Abnormally long and unexpected STP (RPVST+) interface cost on a switch

[Rack&Stack]

I'm seeing a bizarre behavior where one of my switches is showing spanning tree costs much higher than I'd expect. Here's a simple diagram of the network:

The root switch election, root port election, and designated port elections all happen as expected. Where I'm confused is when I look at the cost on DC2. For reference, I'm using spanning-tree pathcost method long on all of these switches.

Here's the spanning tree information for vlan 500 on DC2:

The costs are much higher than I'd expect.. I'd expect to see a cost of 2000 because of the 10Gbps link to the root. It seems as if an extra 1000 was appended to the front of the cost number. I've already checked the interface configurations for this switch, there are no manual costs set.

DC1 looks normal and shows what I'd expect:

Any ideas on what could cause this?

[MHM Cisco World]

it seem that one SW use long and other use short. can you double check the STP pathcost method 

[Rack&Stack]

DC1 and DC2 are both configured with spanning-tree pathcost method long. They should both show the same cost to root but they do not.

[MHM Cisco World]

what about core ?

[Rack&Stack]

The core is on pathcost method short, because it's an old switch that I'm going to replace. But that technically shouldn't matter.. it appears DC2 is the switch at fault because it's interpreting it's own local interface costs as being insanely high for some reason. DC1 isn't having any issues, so that rules out Core in my head.. but I'm possibly missing something. 

[MHM Cisco World]

I dont think So I will run lab and check'

If one SW support short then we may have abnormal behavior'

Let me check 

[MHM Cisco World]

I run lab and one of SW return to use short even if I use long, but that solve after I add command again 
spanning tree pathcost method long 

NOTE:- I dont recommend mix pathcost in same STP domain. 
thanks 

[Rack&Stack]

I know it's not recommended but in this case, I do not think it's the problem. On the 9500-24Y4C switches running 17.6.5, pathcost long is the default method. 

[Reza Sharifi]

Are DC1 and DC2 the same type of switches with the same IOS version? Maybe a bug in DC2's IOS.

HTH

[Rack&Stack]

Catalyst 9500-24Y4C's both on 17.6.5 code. I see the same issue on a different switch (Catalyst 9300 on 16 code) having the same issue that I didn't include in this topology. I'm leaning towards something hidden in the configuration that's doing this.

[Reza Sharifi]

Seems like a bug. If it is a bug, sometimes removing the config and reapplying it fixes things.

HTH