Monitoring Python Applications With Kieker (Poster).

Python is a widely used programming for applications, web services, and, especially, in scientific computing and data science. In the context of our project OceanDSL, which aims to provide DSLs for ocean system models, we need to comprehend existing scientific software based on static and dynamic code analysis. Thus, we decided to provide monitoring support for Python utilizing the existing Kieker analysis toolchain. As the code base is rather extensive, manually injecting probes is not a viable solution. Thus, our implementation relies on code weaving approaches. The Kieker Language Pack for Python follows, in principle, the Kieker architecture for monitoring with a reduced feature set [1]. It comprises (a) event types, (b) code to control probes and data storage, (c) probes to instrument the code, and (d) a techniques to introduce probes into a program without modifying the code manually. Event types for Python can be generated utilizing the Kieker instrumentation record language (IRL) which we extended to support Python [2]. Currently, the event types consist, like their Java counterparts, of a set of constants implementing default values, attributes, a constructor and a serialization method which utilizes a serialization helper to support multiple formats (cf. Listing 1). As the language pack is only used to monitor and log events in Python applications, it does not support features used to deserialize and manage events. However, this can be added later, if needed.