服务器架构设计1
说到服务器,通常可以想象都是一个while(1)无限循环,当某种条件触发之后,则跳出循环。
然而,有些时候,则需要服务器针对某些ses,周期性的执行特定操作,如何实现呢。
一个简单的方法就是,建立session的node list,节点信息包括(time,ses私有变量,回调函数,执行周期interval),主线程周期性的访问这个node list,如果时间条件满足,执行回调函数。
构造了一个简单node list,便于说明。
inline uint64_t get_current_clock(void) { struct timespec tp; clock_gettime(CLOCK_MONOTONIC, &tp); return (uint64_t)tp.tv_sec * 1e3 + (uint64_t)(tp.tv_nsec / 1e6); } static void heap_insert_helper(timer_event_t* heap[], uint32_t idx) { if(idx <= 0) { return; } uint32_t p = (idx - 1) >> 1; if(heap[p]->time <= heap[idx]->time) { return; } else { timer_event_t* tmp = heap[p]; heap[p] = heap[idx]; heap[idx] = tmp; heap_insert_helper(heap, p); } } static void event_heap_insert(timer_ctx_t* ctx, timer_event_t* evt) { timer_event_t** heap = ctx->heap; uint32_t idx = ctx->nevts++; heap[idx] = evt; heap_insert_helper(heap, idx); } static void heap_del_helper(timer_event_t* heap[], uint32_t nevts, uint32_t idx) { if(idx >= nevts) { return; } uint32_t lchild = idx * 2 + 1; uint32_t rchild = idx * 2 + 2; uint32_t i = idx; if(lchild < nevts && (heap[i]->time > heap[lchild]->time)) { i = lchild; } if(rchild < nevts && (heap[i]->time > heap[rchild]->time)) { i = rchild; } if(i != idx) { timer_event_t* tmp = heap[i]; heap[i] = heap[idx]; heap[idx] = tmp; heap_del_helper(heap, nevts, i); } } static timer_event_t* event_heap_pop(timer_ctx_t* ctx) { timer_event_t* evt = ctx->heap[0]; ctx->heap[0] = ctx->heap[--ctx->nevts]; heap_del_helper(ctx->heap, ctx->nevts, 0); return evt; } static void event_re_enqueue(timer_ctx_t* ctx, timer_event_t* evt) { if(evt->enable && evt->interval > 0) { evt->time += evt->interval; if(evt->count > 0) { evt->count--; if(evt->count == 0) { evt->interval = 0; } } event_heap_insert(ctx, evt); } else { mpool_put(evt); } } void timer_run(timer_ctx_t *ctx) { while(ctx->nevts > 0 && ctx->heap[0]->time <= get_current_clock()) { timer_event_t *evt = event_heap_pop(ctx); if(evt->enable && evt->cb) { evt->cb(evt->data); } event_re_enqueue(ctx, evt); } } timer_ctx_t* timer_init(timer_ctx_t* ctx) { if(!ctx) { return NULL; } memset(ctx, 0, sizeof(timer_ctx_t)); //make sure mpool is large enough ctx->mpool = mpool_calloc(MAX_EVENT_COUNT, sizeof(timer_event_t)); // ctx->mpool = mpool_init(sizeof(timer_event_t), // (sizeof(timer_event_t) + sizeof(void*)) * MAX_EVENT_COUNT + 512, // NULL); if(!ctx->mpool) { goto error; } return ctx; error: mpool_cleanup(ctx->mpool); return NULL; } void timer_cleanup(timer_ctx_t *ctx) { if(!ctx) { return; } mpool_cleanup(ctx->mpool); ctx->mpool = NULL; } int timer_register( timer_ctx_t *ctx, timer_cb_t cb, void* data, uint32_t delay, uint32_t interval, uint32_t count) { timer_event_t *evt = NULL; if(ctx->nevts >= MAX_EVENT_COUNT) { goto err; } evt = (timer_event_t*)mpool_get(ctx->mpool); if(!evt) { goto err; } memset(evt, 0, sizeof(timer_event_t)); int idx = mpool_get_idx(ctx->mpool, evt); if(idx < 0) { goto err; } evt->id = idx; evt->time = get_current_clock() + delay; evt->cb = cb; evt->data = data; evt->interval = interval; evt->count = count; evt->enable = 1; event_heap_insert(ctx, evt); return idx; err: if(evt) { mpool_put(evt); } return -1; } void timer_cancel(timer_ctx_t *ctx, int id) { timer_event_t *evt = mpool_get_by_idx(ctx->mpool, id); if(evt) { evt->enable = 0; } }