Skip to content

BatchLogRecordProcessor worker thread deadlock causing force_flush Timeout when shutting down LoggingHandler #3193

@nigel-gott

Description

@nigel-gott

Describe the bug

  • Configure a process with otel logging and BatchLogRecordProcessor and OTLPLogExporter, which sends logs in its own worker thread using requests.
  • Add a opentelemetry.sdk._logs.LoggingHandler using the batch processor to logging to intercept and send all logs
  • In the main thread, clear logging's registered handlers. For example this happens when you try to register a new logging config logging.config.dictConfig(NEW_CONFIG)
  • In the main thread:
    • logging will acquire logging._lock to prevent other concurrent changes to the config
    • logging will call .shutdown on opentelemetry.sdk._logs.LoggingHandler
    • BatchLogRecordProcessor tries to flush it's worker thread
  • Back in the worker thread, OTLPLogExporter will try to export remaining logs to the endpoint using requests
    • requests tries to do logger.debug
    • logger.debug blocks, waiting to acquire logging._lock, but it is now held by the main thread.
  • Finally, the main thread will time out with a Timeout was exceeded in force_flush(). warning.

Steps to reproduce

from opentelemetry._logs import set_logger_provider
from opentelemetry.exporter.otlp.proto.http._log_exporter import OTLPLogExporter
from opentelemetry.sdk._logs import LoggerProvider, LoggingHandler
from opentelemetry.sdk._logs._internal.export import BatchLogRecordProcessor
import logging.config
import logging

# To see the deadlock try using: https://github.com/niccokunzmann/hanging_threads
# pip install hanging_threads 
# 
# from hanging_threads import start_monitoring
#
# monitoring_thread = start_monitoring()


logger_provider = LoggerProvider()
set_logger_provider(logger_provider)
exporter = OTLPLogExporter()
logger_provider.add_log_record_processor(BatchLogRecordProcessor(exporter))
handler = LoggingHandler(
    level=logging.INFO,
    logger_provider=logger_provider,
)

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s [%(levelname)s] %(message)s",
    handlers=[handler, logging.StreamHandler()],
)

logger = logging.getLogger(__name__)
logger.info("test")

## reconfigure!
logging.config.dictConfig(
    {
        "version": 1,
        "disable_existing_loggers": False,
        "formatters": {
            "console": {
                "format": "%(levelname)s %(asctime)s %(name)s.%(funcName)s:%(lineno)s- %("
                "message)s "
            },
        },
        "handlers": {
            "console": {
                "class": "logging.StreamHandler",
                "formatter": "console",
            },
        },
        "root": {
            "handlers": ["console"],
            "level": "INFO",
        },
    }
)
# We now wait for the BatchLogRecordProcessor for the full timeout on its worker thread finishing the flush.
logger.info("test")

What did you expect to see?
The above script ideally shouldn't just block and hang. But i'm not exactly sure what the fix could be, or even if the fix instead should be raised against logging instead.

What did you see instead?
The above script hangs until the otel worker times out with a warning of Timeout was exceeded in force_flush()

What version did you use?
Version: 0.36b0

What config did you use?

receivers:
  otlp:
    protocols:
      grpc: # port 4317
      http: # port 4318

processors:
  batch:
    send_batch_size: 1000

exporters:
  otlp:
    endpoint: "api.honeycomb.io:443"
    headers:
      "x-honeycomb-team": "${env:HONEYCOMB_API_KEY}"
  otlp/metrics:
    endpoint: "api.honeycomb.io:443"
    headers:
      "x-honeycomb-team": "${env:HONEYCOMB_API_KEY}"
  logging:
    verbosity: detailed

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp,logging]
    metrics:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp/metrics,logging]
    logs:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp,logging]

extensions:
  health_check:

Environment
OS: Ubuntu 22.04 LTS

Additional context

All of the above is a constructed example. The real situation I hit this in was with trying to configure Django to work with otel logging. I was accidentally adding the Otel LoggingHandler prior to Django being setup. Django then would configure it's own logging by calling logging.config.dictConfig. All I initially experienced was the entire Django application failing to startup for the length of the timeout.

For future Django users encountering this problem, make sure you configure Django's LOGGING setting itself to work with the Otel handler and don't try to do something custom like I was. Alternatively make sure to setup the otel logging after django.setup has been run (say after the get_xyz_application is called in wsgi.py or asgi.py)

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workinglogging

    Type

    No type

    Projects

    Status

    To do

    Status

    Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions