enforce id-based ordering before attempting to obtain partition locks.

[emmettbutler]

for #253

[ltucker]

I think you may want to recompute this within your retry loop since owned_partition_offsets is mutated by the loop.

[emmettbutler]

@yungchin This could use a sanity check.

[yungchin]

@emmett9001 is the sorting in fetch() strictly necessary? As that code just skips over any partitions it cannot grab a lock on, doesn't that mean there shouldn't be an issue with deadlocks there? Other than that, this looks good to me.

[emmettbutler]

No, I don't think it's strictly necessary. I figured sorting is a good idea in general because it makes the whole system more predictable.

[yungchin]

Ok, yeah that seems fine to me - I don't imagine the sorting is very expensive.

I've not been able to reproduce the bug here, so can't confirm that it's fixed, but I'm convinced that sorting before taking all the locks is a very real improvement either way.