Regression in prometheus metric names introduced in v26.02

[caguado]

Situation

I am testing with the upcoming release of v26.02-rc2. I have noticed a regression in the prometheus exporter where some of the dimensions for metrics are renamed. In those metrics that used a hyphen to compose the node name, an underscore is now used. Please note that this change does not happen in the regular stats module.

Before

# vpp_prometheus_export
nodes_vectors{node="MultiGigaEthernet0_0-rx",index="0",thread="0"} 0
                                      --^

# vpp_get_stats
[0 @ 0]: 0 packets /nodes/MgmtLoopback0_0-rx/vectors

After

# vpp_prometheus_export
nodes_vectors{node="MultiGigaEthernet0_0_rx",index="0",thread="0"} 0
                                      --^

# vpp_get_stats
[0 @ 0]: 0 packets /nodes/MgmtLoopback0_0-rx/vectors

Expectation

No changes to the metric names or dimensions so existing history data and alarms/dashboards are preserved.

Cause

The change below has introduced sanitization to the metric dimension, but I cannot tell from the commit or prior discussion whether this is intended and why.

vpp/src/vpp/app/dump_metrics.c

Line 233 in eece09c

sanitize (tokens[2], lengths[2]);

Please advise as to what the intent is and I can propose a CL in gerrit.

[caguado]

When I remove the sanitization line referred in the description, I see back the metrics emitted with the same dimensions as before the change and no other visible impact on the same prometheus v2 exporter mode.

-		sanitize (tokens[2], lengths[2]);

[JimmyBjorklund]

I submitted a suggestion: https://gerrit.fd.io/r/c/vpp/+/44946/1