-
|
I have a multithreaded application where I have several worker threads and a different thread running the uv_loop. I am writing back to the loop using an async_send cb. However, when I do it that way the uv_write_cb doesn't get called immediately, but it seems to be called the next time some other event callback happens on the loop. On the example below, 'on_write_end' is not called but until there is some other event on the loop. void async_write(uv_async_t* async) {
response_data_t *response_data = async->data;
uv_stream_t *client = response_data->client;
client_request_data *client_data = client->data;
int* buffer = response_data->buffer;
size_t size = response_data->size;
uv_buf_t wrbuf = uv_buf_init((char*)buffer,size);
uv_write_t* write_req = response_data->write_req;
write_req->data = (void*)client_data->client;
uv_write(write_req,client,&wrbuf,1,on_write_end);
client_data->ready_to_write = 1;
} |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 7 replies
-
|
No, that's not expected. 9 out of 10 times it's some kind of thread safety issue in the program but your snippet doesn't provide enough context for me to guess what. NB: I'm assuming that "immediately" doesn't mean you expect on_write_end to have been called by the time uv_write() returns. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your help. I found the issue: I was reusing the uv_write_t struct for several uv_write operations, which caused a weird behavior with the callback. |
Beta Was this translation helpful? Give feedback.
-
|
At the risk of sounding like an idiot, I need to confess that I fell into this rabbit whole again. Let me try to give the most comprehensive example possible. I have a worker thread, and my uv_loop is running on a different thread. The Worker thread processes some data, and when it is finished it sends data back to the uv_loop thread through and async call. Here is a sample of what the worker thread does: The async_write call marshalls the information out, and does a uv_write_ which in turn assings the callback. My main problem is that the 'on_write_finish' callback is not reached but until something else happens on the uv_loop (eg. a client disconnects or there is another read_cb event). I would like to use the callback to free my buffers and other housekeeping operations. There are no other mutex_locks in the uv_loop thread other than what displayed in the example above which should happen before the uv_write is done and after any subsequent call to it again. |
Beta Was this translation helpful? Give feedback.
-
|
Have you considered using uv_queue_work instead? It's integrated into libuv and it allows you to do work in an external thread easily. |
Beta Was this translation helpful? Give feedback.
-
|
I must admit that I am a bit ashamed of the source of the issue, but all seemed to came down for me not calling async events using uv_async_send() mechanism. @saghul, yes I've used them and like them very much. I am using this for a producer-consumer model for my MSc thesis and therefore wanted more control on instantiation of threads. |
Beta Was this translation helpful? Give feedback.
I must admit that I am a bit ashamed of the source of the issue, but all seemed to came down for me not calling async events using uv_async_send() mechanism.
@saghul, yes I've used them and like them very much. I am using this for a producer-consumer model for my MSc thesis and therefore wanted more control on instantiation of threads.