chore(py3): Bump structlog

[evanpurkhiser]

Version 17.1.0 is the first to support python3. We were on 16.1, this is a single release bump.

Notable changes:

• hynek/structlog@cf5d57c - ConsoleRenderer has become the default processor.

  This does not affect us as we have already configured our own processors:

  sentry/src/sentry/runner/initializer.py

  Lines 220 to 226 in 9791f3b

  	"processors": [
  	structlog.stdlib.add_log_level,
  	structlog.stdlib.PositionalArgumentsFormatter(),
  	structlog.processors.format_exc_info,
  	structlog.processors.StackInfoRenderer(),
  	structlog.processors.UnicodeDecoder(),
  	],

• hynek/structlog@705d3a3 - The TimeStamper processor now uses more accurate timestamps.

  We do NOT use this processor

• hynek/structlog@065b69a - ProcessorFormatter is added. This is purely an addition.

  We do NOT use this processor

• hynek/structlog@e4731ba. - The PositionalArgumentsFormatter (which we DO use), has had a minor fix for PositionalArgumentsFormatter does not remove empty list of args hynek/structlog#82

  PositionalArgumentsFormatter with remove_positional_args=True is supposed to remove the positional_args from the event dict. However this only happens when positional_args is truthy which is not the case for the empty list.

  This is a minor change and shouldn't break anything

Pretty much all of the other changes are documentation changes, or minor code changes.

Overall this bump should be trivial.

[joshuarli]

There's barely any coverage. Only tests/sentry/logging/test_handler.py which doesn't do much.

We seem to have customized structlog pretty deeply... have a look at configure_structlog. We use simplejson in our own JSONRenderer, and simplejson vs stdlib json is an entirely separate python3 consideration, so I think bumping structlog is going to need some more investigation and possibly we should block this with simplejson vs stdlib json.

FWIW there is a https://www.structlog.org/en/stable/api.html#structlog.processors.JSONRenderer, I don't know if there's a reason we roll our own. Maybe @JTCunning can shed some light on that one.

[joshuarli]

Other considerations include:

# These are values that come default from logging.LogRecord.
# They are defined here:
# https://github.com/python/cpython/blob/2.7/Lib/logging/__init__.py#L237-L310
throwaways = frozenset(
    (
        "threadName",
        "thread",
        "created",
        "process",
        "processName",
        "args",
        "module",
        "filename",
        "levelno",
        "exc_text",
        "msg",
        "pathname",
        "lineno",
        "funcName",
        "relativeCreated",
        "levelname",
        "msecs",
    )
)

[evanpurkhiser]

Yeah I'm already reading all of this and talking with James :)

[evanpurkhiser]

possibly we should block this with simplejson vs stdlib json.

I think there's no problem bumping structlog to a known python3 compatible version (which is really all this is, since the version bump just brings us to a version where py3 tests are running). When we bump simplejson we can circle back to this and make sure things look OK.

[joshuarli]

possibly we should block this with simplejson vs stdlib json.

I think there's no problem bumping structlog to a known python3 compatible version (which is really all this is, since the version bump just brings us to a version where py3 tests are running). When we bump simplejson we can circle back to this and make sure things look OK.

Yeah, I agree.

[wedamija]

This looks safe to me. Browsing through their changes it's mostly docs and things that shouldn't affect us