[Libevent-users] SIGPIPE in multithreaded program
Hi, I am using libevent in a multithreaded program, the main thread containing the libevent loop, and other threads doing disk I/O. From mail from this list, and the changelogs I was under the impression that libevent handles signals better in a multithreaded environment since 1.3, but now I think I was wrong. Is it necessary to specifically ignore signals in every thread I create? Thanks, Ron smime.p7s Description: S/MIME Cryptographic Signature ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
[Libevent-users] SSL enabling libevent programs, is there an example available?
Hi, subject says it all. If there's no example with libevent, can someone recommend another place to look? I tried Googling but couldn't find a concise example. Thanks, Ron smime.p7s Description: S/MIME Cryptographic Signature ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
[Libevent-users] thread-safety (and performance)
When initiating HTTP requests in a multi-threaded process, under any significant load you will start finding horrible memory corruption problems, sometimes resulting in bad requests and sometimes resulting in core dumps. Also, under any significant load, you'll probably find that you are dropping a large number of connections due to the listen queue overflowing while your code is performing some other task. This is a patch that I think largely deals with these. Also, http_hostportfile() is not thread-safe, so I've attached a thread-safe version that can be used in its place (unfortunately I can't think of any clean way to make the current interface thread-safe without introducing a memory leak). /* TH: remember to free() opaque after you're done with this */ int http_hostportfile_r(const char *url, char **phost, u_short *pport, char **pfile,void **opaque) { struct http_hostportfile_threadsafe_opaque { char shost[1024]; char sfile[1024]; } *opaquedata; opaquedata = malloc(sizeof(*opaquedata)); *opaque = opaquedata; char *p, *p2; int len; u_short port; len = strlen(HTTP_PREFIX); if (strncasecmp(url, HTTP_PREFIX, len)) return (-1); url += len; /* We might overrun */ if (strlcpy(opaquedata-shost, url, sizeof(opaquedata-shost)) = sizeof(opaquedata-shost)) return (-1); p = strchr(opaquedata-shost, '/'); if (p != NULL) { *p = '\0'; p2 = p + 1; } else p2 = NULL; if (pfile != NULL) { /* Generate request file */ if (p2 == NULL) p2 = ; snprintf(opaquedata-sfile, sizeof(opaquedata-sfile), /%s, p2); } p = strchr(opaquedata-shost, ':'); if (p != NULL) { *p = '\0'; port = atoi(p + 1); if (port == 0) return (-1); } else port = HTTP_DEFAULTPORT; if (phost != NULL) *phost = opaquedata-shost; if (pport != NULL) *pport = port; if (pfile != NULL) *pfile = opaquedata-sfile; return (0); } Index: http.c === --- http.c (revision 1) +++ http.c (working copy) @@ -308,7 +308,7 @@ evhttp_make_header_request(struct evhttp_connection *evcon, struct evhttp_request *req) { - static char line[1024]; + char line[1024]; const char *method; evhttp_remove_header(req-output_headers, Accept-Encoding); @@ -378,9 +378,9 @@ evhttp_maybe_add_content_length_header(struct evkeyvalq *headers, long content_length) { + char len[12]; /* XXX: not thread-safe */ if (evhttp_find_header(headers, Transfer-Encoding) == NULL evhttp_find_header(headers, Content-Length) == NULL) { - static char len[12]; /* XXX: not thread-safe */ snprintf(len, sizeof(len), %ld, content_length); evhttp_add_header(headers, Content-Length, len); } @@ -394,7 +394,7 @@ evhttp_make_header_response(struct evhttp_connection *evcon, struct evhttp_request *req) { - static char line[1024]; + char line[1024]; snprintf(line, sizeof(line), HTTP/%d.%d %d %s\r\n, req-major, req-minor, req-response_code, req-response_code_line); @@ -433,7 +433,7 @@ void evhttp_make_header(struct evhttp_connection *evcon, struct evhttp_request *req) { - static char line[1024]; + char line[1024]; struct evkeyval *header; /* @@ -1999,7 +1999,7 @@ if ((fd = bind_socket(address, port)) == -1) return (-1); - if (listen(fd, 10) == -1) { + if (listen(fd, 8192) == -1) { event_warn(%s: listen, __func__); EVUTIL_CLOSESOCKET(fd); return (-1); ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] thread-safety (and performance)
On Mon, Jan 21, 2008 at 04:14:08PM -0800, Tani Hosokawa wrote: snip @@ -1999,7 +1999,7 @@ if ((fd = bind_socket(address, port)) == -1) return (-1); - if (listen(fd, 10) == -1) { + if (listen(fd, 8192) == -1) { event_warn(%s: listen, __func__); EVUTIL_CLOSESOCKET(fd); return (-1); Probably better to use SOMAXCONN instead of an arbitrary number. ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] SSL enabling libevent programs, is there an example available?
Ron Arts wrote: Hi, subject says it all. If there's no example with libevent, can someone recommend another place to look? I tried Googling but couldn't find a concise example. For example, w/ OpenSSL you can turn the socket to be non-blocking and watch for what openssl 'needs' after a failed attempt to read/write... Ex: int ret = SSL_read(ssl, buffer, len); if(ret 0) { int err = SSL_get_error(ret); switch(err) { case SSL_ERROR_WANT_WRITE: /* Hook up libevent to wait on write then retry w/ exact same argument */ break; case SSL_ERROR_WANT_READ: /* Hook up libevent to wait on read then ... */ break; default: /* Something else. */ } } ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] SSL enabling libevent programs, is there an example available?
On Tue, Jan 22, 2008 at 01:09:11AM +0100, Ron Arts wrote: Hi, subject says it all. If there's no example with libevent, can someone recommend another place to look? I tried Googling but couldn't find a concise example. http://25thandclement.com/~william/projects/libevnet.html Download the tarball. OpenSSL bindings are in src/tls.c. There's lots more code in the library for integrating with libevent, but I wrote tls.c many years ago (before libevnet, and possibly before libevent), and it should be easy to rip it out and use however you want. At the very least, tls_read_try(), tls_write_try() and tls_accept_try() should help you out. ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] SIGPIPE in multithreaded program
William Ahern wrote: On Tue, Jan 22, 2008 at 01:06:30AM +0100, Ron Arts wrote: Hi, I am using libevent in a multithreaded program, the main thread containing the libevent loop, and other threads doing disk I/O. From mail from this list, and the changelogs I was under the impression that libevent handles signals better in a multithreaded environment since 1.3, but now I think I was wrong. Is it necessary to specifically ignore signals in every thread I create? In Unix it's necessary to change the default behavior of SIGPIPE if you don't want your process killed. libevent, AFAIK, won't do this for you. If won't independently setup handlers for all the different signals. Most daemon applications ignore SIGPIPE as part of their initialization. The default behavior caters to simple shell applications that can't be bothered to check the return value of a write. #include signal.h struct sigaction sa; sa = sa_initializer; sa.sa_handler = SIG_IGN; sa.sa_flags = 0; sigemptyset(sa.sa_mask); if (0 != sigaction(SIGPIPE, sa, 0)) err(EXIT_FAILURE, sigaction); Oops, I'm sorry, I did not make myself clear, while writing the email I edited it a lot, and forgot to mention that indeed I ignore SIGPIPE in my initialisation code: struct event evsigpipe; event_init(); event_set(evsigpipe, SIGPIPE, EV_SIGNAL|EV_PERSIST, signal_cb, evsigpipe); event_add(evsigpipe, NULL); event_dispatch(); void signal_cb(int fd, short event, void *arg) { struct event *signal = arg; extern int reload_database; switch(EVENT_SIGNAL(signal)) { case SIGTERM: // sigterm fprintf(stderr, Killed); exit(1); break; case SIGHUP: // sighup fprintf(stderr, Got SIGHUP); reload_config = TRUE; break; case SIGUSR1: // reload entire database fprintf(stderr, Got SIGUSR1); reload_database = TRUE; break; default: fprintf(stderr, %s: got signal %d, __func__, EVENT_SIGNAL(signal)); break; } } But my program is still being killed with SIGPIPE occasionally. I am using threads, and I presume sometimes one of the other threads receives the SIGPIPE signal instead of the main thread, and I *think* that in such a case my program exits. But what I meant to ask was: isn't libevent supposed (since 1.3) to handle multithreading and ensure that only one thread receives the signal? Or should I specifically add code at the beginning of each thread to ignore SIGPIPE? Thanks, Ron ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users smime.p7s Description: S/MIME Cryptographic Signature ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users