Support higher performance histogram

[chenBright]

What problem does this PR solve?

Issue Number: resolve

Problem Summary:

bvar::Histogram records through detail::AgentCombiner, the same thread-local
aggregation every other reducer uses. Each operator<< has to look the agent up
in the thread-local AgentGroup and then take the per-agent butil::Lock that
ElementContainer wraps every read and write in. For a histogram that cost is
paid on a path that only ever increments one bucket, a sum and a count.

When bRPC is built with the babylon counter backend (with_babylon_counter),
Adder, IntRecorder and Percentile already record through babylon's
thread-local storage, but Histogram was left on the combiner.

What is changed and the side effects?

Changed:

Side effects:

• Performance effects:

write_perf, 500k records per thread, Linux x86_64, ns per record:

threads	babylon	AgentCombiner
1	6.91	23.40
2	6.98	23.37
4	6.97	24.32
8	7.62	24.32

• Breaking backward compatibility:

---

Check List:

• Please make sure your changes are compilable.
• When providing us with a new feature, it is best to add related tests.
• Please follow Contributor Covenant Code of Conduct.

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

Review effort: Lite
Findings: 5 · 4

Open (9)

• The PR description fields look incomplete/placeholder (e.g., 'Issue Number: resolve', empty… · New
• This PR removes PassiveStatus::reset() (which previously CHECK+abort'ed). That changes the public… · New
• elapsed_ns is stored in long, which can be 32-bit on some platforms and may overflow for… · New
• total_ns uses long and accumulates per-thread elapsed nanoseconds; on 32-bit long this can… · New
• This adds a performance measurement as a regular unit test, with a fairly large workload… · New
• Correct spelling in comment: 'Suming' should be 'Summing'. · New
• Correct wording in comment: 'Inversed' should be 'Inverse' (e.g., 'Inverse operation'). · New
• The comment refers to Percentile, but this is Histogram. Please update the wording to avoid… · New
• last_counts is hard-coded to 4 buckets, which implicitly assumes the schema {10, 20, 30}… · New

What changed in this PR

This PR introduces a higher-performance Histogram backend (guarded by WITH_BABYLON_COUNTER) using thread-local slots, and adds/updates unit tests to validate correctness under concurrency and thread lifecycle behavior.

Changes:

• Add a babylon-backed histogram storage (HistogramSlot / HistogramStorage) and route Histogram recording/snapshotting through it when enabled.
• Optimize bucket lookup by switching BucketSchema::index_of from binary search to a linear scan for small bucket counts.
• Add new concurrency-focused tests plus a write-performance measurement test; refactor ReducerSampler to use tag dispatch.

File	Description
test/​bvar_histogram_unittest.cpp	Adds tests for the new backend, contention/snapshot invariants, dead-thread behavior, and a write perf measurement.
src/​bvar/​passive_status.h	Adjusts virtual destructor and removes a reset() implementation that previously aborted.
src/​bvar/​histogram.h	Adds babylon-backed histogram storage types, switches bucket index lookup to linear scan, and updates combiner/storage plumbing.
src/​bvar/​histogram.cpp	Wires operator<< and get_value() to the selected backend.
src/​bvar/​detail/​sampler.h	Refactors sampling logic to avoid instantiating both branches (tag dispatch).
BUILD.bazel	Adds babylon thread-local dependency when with_babylon_counter is enabled.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

Review effort: Lite
Findings: 2 · 2

Open (4)

• This adds a heavy performance-style test (up to ~4M records plus threading) that will run… · New
• This adds a heavy performance-style test (up to ~4M records plus threading) that will run… · New
• For integral counters (_counts, _num), using fetch_add(..., memory_order_relaxed) would more… · New
• The reported metric is the average of per-thread wall-clock times divided by ops, which is closer… · New

Resolved since last review (9)

• This adds a performance measurement as a regular unit test, with a fairly large workload…
• total_ns uses long and accumulates per-thread elapsed nanoseconds; on 32-bit long this can…
• elapsed_ns is stored in long, which can be 32-bit on some platforms and may overflow for…
• This PR removes PassiveStatus::reset() (which previously CHECK+abort'ed). That changes the public…
• The PR description fields look incomplete/placeholder (e.g., 'Issue Number: resolve', empty…
• last_counts is hard-coded to 4 buckets, which implicitly assumes the schema {10, 20, 30}…
• The comment refers to Percentile, but this is Histogram. Please update the wording to avoid…
• Correct wording in comment: 'Inversed' should be 'Inverse' (e.g., 'Inverse operation').
• Correct spelling in comment: 'Suming' should be 'Summing'.

[wwbmmm]

LGTM