get_dependency

[ghost]

Would it be possible to add a get_dependency method to return the raven client? For times when you want to send a non exception message.

client.captureMessage('Somebody did something sorta strange.')

[mattbennett]

I understand the desire here. I wonder if a better way to do it would be to use a SentryHander and the logging framework.

I've just raised a PR against nameko https://github.com/onefinestay/nameko/pull/313 that loads LOGGING config on service startup.

The following config.yaml configures nameko_sentry and adds a logging SentryHandler at the same time.

SENTRY:
    DSN: &sentry_dsn
        https://user:pass@app.getsentry.com/12345
    CLIENT_CONFIG: &sentry_client_config
        site: 'Test Site'

AMQP_URI: amqp://guest:guest@localhost/
LOGGING:
    version: 1
    disable_existing_loggers: false
    formatters:
        simple:
            format: "%(levelname)s:%(name)s - %(message)s"
    handlers:
        console:
            class: logging.StreamHandler
            formatter: simple
        sentry:
            <<: *sentry_client_config
            dsn: *sentry_dsn
            class: raven.handlers.logging.SentryHandler
        sentry_restricted:
            <<: *sentry_client_config
            dsn: *sentry_dsn
            level: WARNING  # restricted to WARNING and above
            class: raven.handlers.logging.SentryHandler
    loggers:
        sentry:
            level: INFO
            handlers: [sentry]
            propagate: False
    root:
        level: INFO
        handlers: [console,sentry_restricted]

With this particular config, warnings and above in any logger go to sentry, and there's a specific sentry logger will sends info level and above, e.g.

import logging

logging.getLogger("app").info("logs to console only")
logging.getLogger("sentry").info("logs to sentry only")
logging.getLogger("nameko").warn("logs to console and sentry")

The only snag with the above is that the built-in raven.handlers.logging.SentryHandler is incompatible with eventlet. It works but you get a traceback on teardown. This can be fixed by subclassing the built-in one to force the handler to use the eventlet http transport:

from raven.handlers.logging import SentryHandler as RavenSentryHandler
from raven.transport import EventletHTTPTransport

class SentryHandler(RavenSentryHandler):
    def __init__(self, *args, **kwargs):
        kwargs['transport'] = EventletHTTPTransport
        super(SentryHandler, self).__init__(*args, **kwargs)

If this is a useful solve for your problem we could add the above subclass into nameko_sentry.

[ghost]

Using the logging module seems like a more natural approach.

There is one other case for having direct access to the raven client. Typically during an exception or message I'll merge is the user context like so...

self.client.context.merge({'user': player_id})

Which then populates the user detail in sentry. This could probably be handle by exposing a method in the dependency to merge context.

[mattbennett]

Using context.merge before sending an exception doesn't seem quite right. The client object is shared so if you have exceptions in multiple concurrent workers there's no guarantee that the merge will happen at the correct time.

It would be better to pass the user information when you call captureException. As it stands that requires a small modification to nameko-sentry, or a subclass. Using the logging handler though, it looks like you can just provide it using the extra kwarg.

As an aside, are you keeping your user_id in the data attribute of the worker_ctx? There's a general purpose solution whereby all information in the "context data" gets passed into the sentry client.

[mattbennett]

This is addressed by #7.

As of nameko 2.5.1, all worker lifecycle methods happen within the same greenthread, meaning that raven's thread locals keep worker contexts isolated. As a result we can safely expose the client to the worker.

Also note that #7 adds the ability to automatically suck "user" type data out of nameko's WorkerContext.context_data and add it to the appropriate raven context.

I'll write up some docs and cut a release soon.