Add integration and unit tests for monitoring JSON API endpoints

[gimballock]

Closes #329

Exercise every JSON REST API endpoint against live SV2 topologies and strengthen unit-test coverage for edge cases.

Integration tests (monitoring_integration.rs):

• 15 new tests covering /, /api/v1/global, /api/v1/server, /api/v1/server/channels, /api/v1/clients, /api/v1/clients/{id}, /api/v1/clients/{id}/channels, /api/v1/sv1/clients, and /api/v1/sv1/clients/{id} for both Pool and tProxy topologies
• Topology setup helpers (PoolWithSv2Miner, TproxyWithSv1Miner) eliminate duplicated boilerplate across tests
• 404 endpoints assert both HTTP status code and JSON error body via new assert_api_not_found helper

Assertion helpers (prometheus_metrics_assertions.rs):

• fetch_api_json: parse JSON API response
• fetch_api_with_status: return (status_code, json) without panicking on non-2xx, enabling 404 testing
• poll_until_api_field_gte: poll JSON endpoint until a field reaches a threshold (analogous to poll_until_metric_gte for Prometheus)
• assert_api_root, assert_api_global: reusable structure checks
• assert_api_not_found: verify HTTP 404 + error field
• POLL_TIMEOUT constant to reduce repetition

HTTP helper (utils.rs):

• make_get_request_with_status: returns (status, body) without panicking on 4xx responses, unlike make_get_request

Unit tests (http_server.rs):

• 11 new tests for pagination boundaries (limit=0, offset beyond total, limit exceeding MAX_LIMIT), missing data sources returning 404, invalid/non-existent client IDs, SV1 data in global endpoint, and Prometheus metrics with no sources

Dependencies:

• Add serde_json to integration-tests for JSON parsing

[Shourya742]

Shouldn't monitoring module have abstractions for most of these?

[gimballock]

I think the test failure here is the target of PR #338, once we rebase off of that this failure should go away i think.

[Shourya742]

I still dont see any changes requested. Now, we have some random files in the PR

[gimballock]

I still dont see any changes requested. Now, we have some random files in the PR

My bad, all those extra files are removed now.

[gimballock]

Rebased onto main and addressed all review feedback. Summary of changes in the current commit:

All review feedback addressed:

• Typed response structs — GlobalResponse, ServerResponse, ServerChannelsResponse, PaginatedResponse<T>, ClientMetadata, ClientChannelsResponse, Sv1Client defined in prometheus_metrics_assertions.rs; all integration tests use fetch_api_typed instead of stringly-typed serde_json::Value field access
• Route constants — mod routes block in http_server.rs test module defines all route strings; used throughout all unit tests
• Shutdown calls — all integration tests call shutdown_all! or .shutdown().await at the end
• No topology abstractions — each test sets up its own topology explicitly
• No assert helper wrappers — assertions inlined directly in each test
• No stray files — only the 6 files that belong to this PR are changed
• Rebased — single clean commit on top of current main (includes the shares_acknowledged rename from update/fix ServerStandardChannelInfo share accounting #425 and the shutdown_all! macro from Remove broadcast channel #317)

Additional cleanup done during rebase:

• Extracted a single generic poll_until<T, F> internal helper to eliminate duplicated poll loop boilerplate across the 5 typed polling functions
• assert_api_health now uses typed deserialization (fetch_api_typed) instead of string matching

@Shourya742 if you could take a look when you have a chance

[xyephy]

tACK
LGTM.

[xyephy]

Note:when running the monitoring integration tests sequentially (--test-threads=1), 9 tProxy tests fail with Connection refused on the monitoring endpoint. The first tProxy test passes but subsequent ones can't reach the monitoring server — looks like shutdown_all! returns before the port is fully released. Verified the same 10/9 pattern on main (3773dd5), so this is pre-existing and unrelated to this PR.

[gimballock]

Thanks for investigating. I believe this is pre-existing and unrelated to this PR — opened #430 to track it.

Looking at the code, TPROXY_MODE and VARDIFF_ENABLED are process-wide static OnceLock<_> statics in the translator. With test-threads = 1 all tests share one process, so a second TranslatorSv2::start() call would panic inside tokio::spawn (silently swallowed), meaning the translator never starts and the test sees "Connection refused". That would explain the 10/9 pattern, and the reviewer confirmed the same pattern on 3773dd5 before any of our changes.

This PR's tests use isolated ports via get_available_address() and private Registry::new(), so I don't think we're introducing any new conflicts — but happy to dig further if you see something I'm missing.

[GitGab19]

I would say we need to proceed with #338 before evaluating this PR, since #338 is touching things in the ITF as well.

[gimballock]

I rebased and tried to account for all the changes since last update now that #338 is merged

[xyephy]

tACK d71e1a3

Re-tested after #338 rebase