(omega-design-logging)= # Logging ## 1 Overview The OMEGA logging system offers a standardized Application Programming Interface (API) that facilitates the tracking of events occurring during the runtime of all Omega sub-components. Each event comprises a descriptive message along with an indicator denoting its severity level. ## 2 Requirements ### 2.1 Requirement: Basic logging It is a requirement for the OMEGA logging system to furnish Omega sub-components with a mechanism to store log messages onto a file system, based on their corresponding severity levels. ### 2.2 Requirement: Inter-operability with E3SM It is a requirement for the OMEGA logging system to recognize the log root filename defined in E3SM as a part of the input configuration. ### 2.3 Requirement: Multi-processing It is a requirement that each process should have the capability to generate log messages independently. Additionally, there should be a mechanism in place to control the number of processes generating log messages simultaneously. This functionality is desired to ensure efficient management of system logs and prevent potential system overloading due to excessive logging activities. ### 2.4 Requirement: Multi-threading It is a requirement for the OMEGA logging system to provide thread-safe logging capabilities. This functionality is necessary to ensure that concurrent logging activities by multiple threads are executed in a synchronized manner, without causing data corruption or inconsistencies in the generated log files. ### 2.5 Requirement: Omega data types It is a requirement for the OMEGA logging system to support a subset of Omega data types, in order to generate well-formatted log messages that include these data types. This functionality is crucial to ensure that all relevant information is captured in the logs and to facilitate effective debugging and troubleshooting activities. ### 2.6 Requirement: Buffered file input/output It is a requirement for the OMEGA logging system to support both buffered and unbuffered file IO modes. Buffered mode reduces the overhead associated with file IO operations, improving performance and efficiency. Unbuffered mode is crucial for debugging purposes, as it allows for immediate writing of log messages to the file, preventing potential loss of unwritten messages that could lead to confusion. The API should provide an option for the user to choose between buffered and unbuffered modes. ### 2.7 Desired: Formatter It is desired for the OMEGA logging system to support various use-cases of log files by generating log messages through one of the installed message formatters. These formatters should be switchable at both compile-time and runtime to accommodate changing requirements. The default formatter should generate an unstructured text output, while other possible formatters should be able to generate structured output such as JSON. This functionality is essential to ensure flexibility in the logging system, allowing users to tailor the output format to their specific needs. ### 2.8 Desired: Output sinks It is desired for the OMEGA logging system to be capable of transporting log messages to various output sinks such as emails or databases. Thisfunctionality is crucial for ensuring the accessibility and availability of log data, enabling users to view and analyze the logs through multiple channels. ### 2.9 Desired: Local logging It is desired for the OMEGA logging system to provide each sub-component with its own local logger, which can be used to set parameters such as "severity," "message format," and "sink type" independently from other sub-components. This functionality is necessary to ensure that each sub-component has full control over its own logging activities, allowing for more granular and targeted logging output. ## 3 Algorithmic Formulation No algorithm is necessary for the logging system. ## 4 Design The spdlog logging library has been chosen as the core logging capability for the OMEGA logging system. spdlog is a feature-rich, header-only C++ logging library that is widely used and highly regarded in the industry. It offers a comprehensive set of features that satisfy most of the requirements, but additional E3SM and Omega specific requirements must be addressed. ### 4.1 Data types and parameters #### 4.1.1 Parameters The software design will implement a global switch for setting the minimum log severity and message format using predefined global C++ defines. The severity level can be set at compile-time using `-D OMEGA_LOG_LEVEL=`, where `` can be one of the predefined levels: `TRACE`, `DEBUG`, `INFO`, `WARN`, `ERROR`, or `CRITICAL`, where the severity increases from `TRACE` to `CRITICAL`. Log messages that have severity equal to or higher than the specified level will be stored in the log destination. Furthermore, the message format can be set using `-D OMEGA_LOG_PATTERN=` at compile time, where `` represents a format in the form of `%flag`, similar to the strftime function. This pattern defines the layout of the log message. Users can control which MPI ranks generate log files using `-D OMEGA_LOG_TASKS=` at compile time. The `` is either all for all tasks to generate log files, or comma-separated MPI rank numbers, or a range of MPI ranks with a dash. For example, `-D OMEGA_LOG_TASKS=0,2-3` indicates that MPI ranks 0, 2, and 3 generate log files. #### 4.1.2 Class/structs/data types No public data type is necessary for the logging system. ### 4.2 Methods The OMEGA logging system includes multiple macros that allow for logging at various severity levels. These macros guarantee that log messages with severity levels lower than the specified `OMEGA_LOG_LEVEL` will not be compiled. #### 4.2.1 Log write The OMEGA logging system includes macros that allow the user to write log messages with different severity levels. These macros follow the naming convention `LOG_`. For instance, the following macro can be used to write an informational log message: ```c++ LOG_INFO("log message"); ``` The OMEGA logging system provides the following macros for writing log messages with different severity levels: 1. `LOG_CRITICAL`: Used to indicate a critical error or issue that requires immediate attention and could potentially cause data loss or application failure. 2. `LOG_ERROR`: Used to indicate an error or issue that could cause application instability or incorrect behavior. 3. `LOG_WARN`: Used to indicate a warning or potential issue that should be addressed but does not necessarily require immediate attention. 4. `LOG_INFO`: Used to provide general informational messages about the application's operation or progress. 5. `LOG_DEBUG`: Used to provide detailed debugging information for developers to track down issues or analyze behavior. 6. `LOG_TRACE`: Used to provide the most detailed level of information, including function calls and variable values, and is typically used for deep analysis and debugging. In order to enhance the logging capabilities, OMEGA logging system provides each macro with an optional argument of an Omega data type. If the data type is recognized, it will be formatted automatically by the system and appended in the appropriate format to the log message. As an example, the following macro demonstrates how an argument of type `Array1DR8` can be appended to a log message: ```c++ Array1DR8 var; ... LOG_INFO("log message {}", var); ``` ## 5 Verification and Testing ### 5.1 Test OMEGA logging capabilities A unit test should be performed to verify all requirements, except for 2.2 (E3SM interoperability). The unit test should create multiple MPI ranks with multiple threads to ensure that requirement coverage is complete. The unit test should also include tests for both buffered and unbuffered logging capability, with various log message sizes and frequencies, to ensure that both modes work correctly and without loss of data. ### 5.2 Test OMEGA data type support A unit test should be performed to verify Requirement 2.5(Omega data types). ### 5.3 Test E3SM inter-operability Requirement 2.2 will be tested as part of larger system tests and will not have a dedicated unit test.