Re: [libvirt] [RFC 0/2] util: introduce timeout mode in virCommandRun/Async and virKModLoad

On Monday, August 13, 2018 at 10:02 AM, Daniel P. Berrangé wrote: Thanks for your quick reply! I hope I could explain my idea more clear:) The virKModLoad is used for loading external kernel modules which are not under the control of Libvirt. We cannot make sure they're always solid and strong (especially for various pci drivers). If the INIT process of a module is blocked for some reasons, the caller of the virKModLoad will be blocked itself and cannot known what happened internally and have no chance to handle it. Most modules need only dozens of milliseconds to complete loading. On my slow PC, module 'nbd' needs about 2 milliseconds to load and '8021q' needs about 30 milliseconds. And I will try to test other pci drivers. If loading time is longer than a few seconds, we can presume there are some exceptions or errors in the loading process of the module. With timeout mode, virKModLoad can report it to the caller and the caller could decide how to handle it. Timeout mode for virKModLoad is just an insurance which could avoid blocking indefinitely. It's more complicated to call virCommandRun/Async. So timeout mode for virCommandRun/Async is just an optional mode and the developers can decide wether they need to use it according to the command. With virCommandSetTimeout, commands can be set timeout in milliseconds. It can be used under one of these conditions: 1) In synchronous mode, using virCommandRun NOT in daemon mode. By default, virCommandRun waits for the command to complete & exit and then checks its exit status. If the caller uses virCommandSetTimeout before it, the command will be aborted internally when timeout and there is no need to use virCommandAbort to reap the child process. For example: int timeout = 3000; /* in milliseconds */ virCommandSetTimeout(cmd, timeout); if (virCommandRun(cmd, NULL) < 0) {     if (virCommandGetErr(cmd) == ETIME)         ... do something for timeout, e.g. report error ...     return -1; } The argument timeout of the virCommandSetTimeout accepts a positive value. When timeout, virCommandRun return -1 and the caller can use virCommandGetErr to check errorcode. And ETIME means command timeout. Note:virCommandSetTimeout cannot mix with virCommandDaemonize. Or virCommandRun fails directly. 2) In asynchronous mode, the caller can also call virCommandSetTimeout before virCommandRunAsync to limit the running time of the command. The caller uses virCommandWait to wait for the child process to complete & exit. virCommandWait returns -1 if timeout and the caller can use virCommandGetErr to check the reason. For example: char *outbuf = NULL; char *errbuf = NULL; int timeout = 3000; /* in milliseconds */ virCommandSetTimeout(cmd, timeout); virCommandSetOutputBuffer(cmd, &outbuf); virCommandSetErrorBuffer(cmd, &errbuf); virCommandDoAsyncIO(cmd); if (virCommandRunAsync(cmd, NULL) < 0)    return -1; ... do something ... if (virCommandWait(cmd, NULL) < 0) {     if (virCommandGetErr(cmd) == ETIME)         ... do something for timeout, e.g. report error ...     return -1; } Regards, Shilei