neve 0.1.0

A read-only cache for avalanchego's JSON-RPC frontend API, backed by a local block store.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
//! Prometheus metrics: the `GET /metrics` endpoint plus typed recording
//! helpers used across the ingest, backfill, upstream-fetch, and subscription
//! paths.
//!
//! Names follow Prometheus conventions with a `neve_` prefix; histograms are
//! classic (explicit buckets) rather than native.

#![expect(
    clippy::cast_precision_loss,
    reason = "prometheus uses f64 and we're well within f64 bounds for all u64's we're sending"
)]

use std::future::Future;
use std::pin::Pin;
use std::sync::Arc;
use std::task::{Context, Poll};
use std::time::{Instant, SystemTime, UNIX_EPOCH};

use anyhow::{Context as _, Result, anyhow};
use futures_util::FutureExt;
use http::{Method, StatusCode, header};
use jsonrpsee::core::middleware::{Batch, Notification, RpcServiceT};
use jsonrpsee::core::server::MethodResponse;
use jsonrpsee::server::{HttpBody, HttpRequest, HttpResponse};
use jsonrpsee::types::Request;
use metrics::{counter, describe_counter, describe_gauge, describe_histogram, gauge, histogram};
use metrics_exporter_prometheus::{Matcher, PrometheusBuilder, PrometheusHandle};
use tower::{Layer, Service};

use crate::rpc::SubKind;

// ---- metric names ---------------------------------------------------------

const BUILD_INFO: &str = "neve_build_info";
const PROCESS_START_TIME: &str = "neve_process_start_time_seconds";

const INGEST_HEAD_HEIGHT: &str = "neve_ingest_head_height";
const INGEST_CONTIGUOUS_HEIGHT: &str = "neve_ingest_contiguous_height";
const INGEST_BEHIND_BLOCKS: &str = "neve_ingest_behind_blocks";
const INGEST_BLOCKS_TOTAL: &str = "neve_ingest_blocks_total";
const INGEST_LAST_BLOCK_TIMESTAMP: &str = "neve_ingest_last_block_timestamp_seconds";

const RPC_REQUESTS_TOTAL: &str = "neve_rpc_requests_total";
const RPC_REQUEST_DURATION_SECONDS: &str = "neve_rpc_request_duration_seconds";
const RPC_OPEN_CONNECTIONS: &str = "neve_rpc_open_connections";
const RPC_MISDIRECTED_TOTAL: &str = "neve_rpc_misdirected_total";

const UPSTREAM_REQUESTS_TOTAL: &str = "neve_upstream_requests_total";
const UPSTREAM_REQUEST_DURATION_SECONDS: &str = "neve_upstream_request_duration_seconds";
const UPSTREAM_RETRY_AFTER_SECONDS: &str = "neve_upstream_retry_after_seconds";
const UPSTREAM_CONNECTED_SINCE: &str = "neve_upstream_connected_since_seconds";
const UPSTREAM_WS_RECONNECTS_TOTAL: &str = "neve_upstream_ws_reconnects_total";
const UPSTREAM_WS_IDLE_TIMEOUTS_TOTAL: &str = "neve_upstream_ws_idle_timeouts_total";

const SUB_OPEN: &str = "neve_sub_open";
const SUB_LAGGED_TOTAL: &str = "neve_sub_lagged_total";
const SUB_SENT_BYTES_TOTAL: &str = "neve_sub_sent_bytes_total";

// ---- bounded label values -------------------------------------------------

/// Which path persisted a block — the `source` label on `neve_ingest_blocks_total`.
#[derive(Debug)]
pub enum BlockSource {
    /// The live WebSocket ingester (`newHeads`/`newBlocks`).
    Live,
    /// The backfill worker closing a gap.
    Backfill,
}

impl BlockSource {
    const fn as_str(&self) -> &'static str {
        match self {
            Self::Live => "live",
            Self::Backfill => "backfill",
        }
    }
}

/// Outcome of an upstream HTTPS request — the `outcome` label on
/// `neve_upstream_requests_total`. A small fixed enum so the series count can't
/// blow up regardless of what the upstream returns.
#[derive(Debug)]
pub enum UpstreamOutcome {
    /// A usable response (HTTP 2xx with a non-null JSON-RPC `result`).
    Ok,
    /// HTTP 2xx but a null/absent `result` — the block isn't available upstream
    /// yet. Tracks how often live fetches outrun the RPC backend's propagation.
    Empty,
    /// HTTP 429 Too Many Requests.
    TooManyRequests,
    /// HTTP 503 Service Unavailable.
    ServiceUnavailable,
    /// Transport error, non-2xx HTTP status, or response decode failure.
    Error,
}

impl UpstreamOutcome {
    const fn as_str(&self) -> &'static str {
        match self {
            Self::Ok => "ok",
            Self::Empty => "empty",
            Self::TooManyRequests => "429",
            Self::ServiceUnavailable => "503",
            Self::Error => "error",
        }
    }
}

/// Classify an HTTP status into a request outcome. Only the throttle codes map
/// to distinct variants; any other non-success status is `Error`. `Ok`/`Empty`
/// can't be derived from status alone (they depend on the response body), so
/// the success path classifies those itself.
impl From<StatusCode> for UpstreamOutcome {
    fn from(status: StatusCode) -> Self {
        if status == StatusCode::TOO_MANY_REQUESTS {
            Self::TooManyRequests
        } else if status == StatusCode::SERVICE_UNAVAILABLE {
            Self::ServiceUnavailable
        } else {
            Self::Error
        }
    }
}

// ---- recording helpers ----------------------------------------------------

/// Publish the static process-metadata gauges once at startup: `build_info`
/// (constant 1, version + git commit carried in labels, joined onto other
/// series in `PromQL`) and the process start time as unix epoch seconds (let
/// Prometheus derive uptime via `time() - neve_process_start_time_seconds`
/// rather than counting it ourselves). The commit comes from `build.rs`
/// (`"unknown"` outside a git checkout).
pub fn process_metadata() {
    gauge!(
        BUILD_INFO,
        "version" => env!("CARGO_PKG_VERSION"),
        "commit" => env!("NEVE_GIT_COMMIT"),
    )
    .set(1.0);

    let start = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map_or(0.0, |d| d.as_secs_f64());
    gauge!(PROCESS_START_TIME).set(start);
}

/// Create the standard process-stats collector and register its help text once.
/// Exposes the conventional `process_*` series (no `neve_` prefix): CPU seconds,
/// resident/virtual memory, open/max fds, thread count, and the process start
/// time. Call [`metrics_process::Collector::collect`] on the returned handle
/// periodically (from the metrics-upkeep loop) to refresh them. Reads `/proc` on
/// Linux (the deploy target); macOS and other platforms report a subset.
///
/// Note the collector also emits `process_start_time_seconds`, a benign duplicate
/// of our unconditional `neve_process_start_time_seconds` (different name, no
/// collision); we keep ours since it's set at install without depending on the
/// periodic collect.
pub fn process_collector() -> metrics_process::Collector {
    let collector = metrics_process::Collector::default();
    collector.describe();
    collector
}

/// Publish the freshness gauges from one backfill-loop snapshot: the stored
/// tip, the contiguous frontier, and how far behind the upstream tip we are.
pub fn ingest_heights(head: u64, contiguous: u64, behind: u64) {
    gauge!(INGEST_HEAD_HEIGHT).set(head as f64);
    gauge!(INGEST_CONTIGUOUS_HEIGHT).set(contiguous as f64);
    gauge!(INGEST_BEHIND_BLOCKS).set(behind as f64);
}

/// Count one persisted block, tagged by which path stored it.
pub fn block_persisted(source: BlockSource) {
    counter!(INGEST_BLOCKS_TOTAL, "source" => source.as_str()).increment(1);
}

/// Record the block-header timestamp (unix epoch seconds) of the latest live
/// block. Alert on staleness directly via
/// `time() - neve_ingest_last_block_timestamp_seconds > N` — this catches a
/// frozen upstream tip, which `neve_ingest_behind_blocks` (0 whenever we match
/// the tip, stuck or not) cannot. Set from the live path only: backfill writes
/// older blocks whose timestamps would drag the gauge backward.
pub fn last_block_timestamp(secs: u64) {
    gauge!(INGEST_LAST_BLOCK_TIMESTAMP).set(secs as f64);
}

/// Record one served JSON-RPC method call: bump the per-method/status counter
/// and observe its wall-clock latency. `method` is pre-clamped to the registered
/// set (or `"other"`) by the middleware, so label cardinality stays bounded.
pub fn rpc_call(method: &'static str, status: &'static str, secs: f64) {
    counter!(RPC_REQUESTS_TOTAL, "method" => method, "status" => status).increment(1);
    histogram!(RPC_REQUEST_DURATION_SECONDS, "method" => method).record(secs);
}

/// Count one response the `NotFound421` layer rewrote 200→421: a client asked
/// for a block/hash this mirror's tail doesn't hold. A high rate means clients
/// are routinely missing our window and falling back to the full-node pool.
pub fn rpc_misdirected() {
    counter!(RPC_MISDIRECTED_TOTAL).increment(1);
}

/// Record one upstream HTTPS request attempt: its outcome and its wall-clock
/// latency (`secs`). `secs` is per-attempt — the request round-trip including
/// body download/decode, *excluding* any retry backoff sleep that follows.
/// Accepts anything convertible into an `UpstreamOutcome` (e.g. an HTTP
/// `StatusCode`) so call sites needn't spell out the conversion. Latency spans
/// modes that differ by orders of magnitude (sub-ms mirror to 100ms+ public),
/// so the histogram uses a wide geometric ladder ([`UPSTREAM_DURATION_BUCKETS`]).
pub fn upstream_request(outcome: impl Into<UpstreamOutcome>, secs: f64) {
    counter!(UPSTREAM_REQUESTS_TOTAL, "outcome" => outcome.into().as_str()).increment(1);
    histogram!(UPSTREAM_REQUEST_DURATION_SECONDS).record(secs);
}

/// Record a `Retry-After` value (seconds) the upstream asked us to wait.
pub fn upstream_retry_after(secs: u64) {
    histogram!(UPSTREAM_RETRY_AFTER_SECONDS).record(secs as f64);
}

/// Mark the upstream live subscription as (re)established now: set the
/// connected-since gauge to the current unix epoch seconds. Prometheus derives
/// the current session's age as `time() - neve_upstream_connected_since_seconds`;
/// each reset (paired with a `ws_reconnect` bump) marks a fresh session, so a
/// recently-reset gauge plus a climbing reconnect counter reveals flapping.
pub fn upstream_connected() {
    let now = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map_or(0.0, |d| d.as_secs_f64());
    gauge!(UPSTREAM_CONNECTED_SINCE).set(now);
}

/// Count one WebSocket reconnect (a session ended or failed and we looped).
pub fn ws_reconnect() {
    counter!(UPSTREAM_WS_RECONNECTS_TOTAL).increment(1);
}

/// Count one idle-watchdog timeout (no `newHeads` within the configured window).
pub fn ws_idle_timeout() {
    counter!(UPSTREAM_WS_IDLE_TIMEOUTS_TOTAL).increment(1);
}

/// RAII guard for one active subscription. Bumps `neve_sub_open` on creation
/// and drops it on `Drop`, so the gauge is balanced no matter how the
/// subscription loop exits (client disconnect, send error, or a `?` early
/// return). Carries the kind so call sites don't repeat the label.
#[derive(Debug)]
pub struct SubMetricsGuard {
    kind: SubKind,
}

impl SubMetricsGuard {
    /// Increment the active-subscription gauge (decremented on drop).
    pub fn new(kind: SubKind) -> Self {
        gauge!(SUB_OPEN, "kind" => kind.as_str()).increment(1.0);
        Self { kind }
    }

    /// A slow subscriber fell behind the broadcast ring; `n` blocks were skipped.
    pub fn lagged(&self, n: u64) {
        counter!(SUB_LAGGED_TOTAL, "kind" => self.kind.as_str()).increment(n);
    }

    /// Serialized wire bytes pushed to the subscriber after a successful send.
    pub fn sent_bytes(&self, bytes: u64) {
        counter!(SUB_SENT_BYTES_TOTAL, "kind" => self.kind.as_str()).increment(bytes);
    }
}

impl Drop for SubMetricsGuard {
    fn drop(&mut self) {
        gauge!(SUB_OPEN, "kind" => self.kind.as_str()).decrement(1.0);
    }
}

// ---- recorder setup -------------------------------------------------------

/// Bucket bounds (seconds) for `neve_upstream_retry_after_seconds`. `Retry-After`
/// values run seconds-to-minutes, so the buckets are coarse and span up to the
/// 10-minute neighborhood of the default `--max-wait`.
const RETRY_AFTER_BUCKETS: &[f64] = &[0.5, 1.0, 2.0, 5.0, 10.0, 30.0, 60.0, 120.0, 300.0, 600.0];

/// Bucket bounds (seconds) for `neve_upstream_request_duration_seconds`. A
/// geometric ladder (~1.4x, the 1-1.5-2-3-5-7 decade pattern) from a 10ms floor
/// to a 2s near-timeout ceiling. Geometric rather than linear because upstream
/// latency tracks deployment distance: a same-region backend, a cross-country
/// (west↔east, ~60-80ms RTT before TLS/routers) gateway, and a congested path
/// each shift the whole distribution along the ladder — constant *relative*
/// resolution keeps the same handful of buckets across the bulk wherever it
/// lands, instead of a linear ladder that assumes a fixed distance. Floored at
/// 10ms (sub-10ms is below realistic network latency and lumps into the first
/// bucket); topped at 2s with `+Inf` capturing the timeout tail. The series is
/// unlabeled (low cardinality), so the bucket count costs little.
const UPSTREAM_DURATION_BUCKETS: &[f64] = &[
    0.01, 0.015, 0.02, 0.03, 0.05, 0.07, 0.1, 0.15, 0.2, 0.3, 0.5, 0.7, 1.0, 1.5, 2.0,
];

/// Bucket bounds (seconds) for `neve_rpc_request_duration_seconds`. A geometric
/// (~2.5x, 1-2.5-5) ladder from a 25µs floor to a 0.5s ceiling rather than a
/// linear one: the same histogram has to span methods orders of magnitude apart
/// — `eth_chainId` is an in-memory constant (tens of µs) while a heavy
/// `eth_getLogs` or a cold-cache read reaches into the ms — so constant
/// *relative* resolution gives every method a few useful buckets wherever its
/// distribution sits. Span is the measured envelope, not a guess: served-request
/// service time benchmarks at p50 ~0.21ms / p99 ~0.8ms unloaded (c1), and even
/// deep saturation tops out around p99 ~25ms (8-core) / ~77ms (t4g.small) — so
/// the ceiling sits at 0.5s (~6x the worst p99, with `+Inf` catching pathological
/// stalls), and the 25µs floor resolves the fastest in-memory methods rather than
/// jamming them into one bottom bucket. One shared layout (the exporter buckets
/// by metric name, not by the `method` label) keeps the series
/// cross-method-aggregable; a constant-time method crossing into the ms buckets
/// is a saturation signal worth catching.
const RPC_DURATION_BUCKETS: &[f64] = &[
    0.000_025, 0.000_05, 0.000_1, 0.000_25, 0.000_5, 0.001, 0.0025, 0.005, 0.01, 0.025, 0.05, 0.1,
    0.25, 0.5,
];

/// Build the Prometheus recorder, install it as the global `metrics` recorder,
/// describe every series (help text + units), and return a handle for rendering
/// the `/metrics` payload. Histograms get explicit buckets here (classic, not
/// native).
pub fn install() -> Result<PrometheusHandle> {
    let recorder = PrometheusBuilder::new()
        .set_buckets_for_metric(
            Matcher::Full(UPSTREAM_RETRY_AFTER_SECONDS.to_owned()),
            RETRY_AFTER_BUCKETS,
        )
        .context("configuring retry-after histogram buckets")?
        .set_buckets_for_metric(
            Matcher::Full(RPC_REQUEST_DURATION_SECONDS.to_owned()),
            RPC_DURATION_BUCKETS,
        )
        .context("configuring rpc-duration histogram buckets")?
        .set_buckets_for_metric(
            Matcher::Full(UPSTREAM_REQUEST_DURATION_SECONDS.to_owned()),
            UPSTREAM_DURATION_BUCKETS,
        )
        .context("configuring upstream-duration histogram buckets")?
        .build_recorder();
    let handle = recorder.handle();
    metrics::set_global_recorder(recorder)
        .map_err(|e| anyhow!("install global metrics recorder: {e}"))?;
    describe_metrics();
    process_metadata();
    Ok(handle)
}

/// Help text + units for each series. Called once after the recorder is global.
fn describe_metrics() {
    describe_gauge!(
        BUILD_INFO,
        "Build metadata as a constant 1; version and short git commit carried in labels."
    );
    describe_gauge!(
        PROCESS_START_TIME,
        metrics::Unit::Seconds,
        "Process start time (unix epoch seconds). Uptime = time() - this."
    );
    describe_gauge!(
        INGEST_HEAD_HEIGHT,
        "Highest stored block height (the blockstore high-water mark)."
    );
    describe_gauge!(
        INGEST_CONTIGUOUS_HEIGHT,
        "Highest gap-free stored block height."
    );
    describe_gauge!(
        INGEST_BEHIND_BLOCKS,
        "Blocks between the contiguous frontier and the upstream tip (0 = caught up). Primary freshness alerting signal."
    );
    describe_counter!(
        INGEST_BLOCKS_TOTAL,
        "Blocks persisted. Label source={live|backfill}."
    );
    describe_gauge!(
        INGEST_LAST_BLOCK_TIMESTAMP,
        metrics::Unit::Seconds,
        "Block-header timestamp of the latest live block (unix epoch seconds). Staleness = time() - this."
    );
    describe_counter!(
        RPC_REQUESTS_TOTAL,
        "Served JSON-RPC method calls. Labels method (registered eth_* set, else \"other\") and status={ok|error}."
    );
    describe_histogram!(
        RPC_REQUEST_DURATION_SECONDS,
        metrics::Unit::Seconds,
        "Served JSON-RPC method-call latency. Label method (clamped to the registered set)."
    );
    describe_gauge!(
        RPC_OPEN_CONNECTIONS,
        "Open JSON-RPC transport connections currently being served."
    );
    describe_counter!(
        RPC_MISDIRECTED_TOTAL,
        "Responses rewritten 200->421: a requested block/hash is outside this mirror's stored tail."
    );
    describe_counter!(
        UPSTREAM_REQUESTS_TOTAL,
        "Upstream block fetches (backfill HTTPS + live on-socket getBlockByNumber). Label outcome={ok|empty|429|503|error}."
    );
    describe_histogram!(
        UPSTREAM_REQUEST_DURATION_SECONDS,
        metrics::Unit::Seconds,
        "Per-attempt upstream HTTPS request latency (round-trip incl. body decode, excl. retry backoff)."
    );
    describe_histogram!(
        UPSTREAM_RETRY_AFTER_SECONDS,
        metrics::Unit::Seconds,
        "Retry-After delays requested by the upstream on 429/503."
    );
    describe_gauge!(
        UPSTREAM_CONNECTED_SINCE,
        metrics::Unit::Seconds,
        "Unix epoch seconds of the last successful upstream live subscribe. Session age = time() - this."
    );
    describe_counter!(
        UPSTREAM_WS_RECONNECTS_TOTAL,
        "WebSocket reconnects to the upstream."
    );
    describe_counter!(
        UPSTREAM_WS_IDLE_TIMEOUTS_TOTAL,
        "Idle-watchdog timeouts that forced a WebSocket reconnect."
    );
    describe_gauge!(
        SUB_OPEN,
        "Active eth_subscribe subscriptions. Label kind={newHeads|newBlocks|oldBlocks}."
    );
    describe_counter!(
        SUB_LAGGED_TOTAL,
        "Blocks dropped for subscribers that fell behind the broadcast ring. Label kind={newHeads|newBlocks} (live kinds only)."
    );
    describe_counter!(
        SUB_SENT_BYTES_TOTAL,
        "Serialized bytes pushed to subscribers. Label kind={newHeads|newBlocks|oldBlocks}."
    );
}

// ---- `GET /metrics` tower layer -------------------------------------------

/// Tower layer that serves `GET /metrics` from a [`PrometheusHandle`], passing
/// every other request through. Sibling to `health::HealthLayer`.
#[derive(Clone, Debug)]
pub struct MetricsLayer {
    handle: PrometheusHandle,
}

impl MetricsLayer {
    pub const fn new(handle: PrometheusHandle) -> Self {
        Self { handle }
    }
}

impl<S> Layer<S> for MetricsLayer {
    type Service = MetricsService<S>;
    fn layer(&self, inner: S) -> Self::Service {
        MetricsService {
            inner,
            handle: self.handle.clone(),
        }
    }
}

#[derive(Clone, Debug)]
pub struct MetricsService<S> {
    inner: S,
    handle: PrometheusHandle,
}

// Short-circuits `GET /metrics` with the rendered Prometheus payload (text
// exposition content-type); every other request falls through to the inner
// service unchanged.
impl<S> Service<HttpRequest<HttpBody>> for MetricsService<S>
where
    S: Service<HttpRequest<HttpBody>, Response = HttpResponse<HttpBody>> + Clone + Send + 'static,
    S::Future: Send + 'static,
    S::Error: Send + 'static,
{
    type Response = HttpResponse<HttpBody>;
    type Error = S::Error;
    type Future = Pin<Box<dyn Future<Output = Result<Self::Response, Self::Error>> + Send>>;

    fn poll_ready(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
        self.inner.poll_ready(cx)
    }

    fn call(&mut self, req: HttpRequest<HttpBody>) -> Self::Future {
        if req.method() == Method::GET && req.uri().path() == "/metrics" {
            let body = self.handle.render();
            let resp = HttpResponse::builder()
                .status(StatusCode::OK)
                .header(
                    header::CONTENT_TYPE,
                    "text/plain; version=0.0.4; charset=utf-8",
                )
                .body(HttpBody::from(body))
                .expect("static metrics response is valid");
            return std::future::ready(Ok(resp)).boxed();
        }
        self.inner.call(req).boxed()
    }
}

// ---- JSON-RPC method-level middleware -------------------------------------

/// RAII counter for one open JSON-RPC connection. Held behind an `Arc` in the
/// per-connection [`RpcMetricsService`] so the per-call clones share it and the
/// gauge decrements exactly once, when the last clone drops at connection close.
#[derive(Debug)]
struct ConnGuard;

impl ConnGuard {
    fn new() -> Self {
        gauge!(RPC_OPEN_CONNECTIONS).increment(1.0);
        Self
    }
}

impl Drop for ConnGuard {
    fn drop(&mut self) {
        gauge!(RPC_OPEN_CONNECTIONS).decrement(1.0);
    }
}

/// `RpcServiceT` middleware that records per-method call counts, latency, and
/// open-connection count. Built per connection (via `RpcServiceBuilder::layer_fn`
/// in `rpc::serve`), so `ConnGuard` tracks connections and `methods` carries the
/// registered method names used to clamp the `method` label.
#[derive(Clone)]
pub struct RpcMetricsService<S> {
    inner: S,
    /// Registered method names (`&'static str` from `RpcModule::method_names`),
    /// shared across connections. Used to clamp the `method` label so an unknown
    /// method can't blow up label cardinality.
    methods: Arc<[&'static str]>,
    /// Decrements `neve_rpc_open_connections` when the connection closes.
    _conn: Arc<ConnGuard>,
}

impl<S> RpcMetricsService<S> {
    /// Wrap `inner` for one connection, bumping the open-connection gauge.
    pub fn new(inner: S, methods: Arc<[&'static str]>) -> Self {
        Self {
            inner,
            methods,
            _conn: Arc::new(ConnGuard::new()),
        }
    }

    /// Clamp a wire method name to the registered set, or `"other"`, keeping the
    /// `method` label bounded. Linear scan over the dozen registered names.
    fn label(&self, method: &str) -> &'static str {
        self.methods
            .iter()
            .copied()
            .find(|&m| m == method)
            .unwrap_or("other")
    }
}

impl<S> RpcServiceT for RpcMetricsService<S>
where
    S: RpcServiceT<MethodResponse = MethodResponse> + Send + Sync + Clone + 'static,
{
    type MethodResponse = S::MethodResponse;
    type NotificationResponse = S::NotificationResponse;
    type BatchResponse = S::BatchResponse;

    fn call<'a>(&self, req: Request<'a>) -> impl Future<Output = Self::MethodResponse> + Send + 'a {
        let method = self.label(req.method.as_ref());
        let inner = self.inner.clone();
        async move {
            let start = Instant::now();
            let rp = inner.call(req).await;
            let status = if rp.is_success() { "ok" } else { "error" };
            rpc_call(method, status, start.elapsed().as_secs_f64());
            rp
        }
    }

    fn batch<'a>(&self, batch: Batch<'a>) -> impl Future<Output = Self::BatchResponse> + Send + 'a {
        // Batches are forwarded untimed: jsonrpsee re-invokes `call` for each
        // inner request, so the per-method counters/latency still capture them.
        self.inner.batch(batch)
    }

    fn notification<'a>(
        &self,
        n: Notification<'a>,
    ) -> impl Future<Output = Self::NotificationResponse> + Send + 'a {
        self.inner.notification(n)
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    /// Exercise the typed helpers against a thread-local recorder (no global
    /// install, so this is parallel-safe), then assert the rendered exposition
    /// carries the expected names, labels, and bucket lines — locking in the
    /// metric contract dashboards and the spec depend on.
    ///
    /// Not exhaustive by construction: Rust can't enumerate the helpers, so a
    /// new metric isn't covered until it's recorded and asserted here by hand.
    /// Update this test when adding a helper.
    #[test]
    #[expect(
        clippy::too_many_lines,
        reason = "contract test deliberately exercises every helper and asserts every rendered series in one place; splitting would scatter the contract"
    )]
    fn helpers_render_expected_series() {
        let recorder = PrometheusBuilder::new()
            .set_buckets_for_metric(
                Matcher::Full(UPSTREAM_RETRY_AFTER_SECONDS.to_owned()),
                RETRY_AFTER_BUCKETS,
            )
            .expect("bucket config")
            .set_buckets_for_metric(
                Matcher::Full(RPC_REQUEST_DURATION_SECONDS.to_owned()),
                RPC_DURATION_BUCKETS,
            )
            .expect("bucket config")
            .set_buckets_for_metric(
                Matcher::Full(UPSTREAM_REQUEST_DURATION_SECONDS.to_owned()),
                UPSTREAM_DURATION_BUCKETS,
            )
            .expect("bucket config")
            .build_recorder();
        let handle = recorder.handle();

        metrics::with_local_recorder(&recorder, || {
            process_metadata();
            rpc_call("eth_chainId", "ok", 0.001);
            rpc_misdirected();
            ingest_heights(100, 90, 10);
            block_persisted(BlockSource::Live);
            block_persisted(BlockSource::Backfill);
            last_block_timestamp(1_780_000_000);
            upstream_request(UpstreamOutcome::Ok, 0.012);
            upstream_request(UpstreamOutcome::Empty, 0.012);
            upstream_request(UpstreamOutcome::TooManyRequests, 0.012);
            upstream_retry_after(7);
            upstream_connected();
            ws_reconnect();
            ws_idle_timeout();
            let guard = SubMetricsGuard::new(SubKind::NewHeads);
            guard.sent_bytes(512);
            guard.lagged(3);
            // guard dropped here: neve_sub_open returns to 0.
        });

        let out = handle.render();

        // Process metadata: build_info is a constant 1 with version/commit
        // labels; start time is a positive epoch-seconds gauge.
        assert!(
            out.contains(&format!(
                r#"neve_build_info{{version="{}",commit="{}"}} 1"#,
                env!("CARGO_PKG_VERSION"),
                env!("NEVE_GIT_COMMIT"),
            )),
            "{out}"
        );
        assert!(out.contains("neve_process_start_time_seconds "), "{out}");
        // Served-RPC metrics: per-method/status counter, latency histogram, and
        // the 200->421 misdirected counter.
        assert!(
            out.contains(r#"neve_rpc_requests_total{method="eth_chainId",status="ok"} 1"#),
            "{out}"
        );
        assert!(
            out.contains(r#"neve_rpc_request_duration_seconds_bucket{method="eth_chainId""#),
            "{out}"
        );
        assert!(out.contains("neve_rpc_misdirected_total 1"), "{out}");
        // Gauges.
        assert!(out.contains("neve_ingest_head_height 100"), "{out}");
        assert!(out.contains("neve_ingest_behind_blocks 10"), "{out}");
        // Counters with bounded labels.
        assert!(
            out.contains(r#"neve_ingest_blocks_total{source="live"} 1"#),
            "{out}"
        );
        assert!(
            out.contains(r#"neve_ingest_blocks_total{source="backfill"} 1"#),
            "{out}"
        );
        assert!(
            out.contains("neve_ingest_last_block_timestamp_seconds 1780000000"),
            "{out}"
        );
        assert!(
            out.contains(r#"neve_upstream_requests_total{outcome="empty"} 1"#),
            "{out}"
        );
        assert!(
            out.contains(r#"neve_upstream_requests_total{outcome="429"} 1"#),
            "{out}"
        );
        assert!(
            out.contains("neve_upstream_request_duration_seconds_bucket"),
            "{out}"
        );
        assert!(
            out.contains("neve_upstream_request_duration_seconds_count 3"),
            "{out}"
        );
        assert!(
            out.contains("neve_upstream_connected_since_seconds "),
            "{out}"
        );
        assert!(out.contains("neve_upstream_ws_reconnects_total 1"), "{out}");
        assert!(
            out.contains("neve_upstream_ws_idle_timeouts_total 1"),
            "{out}"
        );
        // Histogram with explicit buckets.
        assert!(
            out.contains("neve_upstream_retry_after_seconds_bucket"),
            "{out}"
        );
        assert!(
            out.contains("neve_upstream_retry_after_seconds_count 1"),
            "{out}"
        );
        // Subscription series (open balanced back to 0 after the guard dropped).
        assert!(out.contains(r#"neve_sub_open{kind="newHeads"} 0"#), "{out}");
        assert!(
            out.contains(r#"neve_sub_sent_bytes_total{kind="newHeads"} 512"#),
            "{out}"
        );
        assert!(
            out.contains(r#"neve_sub_lagged_total{kind="newHeads"} 3"#),
            "{out}"
        );
    }
}