RGW性能监控及源码实现

1性能计数器实现

性能计数器启动过程

#src/rgw/rgw_main.cc
int main(int argc, const char **argv)
    ...
  if (g_conf->daemonize) {
    global_init_daemonize(g_ceph_context);
  }
  Mutex mutex("main");
  SafeTimer init_timer(g_ceph_context, mutex);
  init_timer.init();
  mutex.Lock();
  init_timer.add_event_after(g_conf->rgw_init_timeout, new C_InitTimeout);
  mutex.Unlock();

  // Enable the perf counter before starting the service thread
  g_ceph_context->enable_perf_counter();
    ...
  r = rgw_perf_start(g_ceph_context); #启用rgw计数器

性能计数器参数说明

#src/rgw/rgw_common.cc
int rgw_perf_start(CephContext *cct)
{
  PerfCountersBuilder plb(cct, cct->_conf->name.to_str(), l_rgw_first, l_rgw_last);

  plb.add_u64_counter(l_rgw_req, "req", "Requests"); #处理成功请求数量
  plb.add_u64_counter(l_rgw_failed_req, "failed_req", "Aborted requests"); #处理失败请求数量

  plb.add_u64_counter(l_rgw_get, "get", "Gets"); #GET请求数量
  plb.add_u64_counter(l_rgw_get_b, "get_b", "Size of gets");
  plb.add_time_avg(l_rgw_get_lat, "get_initial_lat", "Get latency");
  plb.add_u64_counter(l_rgw_put, "put", "Puts"); #PUT请求数量
  plb.add_u64_counter(l_rgw_put_b, "put_b", "Size of puts");
  plb.add_time_avg(l_rgw_put_lat, "put_initial_lat", "Put latency");

  plb.add_u64(l_rgw_qlen, "qlen", "Queue length");
  plb.add_u64(l_rgw_qactive, "qactive", "Active requests queue");

  plb.add_u64_counter(l_rgw_cache_hit, "cache_hit", "Cache hits"); #用于缓存RGW元数据的缓存命中次数
  plb.add_u64_counter(l_rgw_cache_miss, "cache_miss", "Cache miss"); #未命中次数

  plb.add_u64_counter(l_rgw_keystone_token_cache_hit, "keystone_token_cache_hit", "Keystone token cache hits");
  plb.add_u64_counter(l_rgw_keystone_token_cache_miss, "keystone_token_cache_miss", "Keystone token cache miss");

  perfcounter = plb.create_perf_counters();
  cct->get_perfcounters_collection()->add(perfcounter);
  return 0;
}

性能计数器类型定义如下

#src/rgw/rgw_common.cc
enum {
  l_rgw_first = 15000,
  l_rgw_req,
  l_rgw_failed_req,

  l_rgw_get,
  l_rgw_get_b,
  l_rgw_get_lat,

  l_rgw_put,
  l_rgw_put_b,
  l_rgw_put_lat,

  l_rgw_qlen,
  l_rgw_qactive,

  l_rgw_cache_hit,
  l_rgw_cache_miss,

  l_rgw_keystone_token_cache_hit,
  l_rgw_keystone_token_cache_miss,

  l_rgw_last,
};

2如何使用计数器

方式1

root@demo# ceph --admin-daemon /var/run/ceph-client.radosgw.en-zone1.asok perf dump
{
"cct": {
   "total_workers": 32,
   "unhealthy_workers": 0
},
"client.radosgw.en-zone1": {
   "req": 2,
   "failed_req": 2,
   "get": 0,
   "get_b": 0,
   "get_initial_lat": {
       "avgcount": 0,
       "sum": 0.000000000
   },
   "put": 0,
   "put_b": 0,
   "put_initial_lat": {
       "avgcount": 0,
       "sum": 0.000000000
   },
   "qlen": 0,
   "qactive": 0,
   "cache_hit": 0,
   "cache_miss": 2,
   "keystone_token_cache_hit": 0,
   "keystone_token_cache_miss": 0
}

方式2 (使用Dumpling以上版本）

root@demo# ceph daemon client.radosgw.en-zone1 perf dump
{
   "cct": {
       "total_workers": 32,
       "unhealthy_workers": 0
   },
   "client.radosgw.en-zone1": {
       "req": 2,
       "failed_req": 2,
       "get": 0,
       "get_b": 0,
       "get_initial_lat": {
           "avgcount": 0,
           "sum": 0.000000000
       },
       "put": 0,
       "put_b": 0,
       "put_initial_lat": {
           "avgcount": 0,
           "sum": 0.000000000
       },
       "qlen": 0,
       "qactive": 0,
       "cache_hit": 0,
       "cache_miss": 2,
       "keystone_token_cache_hit": 0,
       "keystone_token_cache_miss": 0
   },

3自定义计数器源码实例

1. common/perf_counters: add average time for PERFCOUNTER_TIME https://github.com/ceph/ceph/pull/15478/files
2. mds: add perf counters for file system operations https://github.com/ceph/ceph/pull/14938/files
3. Added new counters for monitoring http status https://github.com/ceph/ceph/pull/10630/files

4总结

性能计数器一般人可能不太关注，但是将计数器数据推送到一些监控系统里面，同时添加一些告警策略，能够显著提升运维质量。同时通过源码层面去扩展计数器类型，从ceph内部去实现一些自定义数据的统计，也是一个不错的功能亮点。