[onewland]

Two refinements to the System Queries node search from #8609, based on how production hostnames and storages actually look.

Folding storages that share nodes. In production, many storages use the same nodes, so the storage line under each node and the storage pickers repeat the same cluster under many names. Storage lists now show one storage per storage set. A storage set belongs to exactly one cluster (snuba/clusters/cluster.py asserts it), so its storages always share their nodes. /clickhouse_nodes now reports each storage's storage_set, and the shortest name in each set stands in for the rest:

Storage set	Shown as
events_analytics_platform	eap_items (for eap_items_2, the downsample_* ones, eap_item_co_occurring_attrs, …)
events_analytics_platform_ro	eap_items_ro (for the _ro versions)
outcomes	outcomes_raw (for outcomes_hourly, outcomes_daily, outcomes_cross_org_hourly, llm_proxy_cost_raw)
metrics	metrics_raw (for metrics_counters, metrics_distributions, metrics_sets, org_metrics_counters)
functions	functions
replays	replays

Folding only changes what's listed:

• The representative is picked from the storages in the list, so a node that lists eap_items_2 but not eap_items shows eap_items_2.
• The selected storage always stays in the picker, so links to e.g. eap_items_downsample_8 still work.
• The storage sent with a query is still chosen from all of the node's storages.

The first commit hard-coded this for eap_items and outcomes by name. The second replaces that with the storage set, which the config already maintains, so new storages fold without any change here. I considered an admin-only alias or hide flag in the storage YAML. It would mostly repeat what set_key already says, and it needs a schema change and flags on about 20 storages. If a real exception comes up, a small hidden override would be enough.

Skipping a host word. Hosts like snuba-events-analytics-platform-arm-1-1 and …-mz-1-1 matched eap-arm-1-1 but not eap-1-1. A typed separator can now skip one whole host word that starts with a letter:

• Each skipped word counts as a boundary miss, so exact …-1-1 hosts still rank first.
• Number words (shards, replicas, ports) are never skipped. Otherwise eap-1 would match every replica-1 host.

NODE_SEARCH.md documents both rules, with new rows in its examples table.

🤖 Generated with Claude Code