|
| 1 | +// Copyright (c) 2024-present, arana-db Community. All rights reserved. |
| 2 | +// |
| 3 | +// Licensed to the Apache Software Foundation (ASF) under one or more |
| 4 | +// contributor license agreements. See the NOTICE file distributed with |
| 5 | +// this work for additional information regarding copyright ownership. |
| 6 | +// The ASF licenses this file to You under the Apache License, Version 2.0 |
| 7 | +// (the "License"); you may not use this file except in compliance with |
| 8 | +// the License. You may obtain a copy of the License at |
| 9 | +// |
| 10 | +// http://www.apache.org/licenses/LICENSE-2.0 |
| 11 | +// |
| 12 | +// Unless required by applicable law or agreed to in writing, software |
| 13 | +// distributed under the License is distributed on an "AS IS" BASIS, |
| 14 | +// WITHOUT WARRANTIES OR CONDITIONS OF KIND, either express or implied. |
| 15 | +// See the License for the specific language governing permissions and |
| 16 | +// limitations under the License. |
| 17 | + |
| 18 | +//! Request-local storage-layer instrumentation. |
| 19 | +//! |
| 20 | +//! These types were originally declared in `runtime::message` (see |
| 21 | +//! <https://github.com/arana-db/kiwi/issues/312>). They live in the `client` |
| 22 | +//! crate so that `storage`, `runtime` and `cmd` can share them without |
| 23 | +//! creating a dependency cycle (`client` is a leaf crate that everything else |
| 24 | +//! already depends on). |
| 25 | +//! |
| 26 | +//! The real collector ([`RealStorageStatsCollector`]) is installed once per |
| 27 | +//! storage request via the [`STORAGE_STATS_COLLECTOR`] task-local and read by |
| 28 | +//! the storage engine at the point where reads, writes and deletes actually |
| 29 | +//! happen, so the byte/key counters reflect true I/O rather than values |
| 30 | +//! inferred from Redis command arguments. |
| 31 | +
|
| 32 | +use std::sync::atomic::{AtomicBool, AtomicU64, Ordering}; |
| 33 | +use std::sync::{Arc, Mutex as StdMutex}; |
| 34 | + |
| 35 | +use serde::{Deserialize, Serialize}; |
| 36 | + |
| 37 | +/// Statistics about storage operations for monitoring. |
| 38 | +#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)] |
| 39 | +pub struct StorageStats { |
| 40 | + /// Number of keys read during the operation |
| 41 | + pub keys_read: u64, |
| 42 | + /// Number of keys written during the operation |
| 43 | + pub keys_written: u64, |
| 44 | + /// Number of keys deleted during the operation |
| 45 | + pub keys_deleted: u64, |
| 46 | + /// Size of data read in bytes (key bytes + value bytes) |
| 47 | + pub bytes_read: u64, |
| 48 | + /// Size of data written in bytes (key bytes + value bytes) |
| 49 | + pub bytes_written: u64, |
| 50 | + /// Size of data removed by delete operations in bytes |
| 51 | + pub bytes_deleted: u64, |
| 52 | + /// Whether the operation hit a cache (e.g. block/moka cache) |
| 53 | + pub cache_hit: bool, |
| 54 | + /// RocksDB compaction level accessed |
| 55 | + pub compaction_level: Option<u32>, |
| 56 | +} |
| 57 | + |
| 58 | +/// Request-local collector for storage-layer instrumentation. |
| 59 | +/// |
| 60 | +/// Implementations must record the *actual* byte sizes measured by the storage |
| 61 | +/// layer, not sizes inferred from Redis command arguments. |
| 62 | +pub trait StorageStatsCollector: Send + Sync { |
| 63 | + /// Record a storage read. `key_bytes` and `value_bytes` are measured by the |
| 64 | + /// storage layer. |
| 65 | + fn record_read(&self, key_bytes: u64, value_bytes: u64); |
| 66 | + |
| 67 | + /// Record a storage write. `key_bytes` and `value_bytes` are measured after |
| 68 | + /// the storage layer accepts the mutation. |
| 69 | + fn record_write(&self, key_bytes: u64, value_bytes: u64); |
| 70 | + |
| 71 | + /// Record a storage delete. `key_bytes` is measured by the storage layer. |
| 72 | + fn record_delete(&self, key_bytes: u64); |
| 73 | + |
| 74 | + /// Return the accumulated statistics for the request. |
| 75 | + fn finish(&self) -> StorageStats; |
| 76 | +} |
| 77 | + |
| 78 | +/// Placeholder collector used until storage-layer instrumentation is wired in. |
| 79 | +#[derive(Debug, Default, Clone, Copy)] |
| 80 | +pub struct NoopStorageStatsCollector; |
| 81 | + |
| 82 | +impl StorageStatsCollector for NoopStorageStatsCollector { |
| 83 | + fn record_read(&self, _key_bytes: u64, _value_bytes: u64) {} |
| 84 | + |
| 85 | + fn record_write(&self, _key_bytes: u64, _value_bytes: u64) {} |
| 86 | + |
| 87 | + fn record_delete(&self, _key_bytes: u64) {} |
| 88 | + |
| 89 | + fn finish(&self) -> StorageStats { |
| 90 | + StorageStats::default() |
| 91 | + } |
| 92 | +} |
| 93 | + |
| 94 | +/// A real per-request collector backed by lock-free atomic counters. |
| 95 | +/// |
| 96 | +/// Safe to share across the (single) task that executes one storage request: |
| 97 | +/// every field is an atomic and `finish` snapshots the accumulated values. |
| 98 | +pub struct RealStorageStatsCollector { |
| 99 | + keys_read: AtomicU64, |
| 100 | + keys_written: AtomicU64, |
| 101 | + keys_deleted: AtomicU64, |
| 102 | + bytes_read: AtomicU64, |
| 103 | + bytes_written: AtomicU64, |
| 104 | + bytes_deleted: AtomicU64, |
| 105 | + cache_hit: AtomicBool, |
| 106 | + compaction_level: StdMutex<Option<u32>>, |
| 107 | +} |
| 108 | + |
| 109 | +impl Default for RealStorageStatsCollector { |
| 110 | + fn default() -> Self { |
| 111 | + Self::new() |
| 112 | + } |
| 113 | +} |
| 114 | + |
| 115 | +impl RealStorageStatsCollector { |
| 116 | + /// Create an empty collector ready to accumulate one request's stats. |
| 117 | + pub fn new() -> Self { |
| 118 | + Self { |
| 119 | + keys_read: AtomicU64::new(0), |
| 120 | + keys_written: AtomicU64::new(0), |
| 121 | + keys_deleted: AtomicU64::new(0), |
| 122 | + bytes_read: AtomicU64::new(0), |
| 123 | + bytes_written: AtomicU64::new(0), |
| 124 | + bytes_deleted: AtomicU64::new(0), |
| 125 | + cache_hit: AtomicBool::new(false), |
| 126 | + compaction_level: StdMutex::new(None), |
| 127 | + } |
| 128 | + } |
| 129 | + |
| 130 | + /// Mark that this request hit a cache (e.g. block/moka cache). |
| 131 | + pub fn mark_cache_hit(&self) { |
| 132 | + self.cache_hit.store(true, Ordering::Relaxed); |
| 133 | + } |
| 134 | + |
| 135 | + /// Record the RocksDB compaction level that served this request. |
| 136 | + pub fn set_compaction_level(&self, level: u32) { |
| 137 | + *self.compaction_level.lock().expect("compaction_level poisoned") = Some(level); |
| 138 | + } |
| 139 | +} |
| 140 | + |
| 141 | +impl StorageStatsCollector for RealStorageStatsCollector { |
| 142 | + fn record_read(&self, key_bytes: u64, value_bytes: u64) { |
| 143 | + self.keys_read.fetch_add(1, Ordering::Relaxed); |
| 144 | + self.bytes_read |
| 145 | + .fetch_add(key_bytes.saturating_add(value_bytes), Ordering::Relaxed); |
| 146 | + } |
| 147 | + |
| 148 | + fn record_write(&self, key_bytes: u64, value_bytes: u64) { |
| 149 | + self.keys_written.fetch_add(1, Ordering::Relaxed); |
| 150 | + self.bytes_written |
| 151 | + .fetch_add(key_bytes.saturating_add(value_bytes), Ordering::Relaxed); |
| 152 | + } |
| 153 | + |
| 154 | + fn record_delete(&self, key_bytes: u64) { |
| 155 | + self.keys_deleted.fetch_add(1, Ordering::Relaxed); |
| 156 | + self.bytes_deleted.fetch_add(key_bytes, Ordering::Relaxed); |
| 157 | + } |
| 158 | + |
| 159 | + fn finish(&self) -> StorageStats { |
| 160 | + StorageStats { |
| 161 | + keys_read: self.keys_read.load(Ordering::Relaxed), |
| 162 | + keys_written: self.keys_written.load(Ordering::Relaxed), |
| 163 | + keys_deleted: self.keys_deleted.load(Ordering::Relaxed), |
| 164 | + bytes_read: self.bytes_read.load(Ordering::Relaxed), |
| 165 | + bytes_written: self.bytes_written.load(Ordering::Relaxed), |
| 166 | + bytes_deleted: self.bytes_deleted.load(Ordering::Relaxed), |
| 167 | + cache_hit: self.cache_hit.load(Ordering::Relaxed), |
| 168 | + compaction_level: *self.compaction_level.lock().expect("compaction_level poisoned"), |
| 169 | + } |
| 170 | + } |
| 171 | +} |
| 172 | + |
| 173 | +tokio::task_local! { |
| 174 | + /// Per-request storage stats collector, scoped for the duration of a single |
| 175 | + /// storage command execution. The storage layer reads it via |
| 176 | + /// [`try_collector`]. |
| 177 | + pub static STORAGE_STATS_COLLECTOR: Arc<dyn StorageStatsCollector + Send + Sync>; |
| 178 | +} |
| 179 | + |
| 180 | +/// Returns the active request-local collector, if one was installed for the |
| 181 | +/// current async task. |
| 182 | +/// |
| 183 | +/// Returns `None` when no collector is scoped (e.g. unit tests that call the |
| 184 | +/// storage engine directly, or background storage tasks), so callers can record |
| 185 | +/// unconditionally without special-casing those paths. |
| 186 | +pub fn try_collector() -> Option<Arc<dyn StorageStatsCollector + Send + Sync>> { |
| 187 | + STORAGE_STATS_COLLECTOR.try_with(|c| c.clone()).ok() |
| 188 | +} |
| 189 | + |
| 190 | +#[cfg(test)] |
| 191 | +mod tests { |
| 192 | + use super::*; |
| 193 | + |
| 194 | + #[test] |
| 195 | + fn real_collector_accumulates_reads_writes_deletes() { |
| 196 | + let c = RealStorageStatsCollector::new(); |
| 197 | + c.record_read(3, 5); |
| 198 | + c.record_write(3, 7); |
| 199 | + c.record_read(2, 4); |
| 200 | + c.record_delete(3); |
| 201 | + |
| 202 | + let stats = c.finish(); |
| 203 | + assert_eq!(stats.keys_read, 2); |
| 204 | + assert_eq!(stats.keys_written, 1); |
| 205 | + assert_eq!(stats.keys_deleted, 1); |
| 206 | + assert_eq!(stats.bytes_read, (3 + 5) + (2 + 4)); |
| 207 | + assert_eq!(stats.bytes_written, 3 + 7); |
| 208 | + assert_eq!(stats.bytes_deleted, 3); |
| 209 | + assert!(!stats.cache_hit); |
| 210 | + } |
| 211 | + |
| 212 | + #[test] |
| 213 | + fn noop_yields_empty_stats() { |
| 214 | + assert_eq!(NoopStorageStatsCollector.finish(), StorageStats::default()); |
| 215 | + } |
| 216 | + |
| 217 | + #[test] |
| 218 | + fn scoped_collector_is_visible_to_try_collector() { |
| 219 | + // Mirrors how `storage_server` installs the collector and how the |
| 220 | + // storage facade reads it via `try_collector` — without needing a real |
| 221 | + // RocksDB instance. |
| 222 | + let runtime = tokio::runtime::Builder::new_current_thread() |
| 223 | + .enable_all() |
| 224 | + .build() |
| 225 | + .unwrap(); |
| 226 | + runtime.block_on(async { |
| 227 | + let collector: Arc<dyn StorageStatsCollector + Send + Sync> = |
| 228 | + Arc::new(RealStorageStatsCollector::new()); |
| 229 | + STORAGE_STATS_COLLECTOR |
| 230 | + .scope(Arc::clone(&collector), async { |
| 231 | + if let Some(c) = try_collector() { |
| 232 | + c.record_write(3, 5); |
| 233 | + c.record_read(2, 4); |
| 234 | + } else { |
| 235 | + panic!("try_collector() must return the scoped collector"); |
| 236 | + } |
| 237 | + }) |
| 238 | + .await; |
| 239 | + let stats = collector.finish(); |
| 240 | + assert_eq!(stats.keys_written, 1); |
| 241 | + assert_eq!(stats.bytes_written, 8); |
| 242 | + assert_eq!(stats.keys_read, 1); |
| 243 | + assert_eq!(stats.bytes_read, 6); |
| 244 | + }); |
| 245 | + } |
| 246 | +} |
0 commit comments