feat(collector): support milvus collector (#3877)

[paultanay]

What's changed?

Reworked the Milvus monitor to use HertzBeat's native protocol: http + parseType: prometheus - no custom module, no new Java classes.

• hertzbeat-manager/src/main/resources/define/app-milvus.yml - monitoring template scraping Milvus's Prometheus endpoint (port 9091, /metrics), covering 12 metric groups: proxy request count, slow query count, search/insert vector counts, search and mutation latency histograms (histogram bucket series), QueryNode collection/entity counts, and RootCoord collection/partition/DML/DDL counts
• home/docs/help/milvus.md - English help documentation, metric tables aligned exactly to the template

Fixes #3877

[Duansg]

Thank you for your work! Before going further, could we reconsider the approach?

Milvus exposes a standard Prometheus /metrics endpoint, and HertzBeat already handles that with protocol: http + parseType: prometheus. Going that route would let app-milvus.yml work without a new hertzbeat-collector-milvus module, a new MilvusProtocol, or changes to Metrics.java / DispatchConstants.

[paultanay]

Thanks for the clear guidance @Duansg! reworked the implementation to use protocol: http with parseType: prometheus. Please take another look!

[Aias00]

Review: feat(collector): support milvus collector (#3877)

Verdict: ✅ APPROVED — useful new monitor template; please address two doc/metric notes before or after merge.

What this PR does

Adds a Milvus vector-DB monitor template (app-milvus.yml) scraping the Prometheus /metrics endpoint on port 9091, plus the matching help doc. Covers proxy request/slow-query/vector counts, search & mutation latency, QueryNode loaded collections/entities, and RootCoord collection/partition/DML/DDL counts across 12 priority groups.

Assessment

• Structure: Valid template — ASF license header, trilingual name/help, sensible params (host/port/timeout), and consistent protocol: http + parseType: prometheus per group. Port default 9091 and path /metrics match Milvus defaults.
• Risk: Low — additive, no code path touched.

Notes (non-blocking)

1. Doc/metric mismatch: home/docs/help/milvus.md lists metrics that the template does not define (e.g. milvus_proxy_sq_latency_sum, milvus_proxy_receive_bytes_count, milvus_proxy_send_bytes_count, milvus_querynode_search_req_latency_sum, milvus_querynode_search_group_nq_sum, milvus_querynode_evicted_memory_size). The "Collected Metrics" tables read like Milvus's full metric surface rather than what this template actually scrapes. Please align the doc tables with the 12 metric names present in app-milvus.yml so users aren't misled.
2. Histogram metric names: milvus_proxy_sq_latency and milvus_proxy_mutation_latency are modeled with an le label (classic histogram-bucket shape). Prometheus histograms expose ..._bucket / ..._sum / ..._count series; a bare milvus_proxy_sq_latency series usually does not exist, so these two groups may silently collect nothing. Please verify Milvus emits a bare series with the le label, or retarget the metric name to milvus_proxy_sq_latency_bucket (and add _sum/_count as needed).

Nice addition overall.

[paultanay]

addressed remaining suggestions
@Aias00 @Duansg could you please take a look! Thanks...