[redis_enterprise_prometheus] Fix dashboard queries and collect two missing metrics

[slorello89]

What does this PR do?

Fixes default-dashboard widgets in redis_enterprise_prometheus that return no data, and collects two metrics that were documented in metadata.csv but missing from the check's metric map.

Check

• Collect rdse2.node_uname_info and rdse2.node_config. Both were already in metadata.csv but absent from metrics.py, so they never emitted. node_uname_info carries the per-node nodename label — the internal hostname the Node dashboard needs — and node_config carries rs_version.

Dashboards

• Database List: database_syncer_config → db_config. The former is a configuration-label placeholder that only exists on Active-Active clusters, and it lives in the opt-in REDIS2.REPLICATION group, so the widget was empty on a default install.
• Database Input/Output: add the missing .count suffix to endpoint_ingress / endpoint_egress. These are Prometheus counters, which OpenMetrics v2 emits as .count.
• Connections: compute current connections from the .total gauges and subtract endpoint_proxy_disconnections.total, rather than differencing two monotonic .count deltas (which yields net change over the interval, not a current count).
• Double rate division: Database Input/Output, Shard Process CPU, and Proxy Threads CLI Session each wrapped an already-rated .as_rate() query in per_second() / derivative(), dividing by time twice.
• Ingress/egress labels were swapped in Database Input/Output — alias egress pointed at the ingress query. The accompanying note text had the definitions backwards too.
• Node Latency: rebuilt as a calculated field over the endpoint_*_requests_latency_histogram sum/count pairs. It previously queried rdse.node_avg_latency — a V1 metric, under the rdse. namespace rather than rdse2.. Histogram buckets run 0.25–16908, so the unit is microseconds and the formula divides by 1000 for milliseconds.
• Cluster Nodes: group by nodename from node_uname_info instead of the internal-hostname tag, which is not present on any /v2 metric.
• Node Network Traffic: removed. Both rdse.node_egress_bytes_median and rdse.node_ingress_bytes_median are V1-only with no V2 equivalent.
• Note widgets: dropped title, folding it into the content as a markdown heading. The Dashboards API rejects title on note widgets, which made every one of these dashboard assets fail to import.

Motivation

Datadog raised a list of redis_enterprise_prometheus dashboard widgets showing no data. Each item was verified against a live Redis Enterprise 8.0.10-76 /v2 endpoint rather than against the repo fixture, which turned out to be stale in places — for example process_resident_memory_bytes and process_virtual_memory_bytes exist in tests/data/metrics.txt but are gone from RS 8.0.10 entirely.

Two findings worth calling out:

• The internal-hostname tag genuinely does not exist on any /v2 metric, but node_uname_info already exposes the same information as nodename, so no upstream Redis change is needed.
• ddev validate dashboards passes both before and after the note-widget fix, so it does not currently catch assets that the Dashboards API will reject on import.

Review checklist

• PR has a meaningful title or PR has the no-changelog label attached
• Feature or bugfix has tests
• Git history is clean
• If PR impacts documentation, docs team has been notified or an issue has been opened on the documentation repo
• If this PR includes a log pipeline, please add a description describing the remappers and processors.

Additional Notes

Dashboard asset changes are not unit-testable, so no tests are added; each query was instead validated against a live cluster. Two caveats on that validation:

• The cluster used had a single database, so the Node Latency formula could not be checked against a real latency value — the microsecond assumption comes from the histogram bucket boundaries. Worth a second look from someone with a busier cluster.
• Three widgets on the Shard dashboard (Resident Memory, Virtual Memory, Process CPU) are still broken and are not addressed here. Resident/Virtual Memory query rdse2.redis_process_*, which has never existed under any name; the viable source is namedprocess_namegroup_memory_bytes{memtype="resident"|"virtual"}. Process CPU has data but groups by {db,redis} on namedprocess_* series that carry neither tag (shards appear as groupname="redis-<n>"). Fixing those changes the widgets' semantics from shard-scoped to process-group-scoped, which felt like a separate discussion.

Replacement for #3136, which was automatically closed as stale.