Limit fedmsg error emails to 100 per system.

This commit is contained in:
Ralph Bean 2015-05-08 19:08:45 +00:00
parent c3005d4340
commit b76f995492

View file

@ -2,6 +2,7 @@
# All of these modules are just used by the ContextInjector below.
import inspect
import hashlib
import logging
import os
import socket
@ -17,6 +18,14 @@ except (OSError, ImportError):
pass
radio_silence = """
*** %i instances of this error seen. No more mail will be sent. ***
""".strip()
seen_errors = {} # This could be a default dict.
error_limit = 100
class ContextInjector(logging.Filter):
""" Logging filter that adds context to log records.
@ -24,12 +33,15 @@ class ContextInjector(logging.Filter):
method that can return True or False. Only records with 'True' will
actually be logged.
Here, we somewhat abuse the concept of a filter. We always return true,
but we use the opportunity to hang important contextual information on the
log record to later be used by the logging Formatter. We don't normally
want to see all this stuff in normal log records, but we *do* want to see
it when we are emailed error messages. Seeing an error, but not knowing
which host it comes from, is not that useful.
Here, we somewhat abuse the concept of a filter. We for the most part
return true, but we use the opportunity to hang important contextual
information on the log record to later be used by the logging Formatter. We
don't normally want to see all this stuff in normal log records, but we *do*
want to see it when we are emailed error messages. Seeing an error, but not
knowing which host it comes from, is not that useful.
After we've seen an error ~100 times, we stop sending email to avoid choking
the world.
http://docs.python.org/2/howto/logging-cookbook.html#filters-contextual
"""
@ -44,6 +56,20 @@ class ContextInjector(logging.Filter):
record.proc_name = current_process.name
record.command_line = " ".join(current_process.cmdline)
record.callstack = self.format_callstack()
record.farewell = ""
key = hashlib.sha256(record.callstack).hexdigest()
if not key in seen_errors:
seen_errors[key] = 0
if seen_errors[key] > error_limit:
return False
seen_errors[key] += 1
if seen_errors[key] > error_limit:
record.farewell = radio_silence % error_limit
return True
@staticmethod
@ -86,6 +112,7 @@ hefty_format = """Message
-------
[%(asctime)s][%(name)10s %(levelname)7s]
%(message)s
%(farewell)s
Process Details
---------------