Get started with error reporting

Context

Having a central place where we can gather errors and warnings from our R code is particularly relevant when we have deployed our code and have users actively making use of the services we have set up. Users will not always bother to report an error, and in the case of API’s where the users are often other machines, this is even more unlikely.

Infrastructure as a Service (IaaS) platforms such as Google Cloud Platform (GCP) typically have monitoring and loggin services available. In the case of GCP this service is called Google Error Reporting. This package offers you a convenience wrapper around the projects.events API to report errors of your services or any other R process that you would like to monitor. The GCP Google Error Reporting service offers a nice interface where you can see the errors and assign them status (Open, Acknowledged, Resolved, Muted) and include a link to an issue.

Message format

The key concept that the Google Error Reporting API uses is that of a message. This message contains all the information about the error that you want to send to the API. As a side note: the message does not need to be an error, it could if you find a need for it also be a confirmation of success, should it help your debugging to have that show up amongst the messages.

For some language Google has implemented functionality to capture the stacktrace and use that as a basis for a report. When using R code we need to create the erorr message, but note that we can still include dynamic content. For instance, we can setup a tryCatch() call and output the e (error) to a Error Reporting call. We will show an example later on.

The message is small json files with nested fields that we send along in the body of the call to the API. To set the values in R we set these as elements in a list, where the complete list can then be transformed to json to comply with the documented ReportedErrorEvent format.

message$message <- message

message$serviceContext$service <- service
message$serviceContext$version <- version

message$context$httpRequest$method <- method
message$context$httpRequest$url <- url
message$context$httpRequest$userAgent <- user_agent
message$context$httpRequest$referrer <- referrer
message$context$httpRequest$responseStatusCode <- response_status_code
message$context$httpRequest$remoteIp <- remote_ip

message$context$user <- user_id

message$context$reportLocation$filePath <- filepath
message$context$reportLocation$lineNumber <- line_number
message$context$reportLocation$functionName <- function_name

For convenience we included the format_error_message() function, that will instantiate an error message as a list with the minimum configuration.

message$eventTime <- format(lubridate::now(), "%Y-%m-%dT%H:%M:%SZ")