Speed throttling precision issues in the latest libcurl

From: Dmitry Karpov via curl-library <curl-library_at_lists.haxx.se>
Date: Tue, 5 Mar 2024 01:04:28 +0000

Hi All,

After running some speed throttling tests on the latest libcurl, I noticed that the speed throttling precision got lost again.

I reported this issue for 8.5.0, and it was fixed for 8.6.0 as part of https://github.com/curl/curl/commit/1da640abb6886aab822ff0c0da71b1df0ca89d0f, which avoided read looping when speed throttling was enabled
by setting curl_off_t max_recv to zero ( curl_off_t max_recv = data->set.max_recv_speed ? 0 : CURL_OFF_T_MAX; ) in readwrite_data() in lib\transfer.c.

But it looks like that part was lost during some recent readwrite_data() refactoring, and this function got back to read looping when the speed limit is set, which depending on network speed and download size can create speed deviations more than 10%.

In order to have more or less precise speed limits (<1% on average), we need to avoid read looping for cases of the speed limit, so the following change should be probably applied:

---
lib/transfer.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/transfer.c b/lib/transfer.c
index 3ae4b61c0..3ad7904de 100644
--- a/lib/transfer.c
+++ b/lib/transfer.c
_at__at_ -495,7 +495,7 _at__at_ static CURLcode readwrite_data(struct Curl_easy *data,
     /* Observe any imposed speed limit */
     if(bytestoread && data->set.max_recv_speed) {
       curl_off_t net_limit = data->set.max_recv_speed - total_received;
-      if(net_limit <= 0)
+      if(net_limit <= 0 || total_received)
         break;
       if((size_t)net_limit < bytestoread)
         bytestoread = (size_t)net_limit;
--
which will break the reading loop after the first read.
Thanks,
Dmitry Karpov

-- 
Unsubscribe: https://lists.haxx.se/mailman/listinfo/curl-library
Etiquette:   https://curl.se/mail/etiquette.html

Received on 2024-03-05