fix logging, add additional warnings

splunk_otel/__init__.py

@@ -20,6 +20,9 @@
 """
 
 from .defaults import _set_otel_defaults
+from .util import _init_logger
+
+_init_logger("splunk_otel")
 
 _set_otel_defaults()
 

splunk_otel/distro.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 import os
 from typing import Any, Dict
 
@@ -23,9 +24,9 @@
 from splunk_otel.profiling import _start_profiling
 from splunk_otel.profiling.options import _Options as ProfilingOptions
 from splunk_otel.tracing import _configure_tracing
-from splunk_otel.util import _get_logger, _is_truthy
+from splunk_otel.util import _is_truthy
 
-logger = _get_logger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class _SplunkDistro(BaseDistro):

splunk_otel/metrics.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 import os
 
 from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter
@@ -20,9 +21,9 @@
 from opentelemetry.sdk.metrics import MeterProvider
 from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader
 
-from splunk_otel.util import _get_logger, _is_truthy
+from splunk_otel.util import _is_truthy
 
-logger = _get_logger(__name__)
+logger = logging.getLogger(__name__)
 
 
 def start_metrics() -> MeterProvider:

splunk_otel/profiling/__init__.py

@@ -14,6 +14,7 @@
 
 import base64
 import gzip
+import logging
 import os
 import sys
 import threading
@@ -36,13 +37,19 @@
 import splunk_otel
 from splunk_otel.profiling import profile_pb2
 from splunk_otel.profiling.options import _Options
-from splunk_otel.util import _get_logger
 from splunk_otel.version import __version__
 
-logger = _get_logger(__name__)
+logger = logging.getLogger(__name__)
 
-thread_states = {}
-batch_processor = None
+
+class Profiler:
+    def __init__(self):
+        self.running = False
+        self.thread_states = {}
+        self.batch_processor = None
+
+
+profiler = Profiler()
 
 
 class StringTable:
@@ -128,7 +135,7 @@ def get_location(frame):
 
         labels = [timestamp_label, event_period_label, thread_id_label]
 
-        trace_context = thread_states.get(thread_id)
+        trace_context = profiler.thread_states.get(thread_id)
         if trace_context:
             (trace_id, span_id) = trace_context
 
@@ -167,8 +174,7 @@ def _profiler_loop(options: _Options):
 
     exporter = OTLPLogExporter(options.endpoint)
     # pylint: disable-next=global-statement
-    global batch_processor
-    batch_processor = BatchLogRecordProcessor(exporter)
+    profiler.batch_processor = BatchLogRecordProcessor(exporter)
 
     while True:
         profiling_stacktraces = []
@@ -208,7 +214,7 @@ def _profiler_loop(options: _Options):
             ),
             instrumentation_scope=InstrumentationScope("otel.profiling", "0.1.0"),
         )
-        batch_processor.emit(log_data)
+        profiler.batch_processor.emit(log_data)
         time.sleep(interval / 1e3)
 
 
@@ -222,7 +228,10 @@ def _wrapped_context_attach(wrapped, _instance, args, kwargs):
 
         if span:
             thread_id = threading.get_ident()
-            thread_states[thread_id] = (span.context.trace_id, span.context.span_id)
+            profiler.thread_states[thread_id] = (
+                span.context.trace_id,
+                span.context.span_id,
+            )
 
     return token
 
@@ -237,20 +246,28 @@ def _wrapped_context_detach(wrapped, _instance, args, kwargs):
             span = prev.get(_SPAN_KEY)
 
             if span:
-                thread_states[thread_id] = (span.context.trace_id, span.context.span_id)
+                profiler.thread_states[thread_id] = (
+                    span.context.trace_id,
+                    span.context.span_id,
+                )
             else:
-                thread_states[thread_id] = None
+                profiler.thread_states[thread_id] = None
         else:
-            thread_states[thread_id] = None
+            profiler.thread_states[thread_id] = None
     return wrapped(*args, **kwargs)
 
 
 def _force_flush():
-    if batch_processor:
-        batch_processor.force_flush()
+    if profiler.batch_processor:
+        profiler.batch_processor.force_flush()
 
 
 def _start_profiling(options):
+    if profiler.running:
+        logger.warning("profiler already running")
+        return
+
+    profiler.running = True
     logger.debug(
         "starting profiling call_stack_interval=%s endpoint=%s",
         options.call_stack_interval,

splunk_otel/profiling/options.py

@@ -12,11 +12,34 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 from os import environ
 from typing import Optional
 
 from opentelemetry.sdk.resources import Resource
 
+logger = logging.getLogger(__name__)
+_DEFAULT_CALL_STACK_INTERVAL = 1_000
+
+
+def _sanitize_interval(interval):
+    if isinstance(interval, int):
+        if interval < 1:
+            logger.warning(
+                "call stack interval has to be positive, got %s, defaulting to %s",
+                interval,
+                _DEFAULT_CALL_STACK_INTERVAL,
+            )
+            return _DEFAULT_CALL_STACK_INTERVAL
+
+        return interval
+
+    logger.warning(
+        "call stack interval not an integer, defaulting to %s",
+        _DEFAULT_CALL_STACK_INTERVAL,
+    )
+    return _DEFAULT_CALL_STACK_INTERVAL
+
 
 class _Options:
     resource: Resource
@@ -49,8 +72,8 @@ def _get_call_stack_interval(interval: Optional[int]) -> int:
             interval = environ.get("SPLUNK_PROFILER_CALL_STACK_INTERVAL")
 
             if interval:
-                return int(interval)
+                return _sanitize_interval(int(interval))
 
-            return 1_000
+            return _DEFAULT_CALL_STACK_INTERVAL
 
-        return interval
+        return _sanitize_interval(interval)

splunk_otel/tracing.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 import os
 import sys
 from typing import Collection, Dict, Optional, Union
@@ -26,9 +27,9 @@
 from pkg_resources import iter_entry_points
 
 from splunk_otel.options import _Options, _SpanExporterFactory
-from splunk_otel.util import _get_logger, _is_truthy
+from splunk_otel.util import _is_truthy
 
-logger = _get_logger(__name__)
+logger = logging.getLogger(__name__)
 
 
 def start_tracing(

splunk_otel/util.py

@@ -36,10 +36,8 @@ def _get_log_level(level):
     return levels[level.lower()]
 
 
-def _get_logger(name):
+def _init_logger(name):
     level = _get_log_level(os.environ.get("OTEL_LOG_LEVEL", "info"))
     logger = logging.getLogger(name)
     logger.setLevel(level)
     logger.addHandler(logging.StreamHandler())
-
-    return logger