This document concerns every package except azure-servicebus and azure-servicemanagement-legacy
This document covers the exceptions that you can get from the Azure SDK for Python, and will help you decide what is worth retrying or what is a critical failure, depending on your application.
Every exception related to the service content will be referenced in the docstring of the method. For instance, for this
you can see that you might receive a
Where the service team provides no specific sub-classing, the basic exception you can expect from an operation is
In some situations however, the exceptions are specialized by package (see azure-batch package which uses
If a specialized exception is used, it will always be a sub-class at some level of
Whether you should retry the same query or not for these exceptions depends on the service and the error content and cannot be generalized on this article.
For instance, an update to a SQL Database might be refused because you have another underlying operation (in this case you can retry); or a create request can be refused
because your selected name does not meet the pattern enforced by the server (in this case retry will never work). Use the error and response attributes of
CloudError to decide.
All these operations can also raise this set of generic exceptions that are defined in the
AuthenticationError: password invalid/expired, etc. Retry will likely not work.
ClientRequestError: requests library raised an exception (connection error, mostly HTTP level issues). Retry may work, though unlikely if your base_url is incorrect.
DeserializationError: unexpected RestAPI answer. Retry will likely not work. Please create an issue on Github if you see this exception.
HttpOperationError: bad HTTP status code from Azure. See inner exception to decide if it’s worth retrying.
SerializationError: unable to serialize your request. Your Python code didn’t respect the expected model. Retry will never work. Please create an issue on Github if you see this exception and are 100% your parameters are correct.
TokenExpiredError: please renew your credentials. Likely retry with new credentials will be ok.
ValidationError: client side check of your request failed. Fix your parameters with expected value. For instance, you didn’t respect the regexp for the account name. This will never work on retry.
An asynchronous operation is an operation that returns an
create_or_update). Using this kind of operation
usually requires two lines:
async_poller = client.network_security_groups.create_or_update(myparameters) result = async_poller.result()
or, if this asynchronous operation is not returning a result:
async_poller = client.network_security_groups.create_or_update(myparameters) async_poller.wait()
Our recommendation is to surround both of the statements with the necessary try/except. More precisely, the first call might fail on the initial call and the second one might fail during polling the status of the operation
Old versions of the packages never failed on the first call, but this behavior was replaced by the one described and you should follow this pattern even for old packages.