Google Summer of Code

radosgw-admin UX and documentation improvements

Mentors: Yuval Lifshitz (ylifshit@ibm.com), Jacques Heunis (jheunis@bloomberg.net) · Difficulty: Advanced · Hours: 350 · Component: RGW · Skills: C++, maybe Python

Documenting radosgw-admin commands is currently a manual and error-prone process: after implementing a new command, the usage text, man page, and admin guide must each be updated by hand, and they can drift apart. This project takes a programmatic approach:

• Use a CLI and argument framework that supports auto-generating context-aware usage documentation
• Investigate auto-generating the man page and admin guides from it
• Make these command descriptions referenceable elsewhere in the documentation
• Maintain backward compatibility with existing behavior throughout

The first two weeks involve researching command line frameworks and learning the internals of radosgw-admin. See the tracker issue, applicant evaluation steps, and rules for AI usage.

Calls a mentee could attend: RGW standup, RGW refactoring meeting.

Kafka security for bucket notifications

Mentor: Yuval Lifshitz (ylifshit@ibm.com) · Difficulty: Advanced · Hours: 350 · Component: RGW · Skills: Python and C++

Bucket notification integration with Kafka is a very useful RGW feature, but some security capabilities needed for such integrations are missing: GSSAPI, OAUTHBEARER, mTLS, and passing in a CA without a file (useful for Rook integration). The main challenge is automating the tests so they run easily in local environments, with a stretch goal of running them in teuthology.

The first two weeks involve getting to know the bucket notifications code and learning about Kafka security. See the applicant evaluation steps and rules for AI usage.

Calls a mentee could attend: RGW daily standup, RGW weekly refactoring meeting.

RGW tcmalloc profiling

Mentor: Yuval Lifshitz (ylifshit@ibm.com) · Difficulty: Advanced · Hours: 90 · Component: RGW · Skills: Python and C++

All Ceph daemons use tcmalloc as their memory allocator, and a recent PR added the ability to profile how tcmalloc performs inside RGW. This project uses that profiling data to tune tcmalloc for RGW's memory patterns: making better use of thread-local caches for small allocations, keeping overall memory consumption low, returning unused memory to the OS, and exploring further runtime or compile-time parameters.

The first two weeks involve understanding the different RGW workloads, simulating them, and establishing a performance baseline. See the applicant evaluation steps and rules for AI usage.

Calls a mentee could attend: RGW daily standup, RGW weekly refactoring meeting.