some questions in dev-sessions about self-heal

[httaotao]

In dev-session #16 and #17, there are some points I am not sure.
1.if one file is dirty that trusted.afr.dirty is not zero and other file is health , then when heal process is running, which brick would choose to be the sources to repair ? Does it choose the first brick of the health brick list ?
2. in the video, it mentioned that afr_selfheal_inodelk fn will get the full lock ,but if there are other fop requests coming , does it will reject the requests ?
thanks.

[itisravi]

1. If only trusted.afr.dirty is non-zero and there is no trusted.afr.xxxx-client-x, then it will use the heuristics in __afr_selfheal_data_finalize_source to pick one as source (ie. afr_mark_largest_file_as_source, afr_mark_biggest_witness_as_source, afr_mark_newest_file_as_source etc.). This is for data heal. Metadata heal will be something similar- see __afr_selfheal_metadata_finalize_source.
2. afr_selfheal_inodelk will first attempt non-blocking lock (F_SETLK). If it fails, it will proceed to do blocking lock (F_SETLKW). So assuming shd is currently holding the lock, any fop requests asking for lock in the overlapping {offset,size} will be put in a blocked queue in the locks xlator. When shd releases the lock, the fop will acquire it. There is no rejection.

[itisravi]

When shd releases the lock, the fop will acquire it. There is no rejection.

Note that this is because the fop (from the mounts) also follows the same pattern - it first does non-blocking lock and if it fails, it goes for blocking lock. See afr_lock() which is called during afr write transaction from the mounts.