New file replica state is depending on the source node

[yiiyama]

PHEDEX/perl_lib/PHEDEX/Infrastructure/FilePump/Agent.pm

Lines 298 to 311 in d2602ab

	&dbexec($dbh, qq{
	merge into t_xfer_replica xr using
	(select xt.to_node, xt.fileid, xtd.time_update, fn.kind
	from t_xfer_task_harvest xth
	join t_xfer_task xt on xt.id = xth.task
	join t_xfer_task_done xtd on xtd.task = xth.task
	join t_adm_node fn on fn.id = xt.from_node
	where xtd.report_code = 0) new
	on (xr.node = new.to_node and xr.fileid = new.fileid)
	when not matched then
	insert (id, node, fileid, state, time_create, time_state)
	values (seq_xfer_replica.nextval, new.to_node, new.fileid,
	case when new.kind = 'MSS' then 0 else 1 end,
	new.time_update, new.time_update)});

New entries to t_xfer_replica are created by FilePump. The state field is set to 0 or 1 depending on the kind of the source node. I think new.kind was supposed to be the kind of the destination node (which would make more sense from the name new too..).

If this is the case, I think we also need to go in to TMDB and set the state of all file replicas on Disk and Buffer nodes to 1. About half of replicas on Disk and almost all of them on Buffer are state 0 at the moment.

Just to make sure, I grepped the entire repository for parts that are dealing with t_xfer_replica.state. The only component that would ever set state to 0 is Stager, which I believe should be handling only MSS nodes.