I spent a few days this week discussing CephFS with some new contributors working at Red Hat, and I managed to record some of it. We decided to share the recordings publicly, but let me be clear about what these are not: *) These are not prepared. There are no slides; I say "um" a lot *) These are not the word of god. I said untrue things and caught it/got caught a few times, but probably still lied somewhere else that wasn't noticed. *) These are not user discussions or designed as an intro to CephFS. If you aren't comfortable following the source code and trying to relate it to concepts in Sage's thesis (http://ceph.com/papers/weil-thesis.pdf) on your own, these won't help you. What these are, is about 8 hours of a fairly one-sided conversation about the CephFS source code and some of its design. All that said, the video links and an outline of their contents: https://bluejeans.com/s/9QN_/ Chapter 1: The messenger/dispatcher model; how mknod works from the client side https://bluejeans.com/s/9QO5/ Chapter 1: client-side structs: Inode, Dentry, Fh, MetaSession server-side: MDS ranks versus daemons, and their major components; MDS dispatch routing; handle_mds_map; Mutation/MDRequestRefs; the beginnings of mknod; traversing the cache Chapter 2: Selecting and acquiring locks; more mknod; bunches of metadata projecting/logging code; some file cap handling https://bluejeans.com/s/9QRO/ Unfortunately, starting from this point I no longer have video of the terminal I was using to display code. Stuff will be a lot less useful without that, but I think I still reference enough function names for dedicated listeners to follow along. :) Chapter 1: setattr code from both sides; file capabilities and flushing from both sides; locks.c (our lock transitions) Chapter 2: metadata on-disk storage; encoding functions Chapter 3: unlink, from the client and server; leading into file purging code (stray directories!); beginning of layouts and backtraces https://bluejeans.com/s/9QTk/ again, no terminal content Chapter 1: purging and backtraces finished; rename on both client and server Chapter 2: This has got to be totally unintelligible rambling about snapshots, since I gave up on displaying code and started drawing on a whiteboard If anybody does manage to watch these and get something out of them, feel free to ask questions either here on the list or on #ceph-devel. :) -Greg -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
