Figure 2 - uploaded by Antonio Mastropaolo
Content may be subject to copyright.
Log level distance in the predictions generated by LANCE. Zero indicates predictions having a correct level.
Source publication
Logging is a practice widely adopted in several phases of the software lifecycle. For example, during software development log statements allow engineers to verify and debug the system by exposing fine-grained information of the running software. While the benefits of logging are undisputed, taking proper decisions about where to inject log stateme...
Contexts in source publication
Context 1
... wrong level prediction made by LANCE could recommend the usage of Info instead of Fatal as well as the usage of Error instead of Fatal. However, these two errors have a different magnitude, with the first completely misleading the developer while the latter resulting in a sub-optimal (but still acceptable) log level decision. Fig. 2 depicts a histogram showing the number of instances in our test set for which the log level prediction had a distance from the target level going from 0 (i.e., the level was correctly predicted) to 5 (i.e., the worst-case scenario indicating a Trace recommended instead of Fatal or vice versa). While we report the results achieved by all models, ...
Context 2
... the level was correctly predicted) to 5 (i.e., the worst-case scenario indicating a Trace recommended instead of Fatal or vice versa). While we report the results achieved by all models, also in this case, we focus our discussion on the bestperforming one (Denoising-task). Note that not all 12,020 instances from our test set are depicted in Fig. 2. Indeed, besides the ones containing syntax errors (281 for the Denoising-task), we also had to exclude 557 instances for which the model did not recommend a valid log level, making impossible the computation of the ...
Context 3
... wrong level prediction made by LANCE could recommend the usage of Info instead of Fatal as well as the usage of Error instead of Fatal. However, these two errors have a different magnitude, with the first completely misleading the developer while the latter resulting in a sub-optimal (but still acceptable) log level decision. Fig. 2 depicts a histogram showing the number of instances in our test set for which the log level prediction had a distance from the target level going from 0 (i.e., the level was correctly predicted) to 5 (i.e., the worst-case scenario indicating a Trace recommended instead of Fatal or vice versa). While we report the results achieved by all models, ...
Context 4
... the level was correctly predicted) to 5 (i.e., the worst-case scenario indicating a Trace recommended instead of Fatal or vice versa). While we report the results achieved by all models, also in this case, we focus our discussion on the bestperforming one (Denoising-task). Note that not all 12,020 instances from our test set are depicted in Fig. 2. Indeed, besides the ones containing syntax errors (281 for the Denoising-task), we also had to exclude 557 instances for which the model did not recommend a valid log level, making impossible the computation of the ...
Similar publications
Developing a software service requires a strict software development life cycle and process. This process demands controlling all application code through source control management as well as a rigorous versioning and branching strategy. However, the platform and infrastructure also benefit from this rigor. Software services must be deployed to a t...