fix(admin): use default HTTP port for by-host connections; don't crash tracing on empty trace

[phacops]

Two snuba-admin bugs that surface once the clickhouse-connect (HTTP) driver is enabled. The driver is chosen at runtime (use_clickhouse_connect_driver) inside the shared connection cache, so admin paths get the HTTP pool transparently — and these two assumptions, which were fine for the native driver, break.

1. By-host connections must use the node's default HTTP port, not the cluster's http_port

_build_validated_pool built the node with http_port=cluster.get_http_port(). But admin always targets a specific node by hostname (e.g. a replica discovered from system.clusters). The cluster's configured http_port belongs to the cluster's query endpoint — which may sit behind a proxy/load balancer — not to an arbitrary individual node.

• Native driver: talks to the node directly on its native port → unaffected.
• HTTP driver: has to reach the node's own ClickHouse HTTP listener, which is the well-known default port.

The node is now built with DEFAULT_CLICKHOUSE_HTTP_PORT. This mirrors the documented "by-host helper" pattern already used by the CLI helpers and the connection cache's own default.

2. The tracing tool 500'd with list index out of range on the HTTP driver

clickhouse-connect doesn't surface the server's send_logs_level output (a known, documented limitation of that path), so trace_output is always empty. summarize_trace_output indexed parsed[0] unconditionally, raising IndexError — which the trace view's catch-all turned into:

An error occurred: {"message":"list index out of range","type":"unknown"}

It now returns an empty TracingSummary when there are no parseable trace lines, so the tracing tool returns its (empty) trace instead of erroring.

Changes

• snuba/admin/clickhouse/common.py — _build_validated_pool builds the by-host node with the default ClickHouse HTTP port instead of cluster.get_http_port().
• snuba/admin/clickhouse/trace_log_parsing.py — summarize_trace_output returns an empty summary on empty/unparseable input.

Tests

• test_by_host_connection_uses_default_http_port (parametrized over all four by-host helpers) asserts the node passed to the connection cache carries the default HTTP port and not the cluster's configured port (verified with a sentinel port so the guard holds even when the test cluster uses the default).
• test_tracing_summary_empty_trace_output asserts summarize_trace_output("") (and whitespace) returns an empty summary rather than raising.

The trace-parsing change was verified locally; the existing test_tracing_summary still passes. ruff check and ruff format --check pass on all changed files.

Legal Boilerplate

Look, I get it. The entity doing business as "Sentry" was incorporated in the State of Delaware in 2015 as Functional Software, Inc. and is gonna need some rights from me in order to utilize my contributions in this here PR. So here's the deal: I retain all rights, title and interest in and to my contributions, and by keeping this boilerplate intact I confirm that Sentry can use, modify, copy, and redistribute my contributions, under Sentry's choice of terms.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Lzu1UmpvvA1S2Ypq2uqzEX

---

Generated by Claude Code

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit a3bfc0e. Configure here.}