[Gluster-devel] Review request for patch: libglusterfs/syncop: Add xdata to all syncop calls

[Jeff Darcy]

> Is it always ok to consider xdata to be valid, even if op_ret < 0?

I would say yes, simply because it's useful.  For example, xdata might
contain extra information explaining the error, or suggesting a next
step.  In NSR, when a non-coordinator receives a request it returns
EREMOTE.  It would be handy for it to return something in xdata to
identify which brick *is* the coordinator, so the client doesn't have
to guess.  It might also be useful to return a transaction ID, even
on error, so that a subsequent retry can be detected as such.  If
xdata is discarded on error, then both the "regardless of error" and
"only on error" use cases aren't satisfied.

> If yes, I will have to update the syncop_*_cbk calls to ref xdata
> if they exist, irrespective of op_ret.
> 
> Also, it can be used in really cool ways, like we can have a
> key called glusterfs.error_origin_xlator set to this->name of the xlator
> where error originated and master xlators (fuse and gfapi) can log / make
> use of it etc.

Great minds think alike.  ;)  There might even be cases where it
would be useful to capture tracebacks or statistics to send back
with an error.