Understanding CURLOPT_NOSIGNAL

We were faced with a libcurl related crash in this release.

The code making DNS requests in a kubernetes cluster crashed when left to run overnight.

Most posts suggested to use CURLOPT_NOSIGNAL set to 1 to solve the issue.

Although its pretty much clear why this issue happens (https://curl.haxx.se/libcurl/c/CURLOPT_NOSIGNAL.html), the effects of setting CURLOPT_NOSIGNAL to 1 were not yet clear to me.

After some study I found the following 

If you have a multi threaded c++ application setting CURLOPT_NOSIGNAL to 1

will do the following 

1. libcurl will no longer install signal handlers. This means you need to install handlers for SIGPIPE and SIGCHILD yourself in your application. Otherwise you may end up in another crash.
2. libcurl will not timeout when a DNS resolution takes too long. So you will need to understand which requests are timing out and kill the request if it takes too long.

The right approach I feel is to compile libcurl with c-ares support and leave the CURLOPT_NOSIGNAL to its default value .i.e. 0

This way the DNS resolution related alarms will not be trigerred and SIGPIPE related handlers will be installed by libcurl.