + attributes = {} |
| 158 | + for kv in proto_attributes: |
| 159 | + value = extract_attribute_value(kv.value) |
| 160 | + if value is not None: |
| 161 | + attributes[kv.key] = value |
| 162 | + return attributes |
| 163 | + |
| 164 | + |
| 165 | +class TelemetryServer: |
| 166 | + """Telemetry hopper server that forwards OTLP traces to Azure Monitor.""" |
| 167 | + |
| 168 | + def __init__(self, port: int = DEFAULT_GRPC_PORT): |
| 169 | + self.port = port |
| 170 | + self.server: Optional[grpc.Server] = None |
| 171 | + |
| 172 | + def _start(self) -> None: |
| 173 | + """Start the telemetry forwarding server.""" |
| 174 | + try: |
| 175 | + self.server = self._create_server() |
| 176 | + self._setup_signal_handlers() |
| 177 | + |
| 178 | + self.server.start() |
| 179 | + logger.info( |
| 180 | + f"Telemetry server listening on port {self.port} for OTLP traces" |
| 181 | + ) |
| 182 | + |
| 183 | + except Exception as e: |
| 184 | + logger.error(f"Failed to start server: {e}") |
| 185 | + raise |
| 186 | + |
| 187 | + def stop(self, grace_period: int = SHUTDOWN_GRACE_PERIOD_SEC) -> None: |
| 188 | + """Stop the telemetry server gracefully.""" |
| 189 | + self.server.stop(grace_period) |
| 190 | + logger.info("Server stopped") |
| 191 | + |
| 192 | + def wait_for_termination(self) -> None: |
| 193 | + self.server.wait_for_termination() |
| 194 | + |
| 195 | + def run(self) -> None: |
| 196 | + self._start() |
| 197 | + self.wait_for_termination() |
| 198 | + |
| 199 | + def _create_server(self) -> grpc.Server: |
| 200 | + server = grpc.server(futures.ThreadPoolExecutor()) |
| 201 | + trace_service_pb2_grpc.add_TraceServiceServicer_to_server( |
| 202 | + TraceServiceHandler(), server |
| 203 | + ) |
| 204 | + server.add_insecure_port(f"[::]:{self.port}") |
| 205 | + return server |
| 206 | + |
| 207 | + def _setup_signal_handlers(self) -> None: |
| 208 | + |
| 209 | + def shutdown_handler(signum, frame): |
| 210 | + logger.info(f"Received signal {signum}, stopping server") |
| 211 | + self.stop() |
| 212 | + |
| 213 | + signal.signal(signal.SIGINT, shutdown_handler) |
| 214 | + signal.signal(signal.SIGTERM, shutdown_handler) |
| 215 | + |
| 216 | + |
| 217 | +if __name__ == "__main__": |
| 218 | + server = TelemetryServer() |
| 219 | + server.run() |
0 commit comments