[onewland]

System Queries made you pick a storage before a host. The storage list repeats each ClickHouse cluster under many storage names, so finding the right host meant scrolling a long list of near-duplicates. The host field is now one autocomplete across every node the admin knows about, whichever storage it comes from.

Local screenshots:

Old-storage based selector remains:

---

The backend is unchanged. It still needs a storage, because the storage is how it finds the cluster's credentials, database and TLS settings, and how it checks that the host is a real node. So picking a node quietly sends a storage that lists it:

1. the storage from a shared link or the last pick, if it lists the node; otherwise
2. querylog, which has no 25s timeout; otherwise
3. the shortest name.

Manual host entry keeps a storage picker for the same reason. Share links and history entries still carry the storage, so existing links keep working.

The search is built around how our hostnames look. The full rules are in snuba/admin/static/utils/NODE_SEARCH.md. In short:

• Separators are optional: eap1 finds …-platform-1-0.
• A run of words can be typed as initials: eap finds events-analytics-platform.
• A typed separator can skip the rest of a word: err-1 finds errors-1-*.
• Results rank matches at the start of a word first, then matches whose word breaks line up best with the search (eap-11 prefers …-11-1 over …-1-1). Distributed-only nodes go last. Matched characters are highlighted.

I considered a library such as fzf-for-js, match-sorter or Fuse.js. These score fuzzy subsequences, so with about 200 similar hostnames, a search like err-1 would also match hosts such as generic-metrics-1-0, ranked lower. Rules this specific to hostnames are simpler to keep exact in about 250 lines of our own than to tune into someone else's scorer.

A second, separate commit makes every admin CustomSelect select its contents on focus, so typing replaces the current value instead of having to delete it first. It ignores the mouseup from the click that focused the field, because Chrome and Safari would otherwise use it to clear the selection again.

Worth a careful look in review:

• Storage choice: storageForNode in clickhouse_node_utils.tsx. It decides which cluster's credentials a query uses.
• Matcher: matchVariantFrom in the same file, the backtracking part of the search.

On 576 hosts the search takes about 1–12ms per keystroke. /clickhouse_nodes still looks up cluster topology once per storage, as it did before.

🤖 Generated with Claude Code

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit b1d7c6d. Configure here.}

[cursor]

By-storage host not restored

Medium Severity

Picking a storage in by-storage mode fills the first host in query state only, and never writes select-host. Reload always restores that key through queryForNode, so an auto-filled host is lost, and an older search-mode host can overwrite the current storage.

Additional Locations (1)

• snuba/admin/static/clickhouse_queries/query_display.tsx#L170-L179

 

^{Reviewed by Cursor Bugbot for commit b1d7c6d. Configure here.}

[cursor]

Manual-to-search drops storage

Medium Severity

setHostEntry only toggles flags. After a host is typed in Manual, switching back to Search keeps host and port but never runs queryForNode, so storage can stay unset while the storage picker is hidden and Execute stays disabled.

Additional Locations (1)

• snuba/admin/static/clickhouse_queries/query_display.tsx#L291-L298

 

^{Reviewed by Cursor Bugbot for commit b1d7c6d. Configure here.}