feat：扩展了结构化日志输出

- 补充了传输层观测指标
  - 增加了 UDP 丢包相关统计
  - README 也同步更新了字段口径

README.md

@@ -211,3 +211,115 @@ put /path/to/file.bin
 - 该场景已经实现 `A -> C -> D -> B` 与 `B -> D -> C -> A` 的桥接转发
 - 但当前 `bridge` 仍是单下游、单身份模型，不是完整的多节点桥接网络
 
+## 日志与指标
+
+### 输出位置
+
+- 终端文本日志：默认输出到 `stderr`，格式为 `key=value`
+- 结构化日志：默认追加到当前工作目录下的 `omni_logs.jsonl`
+- 性能快照统一写在 `component="perf"` 的 JSONL 记录里
+- 终端里会额外出现 `component=perf_udp_loss` 文本行；同一批 UDP 丢包字段已经合并写进对应的 `component="perf"` JSONL 行
+- 周期性性能快照大约每 `1s` 打一次；进程退出时会再打一条 `tag="final"`
+
+### 当前文档口径
+
+下面这些说法要以当前实现为准，不要再按旧文档理解：
+
+- `processing / queue / transmission / propagation / end_to_end` 都是“当前实现下的本地观测值或估算值”，不是严格意义上的物理链路精确测量值。
+- `processing_*` 当前表示本地应用层处理耗时，主要来自文件分片封装、接收端写盘等路径；它不是“某一块硬件 CPU 的完整开销画像”。
+- `queue_*` 和 `transmission_*` 是基于当前吞吐和缓冲/队列状态反推出来的估算值。进程空转时间很长、样本很小或者当前速率很低时，这两个值可能明显偏大。
+- `propagation_*` 当前来自 `min_rtt_ms / 2` 的估算；如果当前协议没有 RTT 样本，这组字段就是 `0`。
+- `end_to_end_*` 当前只在“最终接收文件的 peer”上有值，来源是发送端分片里的 `origin_ts_ms`。发送端、Hub、Bridge 一般是 `0`。
+- UDP 丢包统计只在 `UDP 文件接收侧 peer` 上有值；UDP 发送侧、Hub、Bridge 不会产出这组汇总。
+- `udp_retrans` 字段当前还没有实现应用层 UDP 重传统计，所以现在始终是 `0`。
+- `TCP/KCP` 当前记录的是重传次数/重传字节/累计发送分片，不直接记录“重传频率”这个单独字段。
+- `send_buffer_pct_*` / `recv_buffer_pct_*` 是占用率风格的指标，但不保证永远严格落在 `0-100`；尤其 KCP 等待队列超过窗口时，理论上可以大于 `100`。
+- `send_call_* / recv_call_* / proto_* / processing_*` 有时会是 `0`，常见原因不是没统计，而是当前时间分辨率是毫秒，很多本地操作小于 `1ms`。
+
+### 基础与吞吐字段
+
+| 文档含义 | JSONL 字段 | 单位 | 当前语义 | 什么时候有值 / 为什么会是 0 |
+| --- | --- | --- | --- | --- |
+| 时间戳 | `ts_ms` | ms | 这条日志写出的单调时间戳 | 始终有值 |
+| 日志分类 | `level` / `component` / `tag` | - | `component="perf"` 表示性能快照，`tag` 常见为 `peer_transport_send`、`peer_transport_recv`、`final` | 始终有值 |
+| 身份上下文 | `app` / `proto` / `mode` / `role` / `self_id` | - | 程序名、协议、模式、角色、逻辑 ID | `hub` 的 `self_id` 为空是正常的 |
+| 运行时长 | `elapsed_ms` | ms | 当前进程从启动到本次快照的时长 | 始终有值 |
+| 累计发送字节 | `bytes_sent` | bytes | 当前进程累计发送的协议帧总字节 | 无发送时为 `0` |
+| 累计接收字节 | `bytes_recv` | bytes | 当前进程累计接收的协议帧总字节 | 无接收时为 `0` |
+| 发送次数 | `send_count` | count | 当前进程累计发送帧次数 | 无发送时为 `0` |
+| 接收次数 | `recv_count` | count | 当前进程累计接收帧次数 | 无接收时为 `0` |
+| 瞬时发送带宽 | `tx_current_mbps` | Mbps | 最近一个统计窗口内的发送速率 | 当前窗口没流量时为 `0` |
+| 瞬时接收带宽 | `rx_current_mbps` | Mbps | 最近一个统计窗口内的接收速率 | 当前窗口没流量时为 `0` |
+| 平均发送带宽 | `tx_avg_mbps` | Mbps | 进程启动到当前的平均发送速率 | 从未发送时为 `0` |
+| 平均接收带宽 | `rx_avg_mbps` | Mbps | 进程启动到当前的平均接收速率 | 从未接收时为 `0` |
+| 传输进度字节 | `progress_bytes` | bytes | 当前文件传输已完成字节数 | 没有文件传输时为 `0` |
+| 总工作量 | `total_work_bytes` | bytes | 当前文件总大小 | 没有文件传输时为 `0` |
+| 传输进度百分比 | `progress_pct` | % | `progress_bytes / total_work_bytes * 100` | 没有文件传输时为 `0` |
+
+### 调用耗时与延迟字段
+
+| 文档含义 | JSONL 字段 | 单位 | 当前语义 | 什么时候有值 / 为什么会是 0 |
+| --- | --- | --- | --- | --- |
+| 应用层发送调用 | `send_call_last_ms` / `send_call_min_ms` / `send_call_max_ms` / `send_call_avg_ms` | ms | `peer_transport_send()` 调用耗时 | 没有发送，或每次发送都小于 `1ms` 时可能为 `0` |
+| 应用层接收调用 | `recv_call_last_ms` / `recv_call_min_ms` / `recv_call_max_ms` / `recv_call_avg_ms` | ms | `peer_transport_next_event()` 接收调用耗时 | 没有接收，或每次接收都小于 `1ms` 时可能为 `0` |
+| 协议层发送耗时 | `proto_send_avg_ms` | ms | TCP/UDP/KCP 实际 send 路径耗时 EWMA | 发送很快且小于 `1ms` 时常为 `0` |
+| 协议层接收耗时 | `proto_recv_avg_ms` | ms | TCP/UDP/KCP 实际 recv 路径耗时 EWMA | 接收很快且小于 `1ms` 时常为 `0` |
+| 本地处理耗时 | `processing_avg_ms` / `processing_min_ms` / `processing_max_ms` | ms | 当前进程内的分片封装、写盘等本地处理耗时 | 仅在文件发送/接收路径上采样；操作太快时可能为 `0` |
+| 排队延迟估算 | `queue_avg_ms` / `queue_min_ms` / `queue_max_ms` | ms | 根据发送/接收队列字节数和当前速率估算 | 没有队列样本时为 `0`；长空转后可能偏大 |
+| 传输延迟估算 | `transmission_avg_ms` / `transmission_min_ms` / `transmission_max_ms` | ms | 根据当前速率估算“这些字节推上链路需要多久” | 没有流量样本时为 `0`；小样本/低速率时可能偏大 |
+| 传播延迟估算 | `propagation_avg_ms` / `propagation_min_ms` / `propagation_max_ms` | ms | 基于 `min_rtt_ms / 2` 的估算 | 当前协议没有 RTT 样本时为 `0` |
+| 端到端延迟 | `end_to_end_avg_ms` / `end_to_end_min_ms` / `end_to_end_max_ms` | ms | 发送端分片 `origin_ts_ms` 到接收端处理时刻的差值 | 只在最终接收文件的 `peer` 上有值；发送端 / Hub / Bridge 多数为 `0` |
+
+### 可靠性字段
+
+| 文档含义 | JSONL 字段 | 单位 | 当前语义 | 什么时候有值 / 为什么会是 0 |
+| --- | --- | --- | --- | --- |
+| TCP 重传次数 | `tcp_retrans` | count | 内核 `TCP_INFO` 的累计重传次数 | 仅 TCP 有值；未重传或非 TCP 时为 `0` |
+| TCP 数据段数 | `tcp_data_segs_out` | count | TCP 累计发送数据段数 | 仅 TCP 有值；非 TCP 时为 `0` |
+| TCP 发送字节 | `tcp_data_bytes_sent` | bytes | TCP 累计发送数据字节 | 仅 TCP 有值；非 TCP 时为 `0` |
+| TCP 重传字节 | `tcp_retrans_bytes` | bytes | TCP 累计重传字节 | 仅 TCP 有值；未重传或非 TCP 时为 `0` |
+| UDP 重传次数 | `udp_retrans` | count | 预留字段，当前未实现 | 当前始终为 `0` |
+| KCP 重传次数 | `kcp_retrans` | count | KCP 内部累计重传分片数 | 仅 KCP 有值；未重传或非 KCP 时为 `0` |
+| KCP 数据分片数 | `kcp_data_segs_out` | count | KCP 累计发送分片数 | 仅 KCP 有值；非 KCP 时为 `0` |
+| KCP 发送字节 | `kcp_data_bytes_sent` | bytes | KCP 累计发送分片字节 | 仅 KCP 有值；非 KCP 时为 `0` |
+| KCP 重传字节 | `kcp_retrans_bytes` | bytes | KCP 累计重传字节 | 仅 KCP 有值；未重传或非 KCP 时为 `0` |
+| UDP 预期分片数 | `udp_expected_chunks` | count | UDP 文件接收端预期应收到的总分片数 | 仅 UDP 文件接收侧 `peer` 有值；其他角色为 `0` |
+| UDP 实收分片数 | `udp_received_chunks` | count | UDP 文件接收端实际收到的唯一分片数 | 仅 UDP 文件接收侧 `peer` 有值；其他角色为 `0` |
+| UDP 丢失分片数 | `udp_lost_chunks` | count | 根据 `seq` 推断的缺失分片数 | 完整收到时为 `0`；非 UDP 接收侧也为 `0` |
+| UDP 丢包率 | `udp_loss_rate_pct` | % | `udp_lost_chunks / udp_expected_chunks * 100` | 完整收到时为 `0`；非 UDP 接收侧也为 `0` |
+| UDP 连续丢包区间数 | `udp_loss_burst_count` | count | 丢包区间个数 | 无丢包时为 `0` |
+| UDP 最大连续丢包长度 | `udp_loss_burst_max_len` | count | 单个丢包区间的最大长度 | 无丢包时为 `0` |
+| UDP 丢包区间摘要 | `udp_loss_ranges` | csv string | 例如 `4-6,9,12-13` | 无丢包时为空串 |
+| UDP 丢失序号样本 | `udp_loss_seq_sample` | csv string | 最多记录一部分缺失 `seq` 样本 | 无丢包时为空串 |
+| UDP 接收窗口分布 | `udp_recv_window_dist` | csv string | `window_id:count`，例如 `0:3,1:28` | 仅 UDP 接收侧有值；没有样本时为空串 |
+
+### 资源与算法字段
+
+| 文档含义 | JSONL 字段 | 单位 | 当前语义 | 什么时候有值 / 为什么会是 0 |
+| --- | --- | --- | --- | --- |
+| 发送缓冲区占用 | `send_buffer_pct_last` / `send_buffer_pct_avg` / `send_buffer_pct_max` | % 风格值 | Socket 或 KCP 发送队列占用率样本 | 没有缓冲采样时为 `0`；KCP 拥塞时可能大于 `100` |
+| 接收缓冲区占用 | `recv_buffer_pct_last` / `recv_buffer_pct_avg` / `recv_buffer_pct_max` | % 风格值 | Socket 或 KCP 接收队列占用率样本 | 没有缓冲采样时为 `0` |
+| 拥塞窗口 | `cwnd_last` / `cwnd_avg` / `cwnd_max` | 协议窗口大小 | TCP/KCP 当前拥塞窗口样本 | UDP 没有拥塞窗口，因此为 `0` |
+| RTT | `last_rtt_ms` / `min_rtt_ms` / `max_rtt_ms` | ms | TCP `TCP_INFO` 或 KCP `rx_srtt` 的 RTT 样本 | UDP 当前没有 RTT 探测，因此为 `0` |
+
+### 实测样本
+
+以下值来自本仓库当前实现的本地回环测试，仅用于说明“字段已经能落到 JSONL 且当前名字是什么”，不是固定性能指标。
+
+| 样本 | 关键字段 | 实测值 |
+| --- | --- | --- |
+| `TCP` 点对点直传发送端 | `progress_pct` / `cwnd_last` / `last_rtt_ms` / `tcp_data_bytes_sent` | `100` / `10` / `1` / `287984` |
+| `UDP` Hub 中转接收端 | `progress_pct` / `udp_expected_chunks` / `udp_received_chunks` / `udp_lost_chunks` / `udp_recv_window_dist` | `100` / `3` / `3` / `0` / `0:3` |
+| `KCP` Bridge 桥接节点 | `cwnd_last` / `last_rtt_ms` / `kcp_data_bytes_sent` / `queue_avg_ms` | `2` / `7` / `265` / `33991.342083` |
+| `KCP` 最终接收端 | `progress_pct` / `end_to_end_avg_ms` / `cwnd_last` | `100` / `121.333333` / `2` |
+
+### 为什么有些字段“存在但没有值”
+
+| 情况 | 典型字段 | 说明 |
+| --- | --- | --- |
+| 协议不适用 | `tcp_*` 出现在 UDP/KCP；`kcp_*` 出现在 TCP/UDP；`cwnd_*` 出现在 UDP | 字段统一保留，便于同一份 JSONL 脚本处理；不适用的协议就写 `0` |
+| 还没实现 | `udp_retrans` | 现在没有应用层 UDP 重传，所以这个字段只是预留 |
+| 当前角色不产出 | `udp_expected_chunks`、`udp_loss_*`、`end_to_end_*` | UDP 丢包统计只在最终接收文件的 UDP peer 上有值；`end_to_end_*` 也主要只在最终接收端有值 |
+| 没有采样到 RTT | `last_rtt_ms`、`propagation_*` | UDP 当前没有 RTT 探针；TCP/KCP 只有在拿到对应协议样本后才有值 |
+| 时间分辨率太粗 | `send_call_*`、`proto_send_avg_ms`、`processing_*` | 当前很多路径按毫秒计时，本地回环下大量操作小于 `1ms`，所以会显示 `0` |
+| 当前没有任务 | `progress_*` | 只注册但没有文件传输时，这组字段自然是 `0` |

include/logger.h

@@ -17,6 +17,10 @@ typedef struct OmniMetricSummary {
     double sum;
 } OmniMetricSummary;
 
+#define OMNI_LOGGER_UDP_RANGES_SIZE 512u
+#define OMNI_LOGGER_UDP_SEQ_SAMPLE_SIZE 512u
+#define OMNI_LOGGER_UDP_WINDOW_DIST_SIZE 512u
+
 /* 通过该结构体收集全局统计信息 */
 typedef struct OmniStats {
     uint64_t start_ms;           /* 起始时间（毫秒） */
@@ -44,6 +48,15 @@ typedef struct OmniStats {
     uint64_t kcp_data_segs_out;  /* KCP 累计发送的数据分片数（含重传） */
     uint64_t kcp_data_bytes_sent; /* KCP 累计发送的数据字节（含重传） */
     uint64_t kcp_retrans_bytes;  /* KCP 累计重传的数据字节 */
+    uint64_t udp_expected_chunks; /* UDP 文件接收侧预期分片数 */
+    uint64_t udp_received_chunks; /* UDP 文件接收侧实际收到的分片数 */
+    uint64_t udp_lost_chunks;    /* UDP 文件接收侧推断丢失的分片数 */
+    uint64_t udp_loss_burst_count; /* UDP 丢包区间数量 */
+    uint64_t udp_loss_burst_max_len; /* UDP 最大连续丢包长度 */
+    double udp_loss_rate_pct;    /* UDP 丢包率（百分比） */
+    char udp_loss_ranges[OMNI_LOGGER_UDP_RANGES_SIZE]; /* UDP 丢包区间摘要 */
+    char udp_loss_seq_sample[OMNI_LOGGER_UDP_SEQ_SAMPLE_SIZE]; /* UDP 丢包序号样本 */
+    char udp_recv_window_dist[OMNI_LOGGER_UDP_WINDOW_DIST_SIZE]; /* UDP 接收窗口分布 */
 
     /* 延迟/耗时统计（单位：毫秒） */
     double send_call_avg_ms;     /* omni_send 平均耗时（EWMA） */
@@ -126,6 +139,15 @@ void logger_on_cwnd(double cwnd);
 /* 记录任务总量与当前进度。 */
 void logger_set_transfer_total(uint64_t total_bytes);
 void logger_set_progress(uint64_t progress_bytes);
+void logger_reset_transfer_observability(void);
+void logger_on_udp_loss_summary(uint64_t expected_chunks,
+                                uint64_t received_chunks,
+                                uint64_t lost_chunks,
+                                uint64_t burst_count,
+                                uint64_t burst_max_len,
+                                const char *ranges,
+                                const char *seq_sample,
+                                const char *recv_window_dist);
 
 /* 计算当前吞吐量（返回：字节/秒） */
 double logger_calculate_throughput(void);

src/apps/peer_main.c

@@ -42,6 +42,11 @@ typedef struct PeerRecvFileState {
     uint32_t total_chunks;
     uint64_t total_bytes;
     uint64_t bytes_written;
+    uint32_t received_chunks;
+    uint32_t observed_windows;
+    uint8_t *seq_seen;
+    uint32_t *recv_window_counts;
+    size_t recv_window_cap;
 } PeerRecvFileState;
 
 typedef struct PeerRuntime {
@@ -49,6 +54,7 @@ typedef struct PeerRuntime {
     PeerTransportSession *session;
     PeerMode mode;
     OmniRole role;
+    OmniProtocol proto;
     int running;
     int direct_register_sent;
     char client_id[OMNI_PEER_ID_SIZE];
@@ -175,11 +181,239 @@ static void peer_set_bound_peer(PeerRuntime *rt, const char *peer_id)
 
 static void close_recv_file(PeerRuntime *rt)
 {
-    if (!rt || !rt->rx_file.fp) {
+    if (!rt) {
         return;
     }
+    if (rt->rx_file.fp) {
         fclose(rt->rx_file.fp);
         rt->rx_file.fp = NULL;
+    }
+    free(rt->rx_file.seq_seen);
+    free(rt->rx_file.recv_window_counts);
+    memset(&rt->rx_file, 0, sizeof(rt->rx_file));
+}
+
+static int recv_file_prepare_tracking(PeerRecvFileState *state, uint32_t total_chunks)
+{
+    if (!state) {
+        return OMNI_ERR_PARAM;
+    }
+    if (total_chunks == 0) {
+        return OMNI_OK;
+    }
+    state->seq_seen = (uint8_t *)calloc((size_t)total_chunks + 1u, sizeof(uint8_t));
+    if (!state->seq_seen) {
+        return OMNI_ERR_GENERIC;
+    }
+    return OMNI_OK;
+}
+
+static int recv_file_ensure_window_capacity(PeerRecvFileState *state, uint32_t window_id)
+{
+    size_t need;
+    size_t new_cap;
+    uint32_t *new_counts;
+
+    if (!state) {
+        return OMNI_ERR_PARAM;
+    }
+    need = (size_t)window_id + 1u;
+    if (need <= state->recv_window_cap) {
+        return OMNI_OK;
+    }
+
+    new_cap = (state->recv_window_cap == 0) ? 4u : state->recv_window_cap;
+    while (new_cap < need) {
+        new_cap *= 2u;
+    }
+
+    new_counts = (uint32_t *)realloc(state->recv_window_counts,
+                                     new_cap * sizeof(uint32_t));
+    if (!new_counts) {
+        return OMNI_ERR_GENERIC;
+    }
+    memset(new_counts + state->recv_window_cap,
+           0,
+           (new_cap - state->recv_window_cap) * sizeof(uint32_t));
+    state->recv_window_counts = new_counts;
+    state->recv_window_cap = new_cap;
+    return OMNI_OK;
+}
+
+static void recv_file_note_chunk(PeerRecvFileState *state,
+                                 uint32_t seq,
+                                 uint32_t window_id)
+{
+    if (!state || seq == 0) {
+        return;
+    }
+
+    if (recv_file_ensure_window_capacity(state, window_id) == OMNI_OK) {
+        if ((size_t)window_id < state->recv_window_cap) {
+            state->recv_window_counts[window_id]++;
+        }
+        if (window_id + 1u > state->observed_windows) {
+            state->observed_windows = window_id + 1u;
+        }
+    }
+
+    if (!state->seq_seen) {
+        state->received_chunks++;
+        return;
+    }
+    if (state->seq_seen && seq <= state->total_chunks && !state->seq_seen[seq]) {
+        state->seq_seen[seq] = 1;
+        state->received_chunks++;
+    }
+}
+
+static void append_csv_uint(char *buf, size_t buf_sz, uint32_t value)
+{
+    size_t len;
+
+    if (!buf || buf_sz == 0) {
+        return;
+    }
+    len = strlen(buf);
+    if (len >= buf_sz - 1u) {
+        return;
+    }
+    snprintf(buf + len, buf_sz - len, "%s%u", len == 0 ? "" : ",", (unsigned)value);
+}
+
+static void append_range(char *buf, size_t buf_sz, uint32_t start, uint32_t end)
+{
+    size_t len;
+
+    if (!buf || buf_sz == 0) {
+        return;
+    }
+    len = strlen(buf);
+    if (len >= buf_sz - 1u) {
+        return;
+    }
+    if (start == end) {
+        snprintf(buf + len, buf_sz - len, "%s%u", len == 0 ? "" : ",", (unsigned)start);
+    } else {
+        snprintf(buf + len, buf_sz - len, "%s%u-%u",
+                 len == 0 ? "" : ",",
+                 (unsigned)start,
+                 (unsigned)end);
+    }
+}
+
+static void append_window_count(char *buf,
+                                size_t buf_sz,
+                                uint32_t window_id,
+                                uint32_t count)
+{
+    size_t len;
+
+    if (!buf || buf_sz == 0) {
+        return;
+    }
+    len = strlen(buf);
+    if (len >= buf_sz - 1u) {
+        return;
+    }
+    snprintf(buf + len, buf_sz - len, "%s%u:%u",
+             len == 0 ? "" : ",",
+             (unsigned)window_id,
+             (unsigned)count);
+}
+
+static void peer_record_udp_loss_summary(PeerRuntime *rt, uint32_t total_windows)
+{
+    PeerRecvFileState *state;
+    uint64_t lost_chunks;
+    uint64_t burst_count = 0;
+    uint64_t burst_max_len = 0;
+    char ranges[OMNI_LOGGER_UDP_RANGES_SIZE];
+    char seq_sample[OMNI_LOGGER_UDP_SEQ_SAMPLE_SIZE];
+    char recv_window_dist[OMNI_LOGGER_UDP_WINDOW_DIST_SIZE];
+    uint32_t sample_count = 0;
+
+    if (!rt || rt->proto != OMNI_PROTO_UDP) {
+        return;
+    }
+    state = &rt->rx_file;
+    if (state->total_chunks == 0 || !state->seq_seen) {
+        return;
+    }
+
+    ranges[0] = '\0';
+    seq_sample[0] = '\0';
+    recv_window_dist[0] = '\0';
+    lost_chunks = 0;
+
+    for (uint32_t seq = 1; seq <= state->total_chunks; ++seq) {
+        if (state->seq_seen[seq]) {
+            continue;
+        }
+
+        {
+            uint32_t end = seq;
+
+            while (end + 1u <= state->total_chunks && !state->seq_seen[end + 1u]) {
+                end++;
+            }
+            lost_chunks += (uint64_t)(end - seq + 1u);
+            burst_count++;
+            if ((uint64_t)(end - seq + 1u) > burst_max_len) {
+                burst_max_len = (uint64_t)(end - seq + 1u);
+            }
+            append_range(ranges, sizeof(ranges), seq, end);
+            for (uint32_t missing = seq; missing <= end && sample_count < 16u; ++missing) {
+                append_csv_uint(seq_sample, sizeof(seq_sample), missing);
+                sample_count++;
+            }
+            seq = end;
+        }
+    }
+
+    for (uint32_t window_id = 0;
+         window_id < (uint32_t)state->recv_window_cap &&
+         window_id < total_windows;
+         ++window_id) {
+        if (state->recv_window_counts[window_id] == 0) {
+            continue;
+        }
+        append_window_count(recv_window_dist,
+                            sizeof(recv_window_dist),
+                            window_id,
+                            state->recv_window_counts[window_id]);
+    }
+
+    logger_on_udp_loss_summary(state->total_chunks,
+                               state->received_chunks,
+                               lost_chunks,
+                               burst_count,
+                               burst_max_len,
+                               ranges,
+                               seq_sample,
+                               recv_window_dist);
+}
+
+static void peer_finalize_recv_observability(PeerRuntime *rt, uint32_t total_windows_hint)
+{
+    uint32_t total_windows;
+
+    if (!rt || rt->proto != OMNI_PROTO_UDP) {
+        return;
+    }
+    if (rt->rx_file.transfer_id == 0 || rt->rx_file.total_chunks == 0) {
+        return;
+    }
+
+    total_windows = total_windows_hint;
+    if (total_windows == 0) {
+        total_windows = rt->rx_file.observed_windows;
+    }
+    if (total_windows == 0) {
+        total_windows = 1;
+    }
+
+    peer_record_udp_loss_summary(rt, total_windows);
 }
 
 static uint64_t compute_file_size(FILE *fp)
@@ -389,6 +623,8 @@ static int peer_send_file(PeerRuntime *rt,
     uint32_t total_chunks;
     uint32_t transfer_id;
     uint32_t total_windows = 1;
+    uint64_t transfer_start_ms;
+    uint32_t max_window_id = 0;
     int rc = OMNI_OK;
 
     if (!file_path) {
@@ -419,10 +655,13 @@ static int peer_send_file(PeerRuntime *rt,
         goto out;
     }
 
+    logger_reset_transfer_observability();
     logger_set_transfer_total(total_bytes);
     logger_set_progress(0);
+    transfer_start_ms = omni_now_ms();
 
     for (uint32_t seq = 1; rt->running && !g_stop; ++seq) {
+        uint64_t processing_t0 = omni_now_ms();
         size_t nread = fread(chunk, 1, chunk_size, fp);
 
         if (nread == 0) {
@@ -437,18 +676,24 @@ static int peer_send_file(PeerRuntime *rt,
 
         if (nread > 0) {
             TransferChunkMeta meta;
+            uint64_t chunk_origin_ts_ms = omni_now_ms();
+            uint32_t window_id = (uint32_t)((chunk_origin_ts_ms - transfer_start_ms) / 1000u);
 
             omni_transfer_chunk_meta_encode(&meta,
                                             transfer_id,
                                             seq,
                                             total_chunks,
-                                            0,
+                                            window_id,
                                             total_bytes,
                                             offset,
                                             (uint32_t)nread,
-                                            omni_now_ms());
+                                            chunk_origin_ts_ms);
             memcpy(payload, &meta, TRANSFER_CHUNK_META_SIZE);
             memcpy(payload + TRANSFER_CHUNK_META_SIZE, chunk, nread);
+            if (window_id > max_window_id) {
+                max_window_id = window_id;
+            }
+            logger_on_processing_latency((double)(omni_now_ms() - processing_t0));
             rc = peer_send_inner(rt,
                                  effective_dst,
                                  MSG_TYPE_FILE_CHUNK,
@@ -465,6 +710,7 @@ static int peer_send_file(PeerRuntime *rt,
     if (rc == OMNI_OK && rt->running && !g_stop) {
         TransferEndMeta end_meta;
 
+        total_windows = max_window_id + 1u;
         omni_transfer_end_meta_encode(&end_meta,
                                       transfer_id,
                                       total_chunks,
@@ -587,6 +833,7 @@ static int peer_open_recv_file(PeerRuntime *rt, const TransferChunkMeta *meta)
     }
 
     if (!rt->rx_file.fp || rt->rx_file.transfer_id != meta->transfer_id) {
+        peer_finalize_recv_observability(rt, 0);
         close_recv_file(rt);
         rt->rx_file.fp = fopen(rt->output_path, "wb+");
         if (!rt->rx_file.fp) {
@@ -596,6 +843,13 @@ static int peer_open_recv_file(PeerRuntime *rt, const TransferChunkMeta *meta)
         rt->rx_file.total_chunks = meta->total_chunks;
         rt->rx_file.total_bytes = meta->total_bytes;
         rt->rx_file.bytes_written = 0;
+        if (recv_file_prepare_tracking(&rt->rx_file, meta->total_chunks) != OMNI_OK) {
+            logger_log("WARN", "peer",
+                       "recv_tracking_alloc_failed transfer_id=%u total_chunks=%u",
+                       (unsigned)meta->transfer_id,
+                       (unsigned)meta->total_chunks);
+        }
+        logger_reset_transfer_observability();
         logger_set_transfer_total(meta->total_bytes);
         logger_set_progress(0);
     }
@@ -608,6 +862,8 @@ static void handle_file_chunk_message(PeerRuntime *rt,
                                       uint32_t payload_len)
 {
     TransferChunkMeta meta;
+    uint64_t processing_t0 = omni_now_ms();
+    uint64_t now_ms;
 
     if (payload_len < TRANSFER_CHUNK_META_SIZE) {
         logger_log("WARN", "peer", "short_file_chunk_payload len=%u", (unsigned)payload_len);
@@ -632,6 +888,8 @@ static void handle_file_chunk_message(PeerRuntime *rt,
         return;
     }
 
+    recv_file_note_chunk(&rt->rx_file, meta.seq, meta.window_id);
+
     if (fseeko(rt->rx_file.fp, (off_t)meta.offset_bytes, SEEK_SET) != 0) {
         logger_log("ERROR", "peer", "fseeko_failed offset=%llu errno=%d",
                    (unsigned long long)meta.offset_bytes,
@@ -645,6 +903,11 @@ static void handle_file_chunk_message(PeerRuntime *rt,
 
     rt->rx_file.bytes_written += meta.chunk_bytes;
     logger_set_progress(rt->rx_file.bytes_written);
+    now_ms = omni_now_ms();
+    logger_on_processing_latency((double)(now_ms - processing_t0));
+    if (meta.origin_ts_ms > 0 && now_ms >= meta.origin_ts_ms) {
+        logger_on_end_to_end_latency((double)(now_ms - meta.origin_ts_ms));
+    }
 }
 
 static void handle_file_end_message(PeerRuntime *rt,
@@ -669,6 +932,9 @@ static void handle_file_end_message(PeerRuntime *rt,
         total_chunks = rt->rx_file.total_chunks ? rt->rx_file.total_chunks : end_meta.total_chunks;
         total_bytes = rt->rx_file.total_bytes ? rt->rx_file.total_bytes : end_meta.total_bytes;
     }
+    peer_finalize_recv_observability(rt,
+                                     end_meta.total_windows > 0 ? end_meta.total_windows
+                                                                : rt->rx_file.observed_windows);
 
     fprintf(stdout,
             "[file %s -> %s] transfer_id=%u bytes_written=%llu total_bytes=%llu total_chunks=%u output=%s\n",
@@ -955,6 +1221,7 @@ static void handle_transport_event(PeerRuntime *rt,
             rt->session = NULL;
             rt->direct_register_sent = 0;
             peer_set_bound_peer(rt, "");
+            peer_finalize_recv_observability(rt, 0);
             close_recv_file(rt);
             rt->rx_file.transfer_id = 0;
             rt->rx_file.total_chunks = 0;
@@ -1114,6 +1381,7 @@ int main(int argc, char **argv)
     memset(&rt, 0, sizeof(rt));
     memset(&startup, 0, sizeof(startup));
     rt.mode = mode;
+    rt.proto = proto;
     rt.role = (mode == PEER_MODE_DIRECT && listen_port > 0) ? OMNI_ROLE_SERVER : OMNI_ROLE_CLIENT;
     log_mode = (mode == PEER_MODE_DIRECT) ? "direct" : "hub";
     log_role = (rt.role == OMNI_ROLE_SERVER) ? "server" : "client";
@@ -1220,6 +1488,7 @@ int main(int argc, char **argv)
         }
     }
 
+    peer_finalize_recv_observability(&rt, 0);
     close_recv_file(&rt);
     peer_transport_close(rt.transport);
     logger_print_performance_log("final");

src/core/logger.c

@@ -29,6 +29,25 @@ static char g_ctx_mode[16];
 static char g_ctx_role[16];
 static char g_ctx_self_id[OMNI_PEER_ID_SIZE];
 
+static void metric_reset(OmniMetricSummary *metric);
+
+static void reset_transfer_observability_locked(void)
+{
+    g_stats.total_work_bytes = 0;
+    g_stats.progress_bytes = 0;
+    metric_reset(&g_stats.processing_delay_ms);
+    metric_reset(&g_stats.end_to_end_delay_ms);
+    g_stats.udp_expected_chunks = 0;
+    g_stats.udp_received_chunks = 0;
+    g_stats.udp_lost_chunks = 0;
+    g_stats.udp_loss_burst_count = 0;
+    g_stats.udp_loss_burst_max_len = 0;
+    g_stats.udp_loss_rate_pct = 0.0;
+    g_stats.udp_loss_ranges[0] = '\0';
+    g_stats.udp_loss_seq_sample[0] = '\0';
+    g_stats.udp_recv_window_dist[0] = '\0';
+}
+
 /* 对 common.h 的时间接口做一层薄包装，统一 logger 模块内部的调用入口。 */
 static uint64_t now_ms(void)
 {
@@ -240,6 +259,7 @@ void logger_init(void)
         metric_reset(&g_stats.send_buffer_pct);
         metric_reset(&g_stats.recv_buffer_pct);
         metric_reset(&g_stats.cwnd);
+        reset_transfer_observability_locked();
         copy_context_field(g_ctx_app, sizeof(g_ctx_app), NULL, "unknown");
         copy_context_field(g_ctx_proto, sizeof(g_ctx_proto), NULL, "unknown");
         copy_context_field(g_ctx_mode, sizeof(g_ctx_mode), NULL, "unknown");
@@ -555,6 +575,43 @@ void logger_set_progress(uint64_t progress_bytes)
     pthread_mutex_unlock(&g_mu);
 }
 
+void logger_reset_transfer_observability(void)
+{
+    pthread_mutex_lock(&g_mu);
+    reset_transfer_observability_locked();
+    pthread_mutex_unlock(&g_mu);
+}
+
+void logger_on_udp_loss_summary(uint64_t expected_chunks,
+                                uint64_t received_chunks,
+                                uint64_t lost_chunks,
+                                uint64_t burst_count,
+                                uint64_t burst_max_len,
+                                const char *ranges,
+                                const char *seq_sample,
+                                const char *recv_window_dist)
+{
+    pthread_mutex_lock(&g_mu);
+    g_stats.udp_expected_chunks = expected_chunks;
+    g_stats.udp_received_chunks = received_chunks;
+    g_stats.udp_lost_chunks = lost_chunks;
+    g_stats.udp_loss_burst_count = burst_count;
+    g_stats.udp_loss_burst_max_len = burst_max_len;
+    g_stats.udp_loss_rate_pct = (expected_chunks > 0)
+        ? ((double)lost_chunks * 100.0) / (double)expected_chunks
+        : 0.0;
+    omni_copy_fixed_ascii(g_stats.udp_loss_ranges,
+                          sizeof(g_stats.udp_loss_ranges),
+                          ranges ? ranges : "");
+    omni_copy_fixed_ascii(g_stats.udp_loss_seq_sample,
+                          sizeof(g_stats.udp_loss_seq_sample),
+                          seq_sample ? seq_sample : "");
+    omni_copy_fixed_ascii(g_stats.udp_recv_window_dist,
+                          sizeof(g_stats.udp_recv_window_dist),
+                          recv_window_dist ? recv_window_dist : "");
+    pthread_mutex_unlock(&g_mu);
+}
+
 /* 对外提供一个“当前总吞吐”的便捷计算接口。 */
 double logger_calculate_throughput(void)
 {
@@ -606,6 +663,9 @@ void logger_print_performance_log(const char *tag)
     char ctx_mode[sizeof(g_ctx_mode)];
     char ctx_role[sizeof(g_ctx_role)];
     char ctx_self_id[sizeof(g_ctx_self_id)];
+    char udp_loss_ranges[sizeof(snapshot.udp_loss_ranges)];
+    char udp_loss_seq_sample[sizeof(snapshot.udp_loss_seq_sample)];
+    char udp_recv_window_dist[sizeof(snapshot.udp_recv_window_dist)];
 
     pthread_mutex_lock(&g_mu);
     now = now_ms();
@@ -626,8 +686,15 @@ void logger_print_performance_log(const char *tag)
     omni_copy_fixed_ascii(ctx_mode, sizeof(ctx_mode), g_ctx_mode);
     omni_copy_fixed_ascii(ctx_role, sizeof(ctx_role), g_ctx_role);
     omni_copy_fixed_ascii(ctx_self_id, sizeof(ctx_self_id), g_ctx_self_id);
+    omni_copy_fixed_ascii(udp_loss_ranges, sizeof(udp_loss_ranges), snapshot.udp_loss_ranges);
+    omni_copy_fixed_ascii(udp_loss_seq_sample, sizeof(udp_loss_seq_sample), snapshot.udp_loss_seq_sample);
+    omni_copy_fixed_ascii(udp_recv_window_dist, sizeof(udp_recv_window_dist), snapshot.udp_recv_window_dist);
     pthread_mutex_unlock(&g_mu);
 
+    udp_loss_ranges[sizeof(udp_loss_ranges) - 1u] = '\0';
+    udp_loss_seq_sample[sizeof(udp_loss_seq_sample) - 1u] = '\0';
+    udp_recv_window_dist[sizeof(udp_recv_window_dist) - 1u] = '\0';
+
     progress_pct = 0.0;
     if (snapshot.total_work_bytes > 0) {
         progress_pct = ((double)snapshot.progress_bytes * 100.0) /
@@ -651,7 +718,7 @@ void logger_print_performance_log(const char *tag)
                 "processing_avg_ms=%.3f queue_avg_ms=%.3f transmission_avg_ms=%.3f propagation_avg_ms=%.3f end_to_end_avg_ms=%.3f "
                 "send_buffer_pct=%.2f recv_buffer_pct=%.2f cwnd=%.2f "
                 "last_rtt_ms=%llu min_rtt_ms=%llu max_rtt_ms=%llu "
-                "tcp_retrans=%llu tcp_data_segs_out=%llu tcp_data_bytes_sent=%llu tcp_retrans_bytes=%llu "
+                "tcp_retrans=%llu udp_retrans=%llu tcp_data_segs_out=%llu tcp_data_bytes_sent=%llu tcp_retrans_bytes=%llu "
                 "kcp_retrans=%llu kcp_data_segs_out=%llu kcp_data_bytes_sent=%llu kcp_retrans_bytes=%llu\n",
                 ctx_app[0] ? ctx_app : "unknown",
                 ctx_proto[0] ? ctx_proto : "unknown",
@@ -693,6 +760,7 @@ void logger_print_performance_log(const char *tag)
                 (unsigned long long)((snapshot.min_rtt_ms == UINT64_MAX) ? 0 : snapshot.min_rtt_ms),
                 (unsigned long long)snapshot.max_rtt_ms,
                 (unsigned long long)snapshot.tcp_retrans,
+                (unsigned long long)snapshot.udp_retrans,
                 (unsigned long long)snapshot.tcp_data_segs_out,
                 (unsigned long long)snapshot.tcp_data_bytes_sent,
                 (unsigned long long)snapshot.tcp_retrans_bytes,
@@ -700,10 +768,39 @@ void logger_print_performance_log(const char *tag)
                 (unsigned long long)snapshot.kcp_data_segs_out,
                 (unsigned long long)snapshot.kcp_data_bytes_sent,
                 (unsigned long long)snapshot.kcp_retrans_bytes);
+        if (snapshot.udp_expected_chunks > 0 || snapshot.udp_lost_chunks > 0) {
+            fprintf(fp,
+                    "ts=%llu level=INFO component=perf_udp_loss app=%s proto=%s mode=%s role=%s self_id=%s "
+                    "udp_expected_chunks=%llu udp_received_chunks=%llu udp_lost_chunks=%llu "
+                    "udp_loss_rate_pct=%.2f udp_loss_burst_count=%llu udp_loss_burst_max_len=%llu "
+                    "udp_loss_ranges=%s udp_loss_seq_sample=%s udp_recv_window_dist=%s\n",
+                    (unsigned long long)now,
+                    ctx_app[0] ? ctx_app : "unknown",
+                    ctx_proto[0] ? ctx_proto : "unknown",
+                    ctx_mode[0] ? ctx_mode : "unknown",
+                    ctx_role[0] ? ctx_role : "unknown",
+                    ctx_self_id[0] ? ctx_self_id : "-",
+                    (unsigned long long)snapshot.udp_expected_chunks,
+                    (unsigned long long)snapshot.udp_received_chunks,
+                    (unsigned long long)snapshot.udp_lost_chunks,
+                    snapshot.udp_loss_rate_pct,
+                    (unsigned long long)snapshot.udp_loss_burst_count,
+                    (unsigned long long)snapshot.udp_loss_burst_max_len,
+                    udp_loss_ranges[0] ? udp_loss_ranges : "-",
+                    udp_loss_seq_sample[0] ? udp_loss_seq_sample : "-",
+                    udp_recv_window_dist[0] ? udp_recv_window_dist : "-");
+        }
     }
 
     if (g_json_fp) {
         /* JSONL 输出便于后续脚本或可视化工具离线分析。 */
+        char esc_udp_ranges[sizeof(snapshot.udp_loss_ranges) * 2u];
+        char esc_udp_seq_sample[sizeof(snapshot.udp_loss_seq_sample) * 2u];
+        char esc_udp_window_dist[sizeof(snapshot.udp_recv_window_dist) * 2u];
+
+        json_escape(udp_loss_ranges, esc_udp_ranges, sizeof(esc_udp_ranges));
+        json_escape(udp_loss_seq_sample, esc_udp_seq_sample, sizeof(esc_udp_seq_sample));
+        json_escape(udp_recv_window_dist, esc_udp_window_dist, sizeof(esc_udp_window_dist));
         fprintf(g_json_fp,
                 "{\"ts_ms\":%llu,"
                 "\"level\":\"INFO\","
@@ -764,13 +861,23 @@ void logger_print_performance_log(const char *tag)
                 "\"min_rtt_ms\":%llu,"
                 "\"max_rtt_ms\":%llu,"
                 "\"tcp_retrans\":%llu,"
+                "\"udp_retrans\":%llu,"
                 "\"tcp_data_segs_out\":%llu,"
                 "\"tcp_data_bytes_sent\":%llu,"
                 "\"tcp_retrans_bytes\":%llu,"
                 "\"kcp_retrans\":%llu,"
                 "\"kcp_data_segs_out\":%llu,"
                 "\"kcp_data_bytes_sent\":%llu,"
-                "\"kcp_retrans_bytes\":%llu}\n",
+                "\"kcp_retrans_bytes\":%llu,"
+                "\"udp_expected_chunks\":%llu,"
+                "\"udp_received_chunks\":%llu,"
+                "\"udp_lost_chunks\":%llu,"
+                "\"udp_loss_rate_pct\":%.6f,"
+                "\"udp_loss_burst_count\":%llu,"
+                "\"udp_loss_burst_max_len\":%llu,"
+                "\"udp_loss_ranges\":\"%s\","
+                "\"udp_loss_seq_sample\":\"%s\","
+                "\"udp_recv_window_dist\":\"%s\"}\n",
                 (unsigned long long)now,
                 ctx_app[0] ? ctx_app : "unknown",
                 ctx_proto[0] ? ctx_proto : "unknown",
@@ -828,13 +935,23 @@ void logger_print_performance_log(const char *tag)
                 (unsigned long long)((snapshot.min_rtt_ms == UINT64_MAX) ? 0 : snapshot.min_rtt_ms),
                 (unsigned long long)snapshot.max_rtt_ms,
                 (unsigned long long)snapshot.tcp_retrans,
+                (unsigned long long)snapshot.udp_retrans,
                 (unsigned long long)snapshot.tcp_data_segs_out,
                 (unsigned long long)snapshot.tcp_data_bytes_sent,
                 (unsigned long long)snapshot.tcp_retrans_bytes,
                 (unsigned long long)snapshot.kcp_retrans,
                 (unsigned long long)snapshot.kcp_data_segs_out,
                 (unsigned long long)snapshot.kcp_data_bytes_sent,
-                (unsigned long long)snapshot.kcp_retrans_bytes);
+                (unsigned long long)snapshot.kcp_retrans_bytes,
+                (unsigned long long)snapshot.udp_expected_chunks,
+                (unsigned long long)snapshot.udp_received_chunks,
+                (unsigned long long)snapshot.udp_lost_chunks,
+                snapshot.udp_loss_rate_pct,
+                (unsigned long long)snapshot.udp_loss_burst_count,
+                (unsigned long long)snapshot.udp_loss_burst_max_len,
+                esc_udp_ranges,
+                esc_udp_seq_sample,
+                esc_udp_window_dist);
         fflush(g_json_fp);
     }
 }

src/core/peer_transport.c

@@ -7,9 +7,11 @@
 #include <errno.h>
 #include <netinet/in.h>
 #include <netinet/tcp.h>
+#include <stddef.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/ioctl.h>
 #include <sys/select.h>
 #include <sys/socket.h>
 #include <sys/types.h>
@@ -29,6 +31,8 @@ struct PeerTransportSession {
     socklen_t addr_len;
     ikcpcb *kcp;
     uint32_t kcp_conv;
+    uint32_t *kcp_seg_xmit_seen;
+    size_t kcp_seg_xmit_cap;
     char remote_ip[64];
     uint16_t remote_port;
     struct PeerTransportSession *next;
@@ -44,6 +48,66 @@ struct PeerTransport {
     uint8_t *rx_frame;
 };
 
+#ifdef __linux__
+struct OmniLinuxTcpInfo {
+    uint8_t tcpi_state;
+    uint8_t tcpi_ca_state;
+    uint8_t tcpi_retransmits;
+    uint8_t tcpi_probes;
+    uint8_t tcpi_backoff;
+    uint8_t tcpi_options;
+    uint8_t tcpi_snd_wscale : 4, tcpi_rcv_wscale : 4;
+    uint8_t tcpi_delivery_rate_app_limited : 1, tcpi_fastopen_client_fail : 2;
+    uint32_t tcpi_rto;
+    uint32_t tcpi_ato;
+    uint32_t tcpi_snd_mss;
+    uint32_t tcpi_rcv_mss;
+    uint32_t tcpi_unacked;
+    uint32_t tcpi_sacked;
+    uint32_t tcpi_lost;
+    uint32_t tcpi_retrans;
+    uint32_t tcpi_fackets;
+    uint32_t tcpi_last_data_sent;
+    uint32_t tcpi_last_ack_sent;
+    uint32_t tcpi_last_data_recv;
+    uint32_t tcpi_last_ack_recv;
+    uint32_t tcpi_pmtu;
+    uint32_t tcpi_rcv_ssthresh;
+    uint32_t tcpi_rtt;
+    uint32_t tcpi_rttvar;
+    uint32_t tcpi_snd_ssthresh;
+    uint32_t tcpi_snd_cwnd;
+    uint32_t tcpi_advmss;
+    uint32_t tcpi_reordering;
+    uint32_t tcpi_rcv_rtt;
+    uint32_t tcpi_rcv_space;
+    uint32_t tcpi_total_retrans;
+    uint64_t tcpi_pacing_rate;
+    uint64_t tcpi_max_pacing_rate;
+    uint64_t tcpi_bytes_acked;
+    uint64_t tcpi_bytes_received;
+    uint32_t tcpi_segs_out;
+    uint32_t tcpi_segs_in;
+    uint32_t tcpi_notsent_bytes;
+    uint32_t tcpi_min_rtt;
+    uint32_t tcpi_data_segs_in;
+    uint32_t tcpi_data_segs_out;
+    uint64_t tcpi_delivery_rate;
+    uint64_t tcpi_busy_time;
+    uint64_t tcpi_rwnd_limited;
+    uint64_t tcpi_sndbuf_limited;
+    uint32_t tcpi_delivered;
+    uint32_t tcpi_delivered_ce;
+    uint64_t tcpi_bytes_sent;
+    uint64_t tcpi_bytes_retrans;
+};
+
+static int tcp_info_has_field(socklen_t len, size_t field_end)
+{
+    return (size_t)len >= field_end;
+}
+#endif
+
 static void peer_transport_note_send(size_t bytes)
 {
     logger_on_send(bytes);
@@ -58,6 +122,203 @@ static void peer_transport_note_recv(size_t bytes)
     logger_maybe_print_performance_log("peer_transport_recv");
 }
 
+static void sample_socket_buffers(int fd)
+{
+    int sndbuf = 0;
+    int rcvbuf = 0;
+    socklen_t optlen = sizeof(int);
+    int outq = 0;
+    int inq = 0;
+    double send_pct = 0.0;
+    double recv_pct = 0.0;
+
+    if (fd < 0) {
+        return;
+    }
+
+    if (getsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, &optlen) == 0 && sndbuf > 0) {
+#ifdef TIOCOUTQ
+        if (ioctl(fd, TIOCOUTQ, &outq) == 0 && outq >= 0) {
+            send_pct = ((double)outq * 100.0) / (double)sndbuf;
+            logger_on_send_queue_bytes((size_t)outq);
+        }
+#endif
+    }
+
+    optlen = sizeof(int);
+    if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &optlen) == 0 && rcvbuf > 0) {
+        if (ioctl(fd, FIONREAD, &inq) == 0 && inq >= 0) {
+            recv_pct = ((double)inq * 100.0) / (double)rcvbuf;
+            logger_on_recv_queue_bytes((size_t)inq);
+        }
+    }
+
+    logger_on_buffer_status(send_pct, recv_pct);
+}
+
+static void sample_tcp_info(int fd)
+{
+#ifdef __linux__
+    struct OmniLinuxTcpInfo ti;
+    uint64_t total_retrans = 0;
+    uint64_t data_segs_out = 0;
+    uint64_t bytes_sent = 0;
+    uint64_t bytes_retrans = 0;
+    socklen_t len = sizeof(ti);
+
+    if (fd < 0) {
+        return;
+    }
+    memset(&ti, 0, sizeof(ti));
+    if (getsockopt(fd, IPPROTO_TCP, TCP_INFO, &ti, &len) != 0) {
+        return;
+    }
+
+    if (tcp_info_has_field(len,
+                           offsetof(struct OmniLinuxTcpInfo, tcpi_total_retrans) +
+                           sizeof(ti.tcpi_total_retrans))) {
+        total_retrans = (uint64_t)ti.tcpi_total_retrans;
+    } else {
+        total_retrans = (uint64_t)ti.tcpi_retrans;
+    }
+    if (tcp_info_has_field(len,
+                           offsetof(struct OmniLinuxTcpInfo, tcpi_data_segs_out) +
+                           sizeof(ti.tcpi_data_segs_out))) {
+        data_segs_out = (uint64_t)ti.tcpi_data_segs_out;
+    }
+    if (tcp_info_has_field(len,
+                           offsetof(struct OmniLinuxTcpInfo, tcpi_bytes_sent) +
+                           sizeof(ti.tcpi_bytes_sent))) {
+        bytes_sent = (uint64_t)ti.tcpi_bytes_sent;
+    }
+    if (tcp_info_has_field(len,
+                           offsetof(struct OmniLinuxTcpInfo, tcpi_bytes_retrans) +
+                           sizeof(ti.tcpi_bytes_retrans))) {
+        bytes_retrans = (uint64_t)ti.tcpi_bytes_retrans;
+    }
+
+    logger_on_rtt((uint64_t)(ti.tcpi_rtt / 1000u));
+    logger_on_tcp_transport(total_retrans, data_segs_out, bytes_sent, bytes_retrans);
+    logger_on_cwnd((double)ti.tcpi_snd_cwnd);
+    sample_socket_buffers(fd);
+#else
+    (void)fd;
+#endif
+}
+
+static int kcp_ensure_seg_track_capacity(PeerTransportSession *session, uint32_t sn)
+{
+    size_t need;
+    size_t new_cap;
+    uint32_t *new_seen;
+
+    if (!session) {
+        return OMNI_ERR_PARAM;
+    }
+    need = (size_t)sn + 1u;
+    if (need <= session->kcp_seg_xmit_cap) {
+        return OMNI_OK;
+    }
+
+    new_cap = (session->kcp_seg_xmit_cap == 0) ? 64u : session->kcp_seg_xmit_cap;
+    while (new_cap < need) {
+        new_cap *= 2u;
+    }
+
+    new_seen = (uint32_t *)realloc(session->kcp_seg_xmit_seen,
+                                   new_cap * sizeof(uint32_t));
+    if (!new_seen) {
+        return OMNI_ERR_GENERIC;
+    }
+    memset(new_seen + session->kcp_seg_xmit_cap,
+           0,
+           (new_cap - session->kcp_seg_xmit_cap) * sizeof(uint32_t));
+    session->kcp_seg_xmit_seen = new_seen;
+    session->kcp_seg_xmit_cap = new_cap;
+    return OMNI_OK;
+}
+
+static void sample_kcp_session(PeerTransportSession *session)
+{
+    struct IQUEUEHEAD *p;
+    double send_pct = 0.0;
+    double recv_pct = 0.0;
+
+    if (!session || !session->kcp) {
+        return;
+    }
+
+    for (p = session->kcp->snd_buf.next; p != &session->kcp->snd_buf; p = p->next) {
+        struct IKCPSEG *segment = iqueue_entry(p, struct IKCPSEG, node);
+        uint32_t prev_xmit;
+
+        if (!segment || segment->len == 0) {
+            continue;
+        }
+        if (kcp_ensure_seg_track_capacity(session, segment->sn) != OMNI_OK) {
+            return;
+        }
+
+        prev_xmit = session->kcp_seg_xmit_seen[segment->sn];
+        if (prev_xmit == 0 && segment->xmit > 0) {
+            logger_on_kcp_tx(1u, (uint64_t)segment->len);
+            prev_xmit = 1u;
+        }
+        if (segment->xmit > prev_xmit) {
+            uint32_t delta = segment->xmit - prev_xmit;
+            logger_on_kcp_retrans((uint64_t)delta,
+                                  (uint64_t)delta * (uint64_t)segment->len);
+        }
+        session->kcp_seg_xmit_seen[segment->sn] = segment->xmit;
+    }
+
+    if (session->kcp->rx_srtt > 0) {
+        logger_on_rtt((uint64_t)session->kcp->rx_srtt);
+    }
+    logger_on_cwnd((double)session->kcp->cwnd);
+
+    if (session->kcp->snd_wnd > 0) {
+        send_pct = ((double)ikcp_waitsnd(session->kcp) * 100.0) /
+                   (double)session->kcp->snd_wnd;
+        logger_on_send_queue_bytes((size_t)ikcp_waitsnd(session->kcp) *
+                                   (size_t)session->kcp->mss);
+    }
+    if (session->kcp->rcv_wnd > 0) {
+        recv_pct = ((double)session->kcp->nrcv_que * 100.0) /
+                   (double)session->kcp->rcv_wnd;
+        logger_on_recv_queue_bytes((size_t)session->kcp->nrcv_que *
+                                   (size_t)session->kcp->mss);
+    }
+    logger_on_buffer_status(send_pct, recv_pct);
+}
+
+static void peer_transport_sample_after_send(PeerTransport *transport,
+                                             PeerTransportSession *session)
+{
+    if (!transport || !session) {
+        return;
+    }
+    switch (transport->proto) {
+    case OMNI_PROTO_TCP:
+        sample_tcp_info(session->fd);
+        break;
+    case OMNI_PROTO_UDP:
+        sample_socket_buffers(transport->fd);
+        break;
+    case OMNI_PROTO_KCP:
+        sample_kcp_session(session);
+        break;
+    default:
+        break;
+    }
+}
+
+static void peer_transport_sample_after_recv(PeerTransport *transport,
+                                             PeerTransportSession *session)
+{
+    peer_transport_sample_after_send(transport, session);
+}
+
 const char *peer_transport_proto_name(OmniProtocol proto)
 {
     switch (proto) {
@@ -183,6 +444,7 @@ static void session_free(PeerTransportSession *session)
     if (session->kcp) {
         ikcp_release(session->kcp);
     }
+    free(session->kcp_seg_xmit_seen);
     free(session);
 }
 
@@ -845,6 +1107,10 @@ int peer_transport_send(PeerTransport *transport,
     ssize_t n = -1;
     int rc = OMNI_OK;
     int raw_rc = 0;
+    uint64_t call_t0;
+    uint64_t proto_t0;
+    uint64_t proto_t1;
+    uint64_t call_t1;
 
     if (!transport) {
         return OMNI_ERR_PARAM;
@@ -856,11 +1122,13 @@ int peer_transport_send(PeerTransport *transport,
         return OMNI_ERR_IO;
     }
 
+    call_t0 = omni_now_ms();
     frame = alloc_frame(type, payload, payload_len, &frame_len);
     if (!frame) {
         return OMNI_ERR_GENERIC;
     }
 
+    proto_t0 = omni_now_ms();
     switch (transport->proto) {
     case OMNI_PROTO_TCP:
         n = write_n(session->fd, frame, frame_len);
@@ -891,9 +1159,14 @@ int peer_transport_send(PeerTransport *transport,
         rc = OMNI_ERR_PARAM;
         break;
     }
+    proto_t1 = omni_now_ms();
+    call_t1 = proto_t1;
+    logger_on_proto_send_latency(proto_t1 - proto_t0);
+    logger_on_send_call_latency(call_t1 - call_t0);
 
     if (rc == OMNI_OK) {
         peer_transport_note_send(frame_len);
+        peer_transport_sample_after_send(transport, session);
     } else {
         logger_log("ERROR", "peer_transport",
                    "send_failed proto=%s type=%u remote=%s:%u raw_rc=%d errno=%d",
@@ -1088,21 +1361,39 @@ int peer_transport_next_event(PeerTransport *transport,
                               size_t payload_cap,
                               int timeout_ms)
 {
+    int rc;
+    uint64_t t0;
+    uint64_t t1;
+
     if (!transport || !event || !payload_buf) {
         return OMNI_ERR_PARAM;
     }
 
     event_reset(event);
+    t0 = omni_now_ms();
     switch (transport->proto) {
     case OMNI_PROTO_TCP:
-        return tcp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        rc = tcp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        break;
     case OMNI_PROTO_UDP:
-        return udp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        rc = udp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        break;
     case OMNI_PROTO_KCP:
-        return kcp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        rc = kcp_next_event(transport, event, payload_buf, payload_cap, timeout_ms);
+        break;
     default:
-        return OMNI_ERR_PARAM;
+        rc = OMNI_ERR_PARAM;
+        break;
     }
+    t1 = omni_now_ms();
+    if (rc > 0) {
+        logger_on_proto_recv_latency(t1 - t0);
+        logger_on_recv_call_latency(t1 - t0);
+        if (event->session) {
+            peer_transport_sample_after_recv(transport, event->session);
+        }
+    }
+    return rc;
 }
 
 void peer_transport_close_session(PeerTransport *transport,