[AxiomHandler] Error thrown on .log() rather than initialization when dataset name is invalid

[kytpbs]

When using AxiomHandler with an incorrect dataset name but valid API credentials, the error is not detected during handler initialization. Instead, the error only occurs later when attempting to log a message, throwing: axiom_py.client.AxiomError: API error 404: dataset not found

Simple Code to reproduce:

client = axiom_py.Client(CORRECT_TOKEN)
handler = ErrorHandlingAxiomHandler(client, "INCCORECT_NAMED_DATASET")
logging.error("this line throws an error")

Possible Solutions:

1. Test flush in constructor:

def __init__(self, client: Client, dataset: str, level=NOTSET, interval=1):
    super().__init__()
    # Set urllib3 logging level to warning, check:
    # https://github.com/axiomhq/axiom-py/issues/23
    # This is a temp solution that would stop requests library from
    # flooding the logs with debug messages
    getLogger("urllib3").setLevel(WARNING)
    self.client = client
    self.dataset = dataset
    self.buffer = []
    self.interval = interval
    self.last_flush = time.monotonic()

    # Check if the dataset is valid before starting the handler, if not raise an error.
    try:
        self.client.datasets.get(self.dataset)  # This will raise an error if the dataset does not exist
    except axiom_py.client.AxiomError as e:
        raise axiom_py.client.AxiomError(e.status, axiom_py.client.AxiomError.Response(f"{self.dataset}, does not exist", None))

    # We use a threading.Timer to make sure we flush every second, even
    # if no more logs are emitted.
    self.timer = Timer(self.interval, self.flush)

    # Make sure we flush before the client shuts down
    def before_shutdown():
        self.flush()
        self.timer.cancel()

    client.before_shutdown(before_shutdown)

2. Log by temporarily removing self from root

Logging to root logger as a library is bad practice, but it's way better than throwing an exception when logging fails

def flush(self):
    """Flush sends all logs in the buffer to Axiom."""

    self.last_flush = time.monotonic()

    if len(self.buffer) == 0:
        return

    local_buffer, self.buffer = self.buffer, []
    try:
        self.client.ingest_events(self.dataset, local_buffer)
    except axiom_py.client.AxiomError as e:
        # logging to root logger as a library is really bad practice, but it is possible to do it this way:
        # but its way better than throwing an exception when logging fails
        if self in getLogger().filters:
            # This is a handler to the root logger, so we remove it temporarily to log the error
            getLogger().removeHandler(self)
            logging.error("Failed to log to Axiom, most probably because DB name is wrong", exc_info=e)
            getLogger().addHandler(self)
        else:
            # we don't know which logger this handler is attached to, so we just log the error to root
            # which shouldn't go back to this handler since it is not attached to the root logger
            logging.error("Failed to log to Axiom, most probably because DB name is wrong", exc_info=e)

3. SilentFail:

def flush(self):
   """Flush sends all logs in the buffer to Axiom."""

    self.last_flush = time.monotonic()

    if len(self.buffer) == 0:
        return

    local_buffer, self.buffer = self.buffer, []
    try:
        self.client.ingest_events(self.dataset, local_buffer)
    except axiom_py.client.AxiomError:
        pass  # dataset name is probably set wrong or some other connection error accured

Recommendation

Solution 1 is the most robust as it fails fast and provides clear error messages at configuration time rather than during logging operations. This makes it much easier to debug configuration issues.

Solution 2 is bad practice since its using the root logger, but it is still better than silent fail (solution 3) since you at least get a error message warning you that axiom is not setup correctly

Solution 3 is a very temporary solution and will cause hours if not days of debugging on some people for sure

[kytpbs]

I can implement whichever solution you guys would prefer. Just let me know which I should open a PR with

I have also tested solution 1 by modifying the library code locally