lian/ldb
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

落盘机制修改

Browse Source
This commit is contained in:
1iaan
2026-03-06 11:54:30 +00:00
parent c4e9bedd0a
commit 2e6baf0efe
14 changed files with 1302 additions and 443 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@@ -5,6 +5,7 @@
*.a
/ebpf/libbpf-bootstrap
/doc
/test-redis/results
kvstore

93
README.md
View File

@@ -1,20 +1,5 @@
# 9.1 Kvstore
## 需求
1. ntyco需要作为kvstore的submodule,通过git clone一次下载。 **完成**。
2. README需要包含编译步骤，测试方案与可行性，性能数据。 **完成**。
3. 全量持久化保存数据集。 **BUG FIX**。
4. 持久化的性能数据。 **完成**。
5. 特殊字符，可以解决redis的resp协议。 **完成**。
6. 实现配置文件，把日志级别，端口ip，主从模式，持久化方案。 **完成**。
7. 持久化落盘用io_uring，加载配置文件用mmap。 **完成**。
8. 主从同步的性能，开启与关闭性能做到5%?。
9. 主从同步600w条,出现的coredump。 **BUG FIX**。
10. 主从同步用ebpf实现。 **完成**。
11. 内存池测试qps与虚拟内存，物理内存。 **完成**。
12. 实现一个内存泄露检测组件。 **完成**。
## 环境安装与编译
```shell
# xml
@@ -101,9 +86,9 @@ BATCH (N=9000000) --> time_used=10033 ms, qps=1794079
VIRT 208M
RES 155M
```
![alt text](image11.png)
![alt text](image12.png)
![alt text](image13.png)
![alt text](https://disk.0voice.com/p/wl)
![alt text](https://disk.0voice.com/p/wm)
![alt text](https://disk.0voice.com/p/wO)
#### jemalloc
```shell
@@ -122,9 +107,9 @@ BATCH (N=9000000) --> time_used=9353 ms, qps=1924516
VIRT 356M
RES 119M
```
![alt text](image11.png)
![alt text](image22.png)
![alt text](image23.png)
![alt text](https://disk.0voice.com/p/wl)
![alt text](https://disk.0voice.com/p/wP)
![alt text](https://disk.0voice.com/p/wQ)
#### mypool
```shell
@@ -143,9 +128,9 @@ BATCH (N=3000000) --> time_used=3022 ms, qps=1985440
VIRT 122M
RES 71492
```
![alt text](image31.png)
![alt text](image32.png)
![alt text](image33.png)
![alt text](https://disk.0voice.com/p/wR)
![alt text](https://disk.0voice.com/p/wn)
![alt text](https://disk.0voice.com/p/wo)
### 测试4：主从同步
测试条件：
@@ -177,18 +162,56 @@ average qps:777838
ALL TESTS PASSED.
```
### 面试题
1. 为什么会实现kvstore，使用场景在哪里？
2. reactor, ntyco, io_uring的三种网络模型的性能差异？
3. 多线程的kvstore该如何改进？
4. 私有协议如何设计会更加安全可靠？
5. 协议改进以后，对已有的代码有哪些改变？
6. kv引擎实现了哪些？
7. 每个kv引擎的使用场景，以及性能差异？
8. 测试用例如何实现？并且保证代码覆盖率超过90%
9. 网络并发量如何？qps如何？
10. 能够跟哪些系统交互使用？
## REDIS 对比测试
### 数据口径（2026-03-06 09:00:30）
- 轮次：3 轮（取均值）
- 参数：requests=1000000 pipeline=128 keyspace=1000000 value-size=32
- 数据来源：`test-redis/results/hash_bench_summary_20260306_090030.csv`
### 仅看 mypool 与 Redis
| 系统 | 场景 | set 均值QPS | set 均值us/op | get 均值QPS | get 均值us/op |
|---|---|---:|---:|---:|---:|
| kvstore (mypool) | none | 165554.00 | 6.04 | 169509.33 | 5.91 |
| kvstore (mypool) | incremental(oplog_sync=none) | 168248.67 | 5.96 | 181076.67 | 5.53 |
| kvstore (mypool) | incremental(oplog_sync=every_sec) | 164801.67 | 6.08 | 178469.33 | 5.60 |
| redis | none | 221558.00 | 4.52 | 254807.33 | 3.92 |
| redis | aof_no | 177826.67 | 5.63 | 256099.00 | 3.91 |
| redis | aof_everysec | 179159.67 | 5.59 | 243906.00 | 4.11 |
| redis | aof_always | 66807.33 | 14.97 | 236824.67 | 4.23 |
### oplog 与 AOF 安全性级别
1. `oplog_sync=none` 时，kvstore 仅异步写入，不做周期 fsync，安全级别仍接近 Redis `aof_no`。
2. `oplog_sync=every_sec` 时，kvstore 每秒执行一次“flush + 每个 uring worker 提交 fsync(drain) 并等待回调”，可提供接近 Redis `aof_everysec` 的周期性落盘保证（仍可能丢失最近 1 秒窗口）。
3. 与 Redis 相比，当前 kvstore 仍缺少 AOF 尾部校验/截断等恢复增强机制，崩溃恢复鲁棒性上略弱于 Redis AOF 体系。
### 落盘性能退化（写路径，set）
1. kvstore(mypool)：`none 165554.00 -> incremental(oplog_sync=none) 168248.67`，变化 `+1.63%`（本轮无退化，属抖动范围）。
2. kvstore(mypool)：`none 165554.00 -> incremental(oplog_sync=every_sec) 164801.67`，退化 `0.45%`。
3. redis：`none 221558.00 -> aof_no 177826.67`，退化 `19.74%`。
4. redis：`none 221558.00 -> aof_everysec 179159.67`，退化 `19.14%`。
5. redis：`none 221558.00 -> aof_always 66807.33`，退化 `69.85%`。
6. 对应 set 平均时延(us/op)变化：kvstore every_sec `6.04 -> 6.08`（`+0.61%`），redis aof_no `4.52 -> 5.63`（`+24.72%`），redis aof_everysec `4.52 -> 5.59`（`+23.69%`），redis aof_always `4.52 -> 14.97`（`+231.51%`）。
结论：本轮下 kvstore(mypool) 的 `every_sec` 持久化性能代价远低于 Redis AOF 系列，同时安全性目标已从 `aof_no` 抬升到接近 `aof_everysec` 的级别。
## 调用开销
### gprof Flat Profile（Top 12，按 self time）
| 排名 | 函数 | self time % | self seconds | calls |
|---:|---|---:|---:|---:|
| 1 | `rbtree_node_get_key` | 58.34 | 0.56 | 103209091 |
| 2 | `rbtree_search` | 7.29 | 0.07 | 1874362 |
| 3 | `kvs_keycmp` | 5.73 | 0.06 | 74287566 |
| 4 | `ascii_casecmp` | 3.13 | 0.03 | 21490996 |
| 5 | `task_init` | 3.13 | 0.03 | 1397974 |
| 6 | `mp_page_create` | 3.13 | 0.03 | 23556 |
| 7 | `need` | 2.08 | 0.02 | 14042759 |
| 8 | `parse_i64` | 2.08 | 0.02 | 7029783 |
| 9 | `mp_page_alloc` | 2.08 | 0.02 | 5122487 |
| 10 | `rbtree_node_size` | 2.08 | 0.02 | 1860739 |
| 11 | `submit_write` | 2.08 | 0.02 | 1394599 |
| 12 | `rbtree_insert` | 2.08 | 0.02 | 926531 |
## 项目收获
#### reactor网络模型，用户态网络缓冲区的写法。

29
common/config.c
View File

@@ -56,6 +56,7 @@ static void set_default_config(AppConfig *cfg)
cfg->mode = MODE_MASTER;
cfg->master_port = 8888;
cfg->persistence = PERSIST_NONE;
cfg->oplog_sync_mode = OPLOG_SYNC_NONE;
cfg->allocator = ALLOC_JEMALLOC;
cfg->leak_mode = MEMLEAK_DETECT_OFF;
cfg->replica_mode = REPLICA_DISABLE;
@@ -87,6 +88,16 @@ static void parse_persistence(const char *s, PersistenceType *out)
else if (!strcasecmp(s, "none")) *out = PERSIST_NONE;
}
static void parse_oplog_sync_mode(const char *s, OplogSyncMode *out)
{
if (!s || !out) return;
if (!strcasecmp(s, "every_sec") || !strcasecmp(s, "everysec")) {
*out = OPLOG_SYNC_EVERY_SEC;
} else if (!strcasecmp(s, "none")) {
*out = OPLOG_SYNC_NONE;
}
}
static void parse_allocator(const char *s, AllocatorType *out)
{
if (!s || !out) return;
@@ -188,6 +199,15 @@ const char *persistence_to_string(PersistenceType p)
}
}
const char *oplog_sync_mode_to_string(OplogSyncMode m)
{
switch (m) {
case OPLOG_SYNC_NONE: return "none";
case OPLOG_SYNC_EVERY_SEC: return "every_sec";
default: return "unknown";
}
}
const char *allocator_to_string(AllocatorType a)
{
switch (a) {
@@ -337,6 +357,15 @@ void persist_load(xmlNodePtr *root, AppConfig *out_cfg){
}
}
xmlNodePtr oplog_sync_node = find_child(pers, "oplog_sync");
if (oplog_sync_node) {
xmlChar *txt = xmlNodeGetContent(oplog_sync_node);
if (txt) {
parse_oplog_sync_mode((char *)txt, &out_cfg->oplog_sync_mode);
xmlFree(txt);
}
}
xmlNodePtr array_node = find_child(pers, "array");
if (array_node) {
xmlChar *txt = xmlNodeGetContent(array_node);

7
common/config.h
View File

@@ -20,6 +20,11 @@ typedef enum {
PERSIST_NONE
} PersistenceType;
typedef enum {
OPLOG_SYNC_NONE,
OPLOG_SYNC_EVERY_SEC
} OplogSyncMode;
typedef enum {
REPLICA_DISABLE,
REPLICA_ENABLE
@@ -53,6 +58,7 @@ typedef struct {
char array_file[256];
char rbtree_file[256];
char hash_file[256];
OplogSyncMode oplog_sync_mode;
AllocatorType allocator;
MemLeakDetectMode leak_mode;
@@ -68,6 +74,7 @@ int config_load(const char *filename, AppConfig *out_cfg);
const char *log_level_to_string(LogLevel lvl);
const char *server_mode_to_string(ServerMode mode);
const char *persistence_to_string(PersistenceType p);
const char *oplog_sync_mode_to_string(OplogSyncMode m);
const char *allocator_to_string(AllocatorType a);
const char *leakage_to_string(MemLeakDetectMode a);
const char *replica_to_string(ReplicaMode a);

19
config/config.xml
View File

@@ -1,11 +1,11 @@
<?xml version='1.0' encoding='UTF-8'?>
<?xml version="1.0" encoding="UTF-8"?>
<config>
<server>
<ip>127.0.0.1</ip>
<ip>192.168.220.134</ip>
<port>8888</port>
<mode>master</mode>
<mode>master</mode> <!-- master / slave -->
<!-- 仅当 mode=slave 时使用 -->
<replica>disable</replica>
<master>
<ip>192.168.220.134</ip>
@@ -14,21 +14,22 @@
</server>
<log>
<level>INFO</level>
<level>INFO</level> <!-- DEBUG / INFO / ERROR -->
</log>
<persistence>
<type>incremental</type>
<dir>data/persist_mypool_20260305_072256</dir>
<type>none</type> <!-- incremental / none -->
<dir>data</dir> <!-- 所有持久化文件所在目录 -->
<wal>kvs_oplog.db</wal>
<oplog_sync>none</oplog_sync> <!-- none / every_sec -->
<array>kvs_array.db</array>
<rbtree>kvs_rbtree.db</rbtree>
<hash>kvs_hash.db</hash>
</persistence>
<memory>
<allocator>mypool</allocator>
<leakage>disable</leakage>
<allocator>mypool</allocator> <!-- jemalloc / malloc / mypool -->
<leakage>disable</leakage> <!-- enable/disable-->
</memory>
</config>

138
diskuring/diskuring.c
View File

@@ -284,7 +284,16 @@ static void *worker_main(void *arg) {
break;
}
if (t->op == TASK_WRITE) {
io_uring_prep_writev(sqe, t->fd, t->iovs, t->iovcnt, t->off);
} else if (t->op == TASK_FSYNC) {
io_uring_prep_fsync(sqe, t->fd, t->fsync_flags);
} else {
task_finish(t, -EINVAL);
destroy_queue_push(w->parent, t);
continue;
}
sqe->flags |= (unsigned char)t->sqe_flags;
sqe->user_data = (uint64_t)(uintptr_t)t;
prepared++;
}
@@ -327,8 +336,18 @@ void task_init(task_t *t) {
if (!t) {
return;
}
t->op = TASK_WRITE;
t->fd = -1;
t->off = 0;
t->fsync_flags = 0;
t->sqe_flags = 0;
t->done = 0;
t->res = 0;
t->iovs = NULL;
t->iovcnt = 0;
t->free_iov_bases = 1;
t->on_destroy = NULL;
t->on_destroy_arg = NULL;
t->next = NULL;
}
@@ -355,12 +374,18 @@ void task_destroy(task_t *t) {
return;
}
if (t->on_destroy) {
t->on_destroy(t, t->on_destroy_arg);
}
if (t->iovs) {
if (t->free_iov_bases) {
for (int i = 0; i < t->iovcnt; i++) {
if (t->iovs[i].iov_base) {
kvs_free(t->iovs[i].iov_base);
}
}
}
kvs_free(t->iovs);
}
@@ -678,6 +703,33 @@ static int queue_task_with_backpressure(iouring_ctx_t *ctx, task_t *t) {
return -1;
}
static int queue_task_to_worker_with_backpressure(iouring_ctx_t *ctx, task_t *t, int worker_id) {
iouring_worker_t *w;
if (!ctx || !ctx->workers || !t) {
return -1;
}
if (worker_id < 0 || worker_id >= ctx->worker_nr) {
return -1;
}
w = &ctx->workers[worker_id];
while (atomic_load_explicit(&ctx->stop, memory_order_acquire) == 0) {
int need_notify = 0;
if (spsc_try_push(&w->submit_q, t, &need_notify) == 0) {
if (need_notify) {
worker_notify(w);
}
return 0;
}
cleanup_finished_iouring_tasks(ctx);
sched_yield();
}
return -1;
}
task_t *submit_write(iouring_ctx_t *ctx, int fd, void **bufs, size_t *lens, int count, off_t off) {
task_t *t;
size_t total = 0;
@@ -782,6 +834,92 @@ task_t *submit_write(iouring_ctx_t *ctx, int fd, void **bufs, size_t *lens, int
return t;
}
task_t *submit_write_ref(iouring_ctx_t *ctx, int fd, void **bufs, size_t *lens, int count, off_t off,
int free_iov_bases, task_destroy_cb_t on_destroy, void *on_destroy_arg) {
task_t *t;
size_t total = 0;
if (!ctx || !ctx->workers || !bufs || !lens || count <= 0) {
return NULL;
}
t = (task_t *)kvs_malloc(sizeof(task_t));
if (!t) {
return NULL;
}
task_init(t);
t->op = TASK_WRITE;
t->fd = fd;
t->off = off;
t->iovcnt = count;
t->free_iov_bases = free_iov_bases ? 1 : 0;
t->on_destroy = on_destroy;
t->on_destroy_arg = on_destroy_arg;
t->iovs = (struct iovec *)kvs_malloc(sizeof(struct iovec) * (size_t)count);
if (!t->iovs) {
kvs_free(t);
return NULL;
}
for (int i = 0; i < count; ++i) {
if (!bufs[i] || lens[i] == 0) {
task_destroy(t);
return NULL;
}
if (lens[i] > SIZE_MAX - total) {
task_destroy(t);
return NULL;
}
t->iovs[i].iov_base = bufs[i];
t->iovs[i].iov_len = lens[i];
total += lens[i];
}
if (total == 0) {
task_destroy(t);
return NULL;
}
if (queue_task_with_backpressure(ctx, t) != 0) {
task_destroy(t);
return NULL;
}
return t;
}
task_t *submit_fsync_ref(iouring_ctx_t *ctx, int fd, int worker_id, int drain,
task_destroy_cb_t on_destroy, void *on_destroy_arg) {
task_t *t;
if (!ctx || !ctx->workers || fd < 0) {
return NULL;
}
if (worker_id < 0 || worker_id >= ctx->worker_nr) {
return NULL;
}
t = (task_t *)kvs_malloc(sizeof(task_t));
if (!t) {
return NULL;
}
task_init(t);
t->op = TASK_FSYNC;
t->fd = fd;
t->fsync_flags = 0;
t->sqe_flags = drain ? IOSQE_IO_DRAIN : 0;
t->on_destroy = on_destroy;
t->on_destroy_arg = on_destroy_arg;
if (queue_task_to_worker_with_backpressure(ctx, t, worker_id) != 0) {
task_destroy(t);
return NULL;
}
return t;
}
int uring_task_complete(iouring_ctx_t *ctx) {
if (!ctx || !ctx->workers) {
return 1;

14
diskuring/diskuring.h
View File

@@ -11,18 +11,26 @@
#include <sys/uio.h>
#include <unistd.h>
typedef enum { TASK_READ, TASK_WRITE } task_op_t;
typedef enum { TASK_READ, TASK_WRITE, TASK_FSYNC } task_op_t;
struct task;
typedef void (*task_destroy_cb_t)(struct task *t, void *arg);
typedef struct task {
task_op_t op;
int fd;
off_t off;
unsigned fsync_flags;
unsigned sqe_flags;
int res;
_Atomic int done;
struct iovec *iovs;
int iovcnt;
int free_iov_bases;
task_destroy_cb_t on_destroy;
void *on_destroy_arg;
struct task *next;
} task_t;
@@ -73,6 +81,10 @@ int iouring_init(iouring_ctx_t *ctx, unsigned entries);
void iouring_shutdown(iouring_ctx_t *ctx);
task_t *submit_write(iouring_ctx_t *ctx, int fd, void **bufs, size_t *lens, int count, off_t off);
task_t *submit_write_ref(iouring_ctx_t *ctx, int fd, void **bufs, size_t *lens, int count, off_t off,
int free_iov_bases, task_destroy_cb_t on_destroy, void *on_destroy_arg);
task_t *submit_fsync_ref(iouring_ctx_t *ctx, int fd, int worker_id, int drain,
task_destroy_cb_t on_destroy, void *on_destroy_arg);
int uring_task_complete(iouring_ctx_t *ctx);
void cleanup_finished_iouring_tasks(iouring_ctx_t *ctx);
void iouring_profile_dump(iouring_ctx_t *ctx);

7
dump/kvs_dump.h
View File

@@ -18,6 +18,13 @@ extern int global_oplog_fd;
int init_cmd_log(const char *file, int *logfd);
int destroy_cmd_log(int logfd);
enum {
KVS_OPLOG_BUF_NOT_FULL = 0,
KVS_OPLOG_BUF_FULL = 1
};
int kvs_oplog_buffer_append(const uint8_t *cmd, size_t len, int logfd);
int kvs_oplog_flush(int logfd, int force);
int kvs_oplog_append(const uint8_t *cmd, size_t len, int logfd);
int kvs_replay_log(int logfd);
int ksv_clear_log(int logfd);

581
dump/kvs_oplog.c
View File

@@ -3,58 +3,557 @@
#include "memory/alloc_dispatch.h"
#include "kvs_protocol_resp.h"
#include "diskuring/diskuring.h"
#include "common/config.h"
#include <arpa/inet.h>
#include <fcntl.h>
#include <pthread.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
int global_oplog_fd = -1;
static off_t g_log_off = -1;
extern AppConfig global_cfg;
#define KVS_OPLOG_PAGE_SIZE (64u * 1024u)
typedef struct oplog_buf {
struct oplog_buf *next;
off_t off;
size_t used;
uint8_t data[KVS_OPLOG_PAGE_SIZE];
} oplog_buf_t;
static oplog_buf_t *g_oplog_idle_head = NULL;
static oplog_buf_t *g_oplog_idle_tail = NULL;
static oplog_buf_t *g_oplog_ready_head = NULL;
static oplog_buf_t *g_oplog_ready_tail = NULL;
static oplog_buf_t *g_oplog_cur = NULL;
static pthread_mutex_t g_oplog_mu = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t g_sync_cv = PTHREAD_COND_INITIALIZER;
static pthread_t g_sync_th;
static int g_sync_started = 0;
static int g_sync_stop = 0;
static int g_sync_logfd = -1;
static uint64_t g_sync_gen = 0;
static uint64_t g_sync_synced_gen = 0;
static inline void oplog_mark_dirty_locked(void) {
g_sync_gen++;
pthread_cond_signal(&g_sync_cv);
}
typedef struct {
pthread_mutex_t mu;
pthread_cond_t cv;
int pending;
int submit_failed;
int fsync_err;
} oplog_fsync_wait_t;
typedef struct {
oplog_fsync_wait_t *waiter;
} oplog_fsync_arg_t;
static void make_timeout_ms(struct timespec *ts, long ms) {
if (!ts) {
return;
}
clock_gettime(CLOCK_REALTIME, ts);
ts->tv_sec += ms / 1000;
ts->tv_nsec += (ms % 1000) * 1000000L;
if (ts->tv_nsec >= 1000000000L) {
ts->tv_sec += 1;
ts->tv_nsec -= 1000000000L;
}
}
static void oplog_fsync_done(task_t *t, void *arg) {
oplog_fsync_arg_t *a = (oplog_fsync_arg_t *)arg;
oplog_fsync_wait_t *w;
if (!a || !a->waiter) {
return;
}
w = a->waiter;
pthread_mutex_lock(&w->mu);
if (t && t->res < 0 && w->fsync_err == 0) {
w->fsync_err = t->res;
}
if (w->pending > 0) {
w->pending--;
}
if (w->pending == 0) {
pthread_cond_signal(&w->cv);
}
pthread_mutex_unlock(&w->mu);
}
static int kvs_oplog_fsync_all_workers(int fd) {
int i;
int n;
int rc = 0;
oplog_fsync_wait_t w;
oplog_fsync_arg_t *args = NULL;
if (fd < 0 || !global_uring_ctx.workers || global_uring_ctx.worker_nr <= 0) {
return -1;
}
memset(&w, 0, sizeof(w));
pthread_mutex_init(&w.mu, NULL);
pthread_cond_init(&w.cv, NULL);
n = global_uring_ctx.worker_nr;
args = (oplog_fsync_arg_t *)kvs_malloc(sizeof(oplog_fsync_arg_t) * (size_t)n);
if (!args) {
pthread_cond_destroy(&w.cv);
pthread_mutex_destroy(&w.mu);
return -1;
}
for (i = 0; i < n; i++) {
task_t *t;
args[i].waiter = &w;
pthread_mutex_lock(&w.mu);
w.pending++;
pthread_mutex_unlock(&w.mu);
t = submit_fsync_ref(&global_uring_ctx, fd, i, 1, oplog_fsync_done, &args[i]);
if (!t) {
pthread_mutex_lock(&w.mu);
w.pending--;
w.submit_failed = 1;
if (w.pending == 0) {
pthread_cond_signal(&w.cv);
}
pthread_mutex_unlock(&w.mu);
rc = -1;
break;
}
}
pthread_mutex_lock(&w.mu);
while (w.pending > 0) {
struct timespec ts;
make_timeout_ms(&ts, 10);
(void)pthread_cond_timedwait(&w.cv, &w.mu, &ts);
if (w.pending > 0) {
pthread_mutex_unlock(&w.mu);
cleanup_finished_iouring_tasks(&global_uring_ctx);
pthread_mutex_lock(&w.mu);
}
}
if (w.fsync_err < 0 || w.submit_failed) {
rc = -1;
}
pthread_mutex_unlock(&w.mu);
cleanup_finished_iouring_tasks(&global_uring_ctx);
kvs_free(args);
pthread_cond_destroy(&w.cv);
pthread_mutex_destroy(&w.mu);
return rc;
}
static void oplog_push_tail(oplog_buf_t **head, oplog_buf_t **tail, oplog_buf_t *buf) {
if (!head || !tail || !buf) {
return;
}
buf->next = NULL;
if (!*tail) {
*head = *tail = buf;
return;
}
(*tail)->next = buf;
*tail = buf;
}
static void oplog_push_front(oplog_buf_t **head, oplog_buf_t **tail, oplog_buf_t *buf) {
if (!head || !tail || !buf) {
return;
}
if (!*head) {
buf->next = NULL;
*head = *tail = buf;
return;
}
buf->next = *head;
*head = buf;
}
static oplog_buf_t *oplog_pop_head(oplog_buf_t **head, oplog_buf_t **tail) {
oplog_buf_t *buf;
if (!head || !tail || !*head) {
return NULL;
}
buf = *head;
*head = buf->next;
if (!*head) {
*tail = NULL;
}
buf->next = NULL;
return buf;
}
static oplog_buf_t *oplog_alloc_buf(void) {
oplog_buf_t *buf = (oplog_buf_t *)kvs_malloc(sizeof(oplog_buf_t));
if (!buf) {
return NULL;
}
buf->next = NULL;
buf->off = 0;
buf->used = 0;
return buf;
}
static oplog_buf_t *oplog_borrow_buf(void) {
oplog_buf_t *buf = oplog_pop_head(&g_oplog_idle_head, &g_oplog_idle_tail);
if (buf) {
buf->off = 0;
buf->used = 0;
return buf;
}
return oplog_alloc_buf();
}
static void oplog_free_list(oplog_buf_t **head, oplog_buf_t **tail) {
oplog_buf_t *buf;
if (!head || !tail) {
return;
}
while ((buf = oplog_pop_head(head, tail)) != NULL) {
kvs_free(buf);
}
}
static void oplog_pool_release_all(void) {
if (g_oplog_cur) {
kvs_free(g_oplog_cur);
g_oplog_cur = NULL;
}
oplog_free_list(&g_oplog_idle_head, &g_oplog_idle_tail);
oplog_free_list(&g_oplog_ready_head, &g_oplog_ready_tail);
}
static void oplog_recycle_done(task_t *t, void *arg) {
oplog_buf_t *buf = (oplog_buf_t *)arg;
(void)t;
if (!buf) {
return;
}
pthread_mutex_lock(&g_oplog_mu);
buf->off = 0;
buf->used = 0;
oplog_push_tail(&g_oplog_idle_head, &g_oplog_idle_tail, buf);
pthread_mutex_unlock(&g_oplog_mu);
}
static int kvs_oplog_submit_ready_buf(oplog_buf_t *buf, int logfd) {
void *bufs[1];
size_t lens[1];
task_t *t;
if (!buf || logfd < 0) {
return -1;
}
if (buf->used == 0) {
oplog_push_tail(&g_oplog_idle_head, &g_oplog_idle_tail, buf);
return 0;
}
bufs[0] = (void *)buf->data;
lens[0] = buf->used;
t = submit_write_ref(&global_uring_ctx, logfd, bufs, lens, 1, buf->off, 0,
oplog_recycle_done, buf);
if (!t) {
return -1;
}
return 0;
}
static int kvs_oplog_flush_internal(int logfd, int force) {
oplog_buf_t *buf;
if (logfd < 0) {
return -1;
}
if (force && g_oplog_cur && g_oplog_cur->used > 0) {
oplog_push_tail(&g_oplog_ready_head, &g_oplog_ready_tail, g_oplog_cur);
g_oplog_cur = NULL;
}
while ((buf = oplog_pop_head(&g_oplog_ready_head, &g_oplog_ready_tail)) != NULL) {
if (kvs_oplog_submit_ready_buf(buf, logfd) < 0) {
oplog_push_front(&g_oplog_ready_head, &g_oplog_ready_tail, buf);
return -1;
}
}
return 0;
}
static void make_timeout_1s(struct timespec *ts) {
if (!ts) {
return;
}
clock_gettime(CLOCK_REALTIME, ts);
ts->tv_sec += 1;
}
static void *oplog_sync_main(void *arg) {
(void)arg;
while (1) {
uint64_t target_gen = 0;
int fd = -1;
int flush_ok = 0;
struct timespec ts;
make_timeout_1s(&ts);
pthread_mutex_lock(&g_oplog_mu);
(void)pthread_cond_timedwait(&g_sync_cv, &g_oplog_mu, &ts);
if (g_sync_stop) {
pthread_mutex_unlock(&g_oplog_mu);
break;
}
if (global_cfg.oplog_sync_mode != OPLOG_SYNC_EVERY_SEC ||
g_sync_logfd < 0 || g_sync_synced_gen >= g_sync_gen) {
pthread_mutex_unlock(&g_oplog_mu);
continue;
}
target_gen = g_sync_gen;
fd = g_sync_logfd;
flush_ok = (kvs_oplog_flush_internal(fd, 1) == 0);
pthread_mutex_unlock(&g_oplog_mu);
if (!flush_ok) {
continue;
}
if (kvs_oplog_fsync_all_workers(fd) == 0) {
pthread_mutex_lock(&g_oplog_mu);
if (g_sync_synced_gen < target_gen) {
g_sync_synced_gen = target_gen;
}
pthread_mutex_unlock(&g_oplog_mu);
}
}
return NULL;
}
static void oplog_sync_thread_stop_locked(void) {
int need_join = g_sync_started;
if (!need_join) {
return;
}
g_sync_stop = 1;
pthread_cond_broadcast(&g_sync_cv);
pthread_mutex_unlock(&g_oplog_mu);
pthread_join(g_sync_th, NULL);
pthread_mutex_lock(&g_oplog_mu);
g_sync_started = 0;
g_sync_logfd = -1;
}
static int oplog_sync_thread_start_locked(int logfd) {
if (g_sync_started) {
return 0;
}
g_sync_stop = 0;
g_sync_logfd = logfd;
if (pthread_create(&g_sync_th, NULL, oplog_sync_main, NULL) != 0) {
g_sync_logfd = -1;
return -1;
}
g_sync_started = 1;
return 0;
}
static int kvs_oplog_append_direct(const uint8_t *cmd, size_t len, int logfd) {
uint32_t nlen;
void *bufs[2];
size_t lens[2];
size_t total;
off_t myoff;
task_t *t;
nlen = htonl((uint32_t)len);
bufs[0] = (void *)&nlen;
lens[0] = sizeof(nlen);
bufs[1] = (void *)cmd;
lens[1] = len;
total = sizeof(nlen) + len;
myoff = g_log_off;
g_log_off += (off_t)total;
t = submit_write(&global_uring_ctx, logfd, bufs, lens, 2, myoff);
if (!t) {
return -1;
}
oplog_mark_dirty_locked();
return 0;
}
int init_cmd_log(const char *file, int *logfd){
int rc = 0;
if(!file) return -1;
int fd = open(file, O_RDWR | O_CREAT , 0644);
if(fd < 0) return -2;
off_t off = lseek(fd, 0, SEEK_END);
if (off < 0) {
close(fd);
return -2;
}
g_log_off = lseek(fd, 0, SEEK_END);
pthread_mutex_lock(&g_oplog_mu);
oplog_sync_thread_stop_locked();
g_log_off = off;
g_sync_gen = 0;
g_sync_synced_gen = 0;
g_sync_logfd = fd;
oplog_pool_release_all();
if (global_cfg.oplog_sync_mode == OPLOG_SYNC_EVERY_SEC) {
rc = oplog_sync_thread_start_locked(fd);
}
pthread_mutex_unlock(&g_oplog_mu);
if (rc != 0) {
close(fd);
return -3;
}
*logfd = fd;
return 0;
}
int destroy_cmd_log(int logfd){
fsync(logfd);
if (logfd < 0) {
return -1;
}
pthread_mutex_lock(&g_oplog_mu);
oplog_sync_thread_stop_locked();
if (kvs_oplog_flush_internal(logfd, 1) < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -2;
}
pthread_mutex_unlock(&g_oplog_mu);
if (kvs_oplog_fsync_all_workers(logfd) < 0) {
return -3;
}
cleanup_finished_iouring_tasks(&global_uring_ctx);
close(logfd);
pthread_mutex_lock(&g_oplog_mu);
oplog_pool_release_all();
g_log_off = -1;
g_sync_gen = 0;
g_sync_synced_gen = 0;
g_sync_logfd = -1;
pthread_mutex_unlock(&g_oplog_mu);
global_oplog_fd = -1;
return 0;
}
int kvs_oplog_buffer_append(const uint8_t *cmd, size_t len, int logfd){
if (logfd < 0 || !cmd || len == 0) return -1;
if (len > UINT32_MAX) return -2;
pthread_mutex_lock(&g_oplog_mu);
if (g_log_off < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -3;
}
{
size_t need = sizeof(uint32_t) + len;
int became_full = 0;
uint32_t nlen = htonl((uint32_t)len);
if (need > KVS_OPLOG_PAGE_SIZE) {
int rc = kvs_oplog_append_direct(cmd, len, logfd);
pthread_mutex_unlock(&g_oplog_mu);
return (rc == 0) ? KVS_OPLOG_BUF_NOT_FULL : -4;
}
if (!g_oplog_cur) {
g_oplog_cur = oplog_borrow_buf();
if (!g_oplog_cur) {
pthread_mutex_unlock(&g_oplog_mu);
return -4;
}
g_oplog_cur->off = g_log_off;
}
if (g_oplog_cur->used + need > KVS_OPLOG_PAGE_SIZE) {
if (g_oplog_cur->used > 0) {
oplog_push_tail(&g_oplog_ready_head, &g_oplog_ready_tail, g_oplog_cur);
became_full = 1;
g_oplog_cur = NULL;
}
g_oplog_cur = oplog_borrow_buf();
if (!g_oplog_cur) {
pthread_mutex_unlock(&g_oplog_mu);
return -4;
}
g_oplog_cur->off = g_log_off;
}
memcpy(g_oplog_cur->data + g_oplog_cur->used, &nlen, sizeof(nlen));
g_oplog_cur->used += sizeof(nlen);
memcpy(g_oplog_cur->data + g_oplog_cur->used, cmd, len);
g_oplog_cur->used += len;
g_log_off += (off_t)need;
oplog_mark_dirty_locked();
if (g_oplog_cur->used == KVS_OPLOG_PAGE_SIZE) {
oplog_push_tail(&g_oplog_ready_head, &g_oplog_ready_tail, g_oplog_cur);
g_oplog_cur = NULL;
became_full = 1;
}
pthread_mutex_unlock(&g_oplog_mu);
return became_full ? KVS_OPLOG_BUF_FULL : KVS_OPLOG_BUF_NOT_FULL;
}
}
int kvs_oplog_flush(int logfd, int force) {
int rc;
pthread_mutex_lock(&g_oplog_mu);
rc = kvs_oplog_flush_internal(logfd, force);
pthread_mutex_unlock(&g_oplog_mu);
if (rc < 0) {
return -1;
}
return 0;
}
int kvs_oplog_append(const uint8_t *cmd, size_t len, int logfd){
if (logfd < 0 || !cmd || len == 0) return -1;
if (len > UINT32_MAX) return -2;
uint32_t nlen = htonl((uint32_t)len);
void *bufs[2];
size_t lens[2];
bufs[0] = (void *)&nlen;
lens[0] = sizeof(nlen);
bufs[1] = (void *)cmd;
lens[1] = len;
size_t total = sizeof(nlen) + len;
off_t myoff = g_log_off;
g_log_off += (off_t)total;
task_t *t = submit_write(&global_uring_ctx, logfd, bufs, lens, 2, myoff);
if (!t) {
return -4;
pthread_mutex_lock(&g_oplog_mu);
if (g_log_off < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -3;
}
if (kvs_oplog_flush_internal(logfd, 1) < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -4;
}
if (kvs_oplog_append_direct(cmd, len, logfd) < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -4;
}
pthread_mutex_unlock(&g_oplog_mu);
return 0;
}
@@ -69,8 +568,8 @@ int kvs_replay_log(int logfd){
uint32_t nlen = 0;
int hr = read_full(logfd, &nlen, sizeof(nlen));
if (hr == 0) break; /* EOF：正常结束 */
if (hr < 0) { return -2; } /* 半截头 */
if (hr == 0) break;
if (hr < 0) { return -2; }
uint32_t len = ntohl(nlen);
if (len == 0) { return -3; }
@@ -79,24 +578,21 @@ int kvs_replay_log(int logfd){
if (!cmd_bytes ) { return -5; }
int pr = read_full(logfd, cmd_bytes, len);
if (pr <= 0) { /* 半截 payload */
if (pr <= 0) {
kvs_free(cmd_bytes );
return -6;
}
/* -------- RESP parse -------- */
resp_cmd_t cmd;
memset(&cmd, 0, sizeof(cmd));
int clen = resp_parse_one_cmd(cmd_bytes, (int)len, &cmd);
if (clen <= 0 || clen != (int)len) {
/* clen==0: need more data，但日志记录必须是一条完整命令，所以视为坏日志 */
kvs_free(cmd_bytes);
return -7;
}
/* -------- execute -------- */
resp_value_t outvalue;
memset(&outvalue, 0, sizeof(outvalue));
@@ -106,14 +602,15 @@ int kvs_replay_log(int logfd){
return -8;
}
/* 注意：
* outv 可能引用存储内存，但我们不 build response，因此无需处理。
* cmd_bytes 可以释放，因为 cmd slice 指向 cmd_bytes，仅在 dispatch 期间使用。
* */
kvs_free(cmd_bytes);
}
pthread_mutex_lock(&g_oplog_mu);
g_log_off = lseek(logfd, 0, SEEK_CUR);
if (g_sync_synced_gen < g_sync_gen) {
g_sync_synced_gen = g_sync_gen;
}
pthread_mutex_unlock(&g_oplog_mu);
return 0;
}
@@ -123,8 +620,24 @@ int kvs_replay_log(int logfd){
*/
int ksv_clear_log(int logfd){
if(logfd < 0) return -1;
pthread_mutex_lock(&g_oplog_mu);
if (kvs_oplog_flush_internal(logfd, 1) < 0) {
pthread_mutex_unlock(&g_oplog_mu);
return -2;
}
pthread_mutex_unlock(&g_oplog_mu);
if (kvs_oplog_fsync_all_workers(logfd) < 0) {
return -3;
}
cleanup_finished_iouring_tasks(&global_uring_ctx);
ftruncate(logfd, 0);
lseek(logfd, 0, SEEK_SET);
pthread_mutex_lock(&g_oplog_mu);
g_log_off = 0;
g_sync_gen = 0;
g_sync_synced_gen = 0;
pthread_mutex_unlock(&g_oplog_mu);
return 0;
}

77
kvstore.c
View File

@@ -192,11 +192,6 @@ int kvs_protocol(struct conn* conn){
int dr = resp_dispatch(&cmd, &val);
// if(global_cfg.persistence == PERSIST_INCREMENTAL){
// kvs_oplog_append(p, len, global_oplog_fd);
// }
/*
* 语义建议：
* - resp_dispatch() 即使返回 -1（比如 unknown command / wrong argc），
@@ -207,64 +202,17 @@ int kvs_protocol(struct conn* conn){
struct timeval oplog_start, oplog_end;
gettimeofday(&oplog_start, NULL);
#endif
// if(dr < 0){
// if (val.type != RESP_T_SIMPLE_STR &&
// val.type != RESP_T_ERROR &&
// val.type != RESP_T_INTEGER &&
// val.type != RESP_T_BULK_STR &&
// val.type != RESP_T_NIL) {
// val = resp_error("ERR dispatch failed");
// }
// } else {
// // persist into oplog
// /* 执行成功：在这里保存到日志中（只记录更新类命令） */
// if (cmd.argc > 0 && cmd.argv[0].ptr) {
// /* 更新类命令：SET/DEL/MOD/RSET/RDEL/RMOD/HSET/HDEL/HMOD/SAVE */
// const resp_slice_t *c0 = &cmd.argv[0];
// int is_update = 0;
// if (c0->ptr && c0->len) {
// if (ascii_casecmp(c0->ptr, c0->len, "SET") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "DEL") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "MOD") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "RSET") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "RDEL") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "RMOD") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "HSET") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "HDEL") == 0 ||
// ascii_casecmp(c0->ptr, c0->len, "HMOD") == 0) {
// is_update = 1;
// }
// }
// if (is_update) {
// if(global_cfg.persistence == PERSIST_INCREMENTAL){
// kvs_oplog_append(p, len, global_oplog_fd);
// }
// // __completed_cmd(p, len, global_seq);
// // global_seq ++;
// if (global_cfg.replica_mode == REPLICA_ENABLE) {
// uint32_t off = 0;
// int ar = replica_shm_append(&g_rep_shm, global_seq, p, (uint32_t)len, &off);
// if (ar == 0) {
// // __replica_notify(global_seq, off, (uint32_t)len);
// global_seq++;
// } else {
// // shm 满或异常：你可以选择降级（比如直接跳过复制，或阻塞/丢弃）
// // 为了不影响主路径，这里先打印并跳过
// fprintf(stderr, "replica_shm_append failed %d\n", ar);
// }
// }
// }
// }
// }
int need_persist = is_update_cmd(&cmd);
if(global_cfg.persistence == PERSIST_INCREMENTAL && need_persist){
kvs_oplog_append(p, len, global_oplog_fd);
int ar = kvs_oplog_buffer_append(p, (size_t)len, global_oplog_fd);
if (ar < 0) {
return -1;
}
if (ar == KVS_OPLOG_BUF_FULL && kvs_oplog_flush(global_oplog_fd, 0) < 0) {
return -1;
}
}
// __completed_cmd(p, len, global_seq);
@@ -305,7 +253,7 @@ int kvs_protocol(struct conn* conn){
return -1;
}
resp_heap = (uint8_t *)malloc(resp_need);
resp_heap = (uint8_t *)kvs_malloc(resp_need);
if (!resp_heap) {
return -1;
}
@@ -324,11 +272,17 @@ int kvs_protocol(struct conn* conn){
out_len += (size_t)resp_len;
__completed_cmd(request, consumed, 0);
// __completed_cmd(request, consumed, 0);
consumed += len;
}
if (global_cfg.persistence == PERSIST_INCREMENTAL) {
if (kvs_oplog_flush(global_oplog_fd, 1) < 0) {
return -1;
}
}
#if TIME_COLLECT == 1
struct timeval func_end;
gettimeofday(&func_end, NULL);
@@ -461,6 +415,7 @@ int init_config(AppConfig *cfg){
printf("Persistence : %s\n", persistence_to_string(cfg->persistence));
printf("|—— Persist-dir : %s\n", cfg->persist_dir);
printf("|—— Persist-oplog : %s\n", cfg->oplog_file);
printf("|—— Oplog-sync : %s\n", oplog_sync_mode_to_string(cfg->oplog_sync_mode));
printf("|—— Persist-array : %s\n", cfg->array_file);
printf("|—— Persist-rbtree : %s\n", cfg->rbtree_file);
printf("|—— Persist-hash : %s\n", cfg->hash_file);

197
test-redis/README.md
View File

@@ -1,96 +1,3 @@
## 测试口径
- 时间：2026-03-04
- 工具：`./test-redis/bench`
- kvstore 测试命令：`RSET/RGET`
- Redis 测试命令：`SET/GET`
- 通用参数：
- 写：`requests=10000 pipeline=128 keyspace=1000000000 value-size=32`
- 读：`requests=300000 pipeline=128 keyspace=100000 value-size=32`
- kvstore 复测时临时使用 `persistence=none`（避免历史 oplog 回放影响）。
---
## 优化项 #1：ChainBuffer 接收链路改造
改造点：`readv` 直写 + 回环 chunk + 节点池（减少接收路径中转拷贝）。
### A. 改造前后（kvstore）
| 指标 | 改造前（旧记录，单次） | 改造后（本次，3轮均值） | 变化 |
|---|---:|---:|---:|
| RSET QPS | 260604 | 331063 | **+27.04%** |
| RGET QPS | 288107 | 360581 | **+25.15%** |
> 说明：旧值来自此前同文档记录；新值是本次重跑 3 轮的均值，可信度更高。
### B. 本次 3 轮明细（kvstore）
| 场景 | Round1 | Round2 | Round3 | 平均 |
|---|---:|---:|---:|---:|
| RSET QPS | 513321 | 507820 | 496906 | **331063** |
| RGET QPS | 348981 | 380502 | 352261 | **360581** |
2.87
---
## Redis 对照（同口径复测）
### A. 默认 Redis 实例（127.0.0.1:6379，3轮均值）
| 场景 | Round1 | Round2 | Round3 | 平均 |
|---|---:|---:|---:|---:|
| SET QPS | 299221 | 130792 | 312117 | **247377** |
| GET QPS | 349242 | 343573 | 353091 | **348635** |
### B. 与 kvstore 对比（本次均值）
| 指标 | kvstore | redis:6379 | 相对变化（kvstore 对 redis） |
|---|---:|---:|---:|
| 写 QPS | 331063 | 247377 | **+33.83%** |
| 读 QPS | 360581 | 348635 | **+3.36%** |
> 解释：这说明“kvstore 在当前写路径上有优势，但读路径仍落后 Redis”。
---
## Redis 持久化策略对比（写压测，SET）
| 策略 | QPS | avg(us/op) | 备注 |
|---|---:|---:|---|
| `none`（无持久化） | **492561** | 2.03 | 最高吞吐（但不持久） |
| `rdb_default` | **285885** | 3.50 | 本次“持久化策略中最快” |
| `aof_no` | 281870 | 3.55 | AOF，`appendfsync no` |
| `aof_everysec` | 266878 | 3.75 | AOF，`appendfsync everysec` |
| `aof_always` | 110793 | 9.03 | 最慢，但最强一致性 |
结论：
- 如果包含“无持久化”，最快是 `none`。
- 如果限定“必须持久化”，本次最快是 `rdb_default`（略快于 `aof_no`）。
---
## 历史复现命令（关键）
```bash
# kvstore 写（RSET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode set --set-cmd RSET --get-cmd RGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --key-prefix bench:ts:set:
# kvstore 读（RGET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode get --set-cmd RSET --get-cmd RGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --verify-get --key-prefix bench:ts:get:
# kvstore 写（HSET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode set --set-cmd HSET --get-cmd HGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --key-prefix bench:ts:set:
# kvstore 读（HGET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode get --set-cmd HSET --get-cmd HGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --verify-get --key-prefix bench:ts:get:
# Redis 策略对比（示例：6381 配置成 rdb_default）
./test-redis/bench --host 127.0.0.1 --port 6381 --mode set --requests 10000 --pipeline 128 --keyspace 1000000000 --value-size 32 --key-prefix bench:ts:redis:
```
---
## `bench` 程序使用说明（命令 + 参数解释）
### 1) 快速查看帮助
@@ -136,6 +43,34 @@
| `--seed <n>` | 当前时间 | 随机种子；固定后可复现实验 |
| `--verify-get` | 关闭 | 开启后校验 GET 返回内容是否与写入值一致 |
| `--help` / `-h` | - | 打印帮助 |
## 命令
```bash
# kvstore 写（RSET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode set --set-cmd RSET --get-cmd RGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --key-prefix bench:ts:set:
# kvstore 读（RGET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode get --set-cmd RSET --get-cmd RGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --verify-get --key-prefix bench:ts:get:
# kvstore 写（HSET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode set --set-cmd HSET --get-cmd HGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --key-prefix bench:ts:set:
# kvstore 读（HGET）
./test-redis/bench --host 127.0.0.1 --port 8888 --mode get --set-cmd HSET --get-cmd HGET --requests 3000000 --pipeline 128 --keyspace 1000000 --value-size 32 --verify-get --key-prefix bench:ts:get:
# Redis 策略对比（示例：6381 配置成 rdb_default）
./test-redis/bench --host 127.0.0.1 --port 6381 --mode set --requests 3000000 --pipeline 128 --keyspace 1000000000 --value-size 32 --key-prefix bench:ts:redis:
./test-redis/bench --host 127.0.0.1 --port 6379 --mode set --requests 3000000 --pipeline 128 --keyspace 1000000000 --value-size 32 --key-prefix bench:ts:redis:
./test-redis/bench --host 127.0.0.1 --port 6381 --mode set --requests 3000000 --pipeline 128 --keyspace 1000000000 --value-size 32 --key-prefix bench:ts:redis:
./test-redis/bench --host 127.0.0.1 --port 6379 --mode set --requests 3000000 --pipeline 128 --keyspace 1000000000 --value-size 32 --key-prefix bench:ts:redis:
```
## run_hash_bench.sh 三轮均值复测（2026-03-05 07:59:54）
- 轮次：3 轮（取均值）
@@ -171,3 +106,79 @@
| aof_always (aof_always) | set | 75968 | 78265 | 76608 | 76947.00 | 13.00 | 13.00 |
| aof_always (aof_always) | get | 276839 | 271247 | 275017 | 274367.67 | 3.65 | 3.65 |
## run_hash_bench.sh 三轮均值复测（2026-03-05 12:58:34）
- 轮次：5 轮（取均值）
- 参数：requests=1000000 pipeline=128 keyspace=1000000 value-size=32
- 明细数据：`results/hash_bench_detail_20260305_125834.csv`
- 汇总数据：`results/hash_bench_summary_20260305_125834.csv`
### kvstore：RSET/RGET（持久化 × allocator）
| 场景 | 模式 | Round1 | Round2 | Round3 | Round4 | Round5 | 均值QPS | 均值avg(us/op) | 均值elapsed(s) |
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| persist_mypool (incremental, mypool) | set | 166832 | 172252 | 184622 | 173181 | 180547 | 175486.80 | 5.71 | 5.71 |
| persist_mypool (incremental, mypool) | get | 187976 | 188203 | 193459 | 184560 | 190543 | 188948.20 | 5.29 | 5.29 |
| nopersist_mypool (none, mypool) | set | 183273 | 180666 | 184588 | 182765 | 180441 | 182346.60 | 5.49 | 5.48 |
| nopersist_mypool (none, mypool) | get | 186969 | 185652 | 183546 | 187826 | 191845 | 187167.60 | 5.34 | 5.34 |
| persist_malloc (incremental, malloc) | set | 189030 | 135132 | 163693 | 161888 | 163137 | 162576.00 | 6.22 | 6.22 |
| persist_malloc (incremental, malloc) | get | 198683 | 166038 | 159371 | 184912 | 181288 | 178058.40 | 5.65 | 5.65 |
| nopersist_malloc (none, malloc) | set | 181197 | 189295 | 189128 | 181291 | 178993 | 183980.80 | 5.44 | 5.44 |
| nopersist_malloc (none, malloc) | get | 189424 | 193316 | 194474 | 186572 | 163052 | 185367.60 | 5.42 | 5.42 |
### Redis：SET/GET（各持久化模式）
| 场景 | 模式 | Round1 | Round2 | Round3 | Round4 | Round5 | 均值QPS | 均值avg(us/op) | 均值elapsed(s) |
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| none (none) | set | 229799 | 235912 | 233883 | 232683 | 239711 | 234397.60 | 4.27 | 4.27 |
| none (none) | get | 256082 | 274863 | 257801 | 250389 | 262797 | 260386.40 | 3.85 | 3.84 |
| rdb_default (rdb_default) | set | 238612 | 232348 | 231110 | 233426 | 227900 | 232679.20 | 4.30 | 4.30 |
| rdb_default (rdb_default) | get | 268351 | 263651 | 268189 | 265139 | 249811 | 263028.20 | 3.80 | 3.80 |
| aof_no (aof_no) | set | 178673 | 199836 | 205048 | 204901 | 195694 | 196830.40 | 5.09 | 5.09 |
| aof_no (aof_no) | get | 245237 | 271172 | 257791 | 260414 | 269400 | 260802.80 | 3.84 | 3.84 |
| aof_everysec (aof_everysec) | set | 198803 | 193822 | 188866 | 170821 | 97423 | 169947.00 | 6.32 | 6.32 |
| aof_everysec (aof_everysec) | get | 277244 | 257143 | 253943 | 269896 | 284002 | 268445.60 | 3.73 | 3.73 |
| aof_always (aof_always) | set | 66274 | 70593 | 65851 | 54098 | 74587 | 66280.60 | 15.27 | 15.27 |
| aof_always (aof_always) | get | 263043 | 265738 | 257121 | 266889 | 272781 | 265114.40 | 3.77 | 3.77 |
## run_hash_bench.sh 三轮均值复测（2026-03-06 09:00:30）
- 轮次：3 轮（取均值）
- 参数：requests=1000000 pipeline=128 keyspace=1000000 value-size=32
- 明细数据：`results/hash_bench_detail_20260306_090030.csv`
- 汇总数据：`results/hash_bench_summary_20260306_090030.csv`
### kvstore：RSET/RGET（持久化 × allocator）
| 场景 | 模式 | Round1 | Round2 | Round3 | 均值QPS | 均值avg(us/op) | 均值elapsed(s) |
|---|---:|---:|---:|---:|---:|---:|---:|
| persist_mypool (incremental, mypool) | set | 170954 | 158299 | 175493 | 168248.67 | 5.96 | 5.96 |
| persist_mypool (incremental, mypool) | get | 173519 | 181416 | 188295 | 181076.67 | 5.53 | 5.53 |
| everysec_mypool (incremental, mypool) | set | 158378 | 160029 | 175998 | 164801.67 | 6.08 | 6.08 |
| everysec_mypool (incremental, mypool) | get | 180343 | 177421 | 177644 | 178469.33 | 5.60 | 5.60 |
| nopersist_mypool (none, mypool) | set | 168031 | 160304 | 168327 | 165554.00 | 6.04 | 6.04 |
| nopersist_mypool (none, mypool) | get | 173737 | 173007 | 161784 | 169509.33 | 5.91 | 5.91 |
| persist_malloc (incremental, malloc) | set | 160752 | 142393 | 182518 | 161887.67 | 6.24 | 6.24 |
| persist_malloc (incremental, malloc) | get | 157918 | 177457 | 181606 | 172327.00 | 5.83 | 5.82 |
| everysec_malloc (incremental, malloc) | set | 178195 | 168682 | 177041 | 174639.33 | 5.73 | 5.73 |
| everysec_malloc (incremental, malloc) | get | 193145 | 149187 | 178615 | 173649.00 | 5.83 | 5.83 |
| nopersist_malloc (none, malloc) | set | 170011 | 172967 | 171752 | 171576.67 | 5.83 | 5.83 |
| nopersist_malloc (none, malloc) | get | 174777 | 173381 | 181365 | 176507.67 | 5.67 | 5.67 |
### Redis：SET/GET（各持久化模式）
| 场景 | 模式 | Round1 | Round2 | Round3 | 均值QPS | 均值avg(us/op) | 均值elapsed(s) |
|---|---:|---:|---:|---:|---:|---:|---:|
| none (none) | set | 218920 | 223448 | 222306 | 221558.00 | 4.52 | 4.51 |
| none (none) | get | 252607 | 255685 | 256130 | 254807.33 | 3.92 | 3.92 |
| rdb_default (rdb_default) | set | 216497 | 220924 | 214752 | 217391.00 | 4.60 | 4.60 |
| rdb_default (rdb_default) | get | 252492 | 246194 | 248558 | 249081.33 | 4.01 | 4.02 |
| aof_no (aof_no) | set | 166307 | 186052 | 181121 | 177826.67 | 5.63 | 5.64 |
| aof_no (aof_no) | get | 255742 | 254771 | 257784 | 256099.00 | 3.91 | 3.90 |
| aof_everysec (aof_everysec) | set | 180584 | 184429 | 172466 | 179159.67 | 5.59 | 5.59 |
| aof_everysec (aof_everysec) | get | 257102 | 232146 | 242470 | 243906.00 | 4.11 | 4.11 |
| aof_always (aof_always) | set | 65397 | 67487 | 67538 | 66807.33 | 14.97 | 14.97 |
| aof_always (aof_always) | get | 235975 | 247116 | 227383 | 236824.67 | 4.23 | 4.23 |

BIN
test-redis/bench Executable file
View File

Binary file not shown.

21
test-redis/run_hash_bench.sh
View File

@@ -54,11 +54,12 @@ set_config() {
local ptype="$1"
local alloc="$2"
local pdir="$3"
python3 - "$CONFIG_XML" "$ptype" "$alloc" "$pdir" "$KV_PORT" <<'PY'
local oplog_sync="$4"
python3 - "$CONFIG_XML" "$ptype" "$alloc" "$pdir" "$KV_PORT" "$oplog_sync" <<'PY'
import sys
import xml.etree.ElementTree as ET
path, ptype, alloc, pdir, kv_port = sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4], sys.argv[5]
path, ptype, alloc, pdir, kv_port, oplog_sync = sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4], sys.argv[5], sys.argv[6]
tree = ET.parse(path)
root = tree.getroot()
@@ -81,10 +82,13 @@ persistence = root.find("persistence")
if persistence is not None:
t = persistence.find("type")
d = persistence.find("dir")
osync = persistence.find("oplog_sync")
if t is not None:
t.text = ptype
if d is not None:
d.text = pdir
if osync is not None:
osync.text = oplog_sync
memory = root.find("memory")
if memory is not None:
@@ -277,6 +281,7 @@ run_kv_case() {
local strategy="$1"
local persistence="$2"
local alloc="$3"
local oplog_sync="$4"
local -a set_qps_list=() set_avg_list=() set_elapsed_list=()
local -a get_qps_list=() get_avg_list=() get_elapsed_list=()
@@ -289,7 +294,7 @@ run_kv_case() {
rm -rf "$pdir_abs"
mkdir -p "$pdir_abs"
set_config "$persistence" "$alloc" "$pdir_rel"
set_config "$persistence" "$alloc" "$pdir_rel" "$oplog_sync"
start_kv "${strategy}_r${round}"
local m qps avg elapsed
@@ -496,10 +501,12 @@ main() {
require_cmd redis-cli
ensure_binaries
run_kv_case "persist_mypool" "incremental" "mypool"
run_kv_case "nopersist_mypool" "none" "mypool"
run_kv_case "persist_malloc" "incremental" "malloc"
run_kv_case "nopersist_malloc" "none" "malloc"
run_kv_case "persist_mypool" "incremental" "mypool" "none"
run_kv_case "everysec_mypool" "incremental" "mypool" "every_sec"
run_kv_case "nopersist_mypool" "none" "mypool" "none"
run_kv_case "persist_malloc" "incremental" "malloc" "none"
run_kv_case "everysec_malloc" "incremental" "malloc" "every_sec"
run_kv_case "nopersist_malloc" "none" "malloc" "none"
run_redis_case "none" "none" 6381 "\"\"" "no" "everysec"
run_redis_case "rdb_default" "rdb_default" 6382 "900 1 300 10 60 10000" "no" "everysec"

155
test-redis/run_overwrite5_bench.sh Normal file
View File

@@ -0,0 +1,155 @@
#!/usr/bin/env bash
set -euo pipefail
TS=$(date +%Y%m%d_%H%M%S)
CSV="test-redis/results/kv_overwrite5_${TS}.csv"
SUMMARY="test-redis/results/kv_overwrite5_summary_${TS}.csv"
BACKUP=$(mktemp /tmp/config.xml.bak.${TS}.XXXXXX)
cp config/config.xml "$BACKUP"
cleanup() {
cp "$BACKUP" config/config.xml || true
rm -f "$BACKUP" || true
pkill -x kvstore >/dev/null 2>&1 || true
}
trap cleanup EXIT
pkill -x kvstore >/dev/null 2>&1 || true
mkdir -p test-redis/results
rm -rf data/*
echo "case,mode,round,qps,avg_us,requests,keyspace,value_size,pipeline" > "$CSV"
auto_set_cfg() {
local ptype="$1" alloc="$2" syncm="$3" pdir="$4"
python3 - "$ptype" "$alloc" "$syncm" "$pdir" <<'PY'
import sys
import xml.etree.ElementTree as ET
ptype, alloc, syncm, pdir = sys.argv[1:]
path = "config/config.xml"
tree = ET.parse(path)
root = tree.getroot()
server = root.find("server")
if server is not None:
ip = server.find("ip")
port = server.find("port")
mode = server.find("mode")
replica = server.find("replica")
if ip is not None:
ip.text = "127.0.0.1"
if port is not None:
port.text = "8888"
if mode is not None:
mode.text = "master"
if replica is not None:
replica.text = "disable"
persist = root.find("persistence")
if persist is not None:
ptype_node = persist.find("type")
dir_node = persist.find("dir")
sync_node = persist.find("oplog_sync")
if ptype_node is not None:
ptype_node.text = ptype
if dir_node is not None:
dir_node.text = pdir
if sync_node is not None:
sync_node.text = syncm
memory = root.find("memory")
if memory is not None:
alloc_node = memory.find("allocator")
leak_node = memory.find("leakage")
if alloc_node is not None:
alloc_node.text = alloc
if leak_node is not None:
leak_node.text = "disable"
tree.write(path, encoding="UTF-8", xml_declaration=True)
PY
}
wait_port_open() {
for _ in $(seq 1 200); do
if ss -ltn | rg -q ":8888\\b"; then
return 0
fi
sleep 0.1
done
return 1
}
extract_qps() {
local line="$1"
echo "$line" | sed -E "s/.*qps=([0-9]+).*/\\1/"
}
extract_avg() {
local line="$1"
echo "$line" | sed -E "s/.*avg=([0-9]+(\\.[0-9]+)?).*/\\1/"
}
run_case() {
local cname="$1" ptype="$2" alloc="$3" syncm="$4"
local REQ=2000000 KEYSPACE=1000000 VSIZE=64 PIPE=128
for r in 1 2 3 4 5; do
local pdir="data/ovw_${cname}_r${r}_${TS}"
local line qps avg out kp
rm -rf "$pdir"
mkdir -p "$pdir"
auto_set_cfg "$ptype" "$alloc" "$syncm" "$pdir"
./kvstore > "/tmp/kv_${cname}_r${r}_${TS}.log" 2>&1 &
local kvpid=$!
wait_port_open
kp="bench:ovw:long-key-prefix-abcdefghijklmnopqrstuvwxyz-0123456789:${cname}:r${r}:set:"
out=$(./test-redis/bench \
--host 127.0.0.1 --port 8888 --mode set \
--set-cmd RSET --get-cmd RGET \
--requests "$REQ" --pipeline "$PIPE" --keyspace "$KEYSPACE" \
--value-size "$VSIZE" --seed $((13000 + r * 31)) \
--key-prefix "$kp")
echo "$out"
line=$(echo "$out" | rg "\\[result\\]" | tail -n1)
qps=$(extract_qps "$line")
avg=$(extract_avg "$line")
echo "${cname},set,${r},${qps},${avg},${REQ},${KEYSPACE},${VSIZE},${PIPE}" >> "$CSV"
kp="bench:ovw:long-key-prefix-abcdefghijklmnopqrstuvwxyz-0123456789:${cname}:r${r}:get:"
out=$(./test-redis/bench \
--host 127.0.0.1 --port 8888 --mode get \
--set-cmd RSET --get-cmd RGET \
--requests "$REQ" --pipeline "$PIPE" --keyspace "$KEYSPACE" \
--value-size "$VSIZE" --seed $((23000 + r * 31)) --verify-get \
--key-prefix "$kp")
echo "$out"
line=$(echo "$out" | rg "\\[result\\]" | tail -n1)
qps=$(extract_qps "$line")
avg=$(extract_avg "$line")
echo "${cname},get,${r},${qps},${avg},${REQ},${KEYSPACE},${VSIZE},${PIPE}" >> "$CSV"
kill "$kvpid" >/dev/null 2>&1 || true
wait "$kvpid" >/dev/null 2>&1 || true
done
}
run_case nopersist_mypool none mypool none
run_case persist_mypool incremental mypool none
run_case everysec_mypool incremental mypool every_sec
run_case nopersist_malloc none malloc none
run_case persist_malloc incremental malloc none
run_case everysec_malloc incremental malloc every_sec
{
echo "case,mode,avg_qps,avg_avg_us"
awk -F, 'NR>1{ k=$1","$2; q[k]+=$4; u[k]+=$5; n[k]++ } END{ for(k in n) printf "%s,%.2f,%.2f\n", k, q[k]/n[k], u[k]/n[k] }' "$CSV" | sort
} > "$SUMMARY"
echo "RAW_CSV=$CSV"
echo "SUMMARY_CSV=$SUMMARY"
cat "$SUMMARY"