Starting a transaction can sometimes fail

[jjatria]

The code in Plack::Middleware::NewFangle assumes that start_web_transaction always returns a valid transaction. This is partially the case: it will always return an object you can call methods on, and the methods will only do useful things when the transaction is valid.

However, starting a transaction can sometimes fail. In those cases, you get a null transaction, and since the code does not check whether the transaction is valid or not, you end up with a lot of noise in the logs:

2021-10-28 14:53:11.882 +0100 (11 11) error: unable to start transaction
2021-10-28 14:53:11.882 +0100 (11 11) error: unable to add attribute for a NULL transaction
2021-10-28 14:53:11.882 +0100 (11 11) error: unable to add attribute for a NULL transaction
2021-10-28 14:53:11.882 +0100 (11 11) error: unable to add attribute for a NULL transaction
2021-10-28 14:53:11.882 +0100 (11 11) error: unable to start segment with NULL transaction

However, it does not seem like there is an easy way to check whether the transaction is valid or not: there is no method to check it in NewFangle, and the documentation for start_web_transaction does not actually mention the possibility of it retuning an invalid transaction:

Given an application pointer and transaction name, this function begins timing a new transaction. It returns a valid pointer to an active New Relic transaction, newrelic_txn_t. The return value of this function may be used as an input parameter to functions that modify an active transaction.