Niels Provos

Nick announced the release of Libevent-2.0.7-rc today. Here is an excerpt from his email:

Thanks to everybody who reported and fixed bugs in Libevent 2.0.6-rc,
Libevent 2.0.7-rc should be much more stable and portable, especially
for people using IOCP, Windows, rate-limiting, or threads.

There are also numerous small bugfixes thoughout the codebase (though
still not, alas, in the http stuff).

For a complete list of changes, just see the ChangeLog included with
the source distribution.

You can download source code of libevent releases from monkey.org

Libevent 2.0.4-alpha is now available for download:

http://monkey.org/~provos/libevent-2.0.4-alpha.tar.gz
http://monkey.org/~provos/libevent-2.0.4-alpha.tar.gz.sig

The complete change list is available here.

Some of the feature improvements include:

• bufferevents can now be rate limited
• http connections can now resolve host names asynchronously
• a facility for lock debugging
• arc4random() for evdns

However, we (that means mostly Nick) have also made a large number of bug fixes and stability improvements across many platforms. Many thanks to everyone who helped by providing bug reports and patches including Brodie Thiesfield, Dagobert Michelsen, Evan Jones, Joachim Bauch, Pavel Plesov, Roman Puls, Sebastian Hahn, William Ahern, Yasuoka Masahiko and Zhuang Yuyao.

In a separate email, Nick also provided a much more verbose description of what all changed.

Tom Pusateri reported success with using OpenSSL client certificates and libevent's builtin OpenSSL support. Here is what he wrote on the mailing list:

I tried 2.0.3 alpha against the Apple Push notification feedback service which requires a client key/certificate and it works great.

One hint... Make sure you add the key and cert to the SSL context before calling SSL_new(). Otherwise, you'll get an error that looks like:

sslv3 alert handshake failure in SSL routines SSL3_READ_BYTES

Here's the working code:

static void
init_feedback_service(struct event_base *ev_base,
    struct evdns_base *dns)
{
   int rc;
   struct bufferevent *bev;
   SSL_CTX *ssl_ctx;
   SSL *ssl;

   ssl_ctx = SSL_CTX_new(SSLv3_method());

   rc = SSL_CTX_use_certificate_file(ssl_ctx, "my_apple_cert_key.pem",
       SSL_FILETYPE_PEM);
   if (rc != 1) {
       errx(EXIT_FAILURE, "Could not load certificate file");
   }
   rc = SSL_CTX_use_PrivateKey_file(ssl_ctx, "my_apple_cert_key.pem",
       SSL_FILETYPE_PEM);
   if (rc != 1) {
       errx(EXIT_FAILURE, "Could not load private key file");
   }

   ssl = SSL_new(ssl_ctx);
   bev = bufferevent_openssl_socket_new(ev_base, -1, ssl,
       BUFFEREVENT_SSL_CONNECTING, BEV_OPT_CLOSE_ON_FREE);
   bufferevent_setcb(bev, feedback_read_cb, NULL,
       feedback_event_cb, NULL);
   rc = bufferevent_socket_connect_hostname(bev, dns, AF_INET,
       "feedback.sandbox.push.apple.com", 2196);
   if (rc < 0) {
       warnx("could not connect to feedback service: %s",
             evutil_socket_error_to_string(EVUTIL_SOCKET_ERROR()));
       bufferevent_free(bev);
       return;
   }
   bufferevent_enable(bev, EV_READ);
}

It has been a while since the last alpha release of libevent-2.0. Yesterday, we released 2.0.3-alpha which can be downloaded from

http://monkey.org/~provos/libevent-2.0.3-alpha.tar.gz

Please, give it a spin and let us know if you run into any problems. There have been a lot of changes since the last release, mostly due to Nick's hard work. Here are just some highlights, the ChangeLog contains the full story:

- SSL/TLS support on bufferevents, using the OpenSSL library
- Improved searching on evbuffer objects
- Improved support for Windows
- More efficient memory allocation for event_bases that use epoll
- Improved thread-safety
- The IOCP bufferevent backend is now exposed on Windows; many thanks to Christopher Davis for his work.

Many thanks to everyone who helped with patches and bug reports including Rocco Carbone, Brodie Thiesfield, Caitlin Mercer, David Reiss, Alexander Pronchenkov, Jacek Masiulaniec, Ka-Hing Cheung, Christopher Davis, Ferenc Szalai, and Ryan Phillips.

Edited to fix the link.

We just released a new stable version of Libevent that fixes the following problems:

• If the kernel tells us that there are a negative number of bytes to read from a socket, do not believe it. Fixes bug 2841177; found by Alexander Pronchenkov.
• Do not allocate the maximum event queue and fd array for the epoll backend at startup. Instead, start out accepting 32 events at a time, and double the queue's size when it seems that the OS is generating events faster than we're requesting them. Saves up to 512K per epoll-based event_base. Resolves bug 2839240.
• Fix compilation on Android, which forgot to define fd_mask in its sys/select.h
• Do not drop data from evbuffer when out of memory; reported by Jacek Masiulaniec
• Rename our replacement compat/sys/_time.h header to avoid build a conflict on HPUX; reported by Kathryn Hogg.
• Build kqueue.c correctly on GNU/kFreeBSD platforms. Patch pulled upstream from Debian.
• Fix a problem with excessive memory allocation when using multiple event priorities.
• When running set[ug]id, don't check the environment. Based on a patch from OpenBSD.

A new alpha release of libevent 2.0 is on its way, too. Thanks to everyone who submitted patches and bug reports.

The source code is available at http://www.monkey.org/~provos/libevent-1.4.13-stable.tar.gz. Don't forget to verify the signature.

Nick just announced two new libevent releases. Here is his summary.

Libevent 1.4.12-stable:
You can find the source in the usual place:

http://monkey.org/~provos/libevent-1.4.12-stable.tar.gz

This is a bugfix-only release, and some of the bugs were kind of nasty. I'd recommend that you upgrade, especially if you are writing code that uses epoll or evdns.

Changes in 1.4.12-stable:

• Try to contain degree of failure when running on a win32 version so heavily firewalled that we can't fake a socketpair.
• Fix an obscure timing-dependent, allocator-dependent crash in the evdns code.
• Use _VA_ARGS_ syntax for varargs macros in event_rpcgen when compiler is not GCC.
• Activate fd events in a pseudorandom order with O(N) backends, so that we don't systematically favor low fds (select) or earlier-added fds (poll, win32).
• Fix another pair of fencepost bugs in epoll.c. [Patch from Adam Langley.]
• Do not break evdns connections to nameservers when our IP changes.
• Set truncated flag correctly in evdns server replies.
• Disable strict aliasing with GCC: our code is not compliant with it.

Libevent-2.0.2-alpha:
The first alpha release in the long-promised Libevent 2.0 series is finally out. You can download Libevent 2.0.2-alpha from:

http://monkey.org/~provos/libevent-2.0.2-alpha.tar.gz

This is an alpha release. Libevent 2.0 is not finished. There will be bugs, and we make no promises about the stability of any APIs introduced in the 2.0.x-alpha releases. When you find bugs, please let us know.

Libevent 2.0 is intended to be backward compatible with the Libevent 1.4 APIs[*]. Any program that worked with Libevent 1.4 should still work with Libevent 2.0, unless we screwed up. Please test your programs when you have a chance, so that if we did screw up, we can notice soon.
[*] Unless you were messing around with the internals of internal structures.

This release adds many new features to the previous alpha release, and fixes many bugs. See the ChangeLog for full details. Highlights include:

• evdns is now threadsafe, with locking support
• There's an evconnlistener type that you can use to abstract cross-platform differences in accepting connections.
• The evbuffer interface (and therefore bufferevents) now supports zero-copy much better.
• About a zillion fixes for tricky bugs in the new Libevent 2.0.1-alpha code.

Special thanks to everybody who helped find bugs and improve the code, especially James Mansion, Zack Weinberg, and Joachim Bauch.

In preparation for CodeCon, Nick and I wanted to see how HTTP performance differs between Libevent 1.4 and Libevent 2.0. HTTP is a good test case as it exercises many of the optimized components. Here is a preliminary result.

The libevent HTTP server is serving 200,000 bytes of content for each request. Apache's benchmark tool ab was used to make 15,000 requests with 40 requests happening in parallel.

• 1.4.10:
  Requests per second: 1450.79 [#/sec] (mean)
• 2.0:
  Requests per second: 1961.99 [#/sec] (mean)
• 2.0 (evbuffer_add_reference):
  Requests per second: 3979.31 [#/sec] (mean)

In Libevent 2.0, the evbuffer interface was rewritten to avoid memory copies where possible. This seems to result in a 35% performance improvement. The evbuffer_add_reference() API allows external memory to be associated with an evbuffer and thus avoids another memory copy. This results in about 100% performance increase. In comparison to Libevent 1.4, this is almost 175% faster.

In the meantime, Nick is working on making IOCP available for Windows.

We just released libevent 1.4.9-stable. You can download the source from the usual place:

http://monkey.org/~provos/libevent-1.4.9-stable.tar.gz

This release fixes a number of bugs:

• fixed several memory leaks in the HTTP layer.
• fixed signal handling for multi-threaded applications.
• fixed issues with the timer cache when leaving/entering the event loop.

Check here for a more detailed change list.

Thanks to Dean McNamee, Victor Chang, Alejo Sanchez, Richard Jones, Robin Haberkorn and everyone else who reported bugs or supplied patches.

This is another bug fix release. Here are the problems fixed in this one:

• Match the query in DNS replies to the query in the request; from Vsevolod Stakhov.
• Fix a merge problem in which name_from_addr returned pointers to the stack; found by Jiang Hong.
• Do not remove Accept-Encoding header

You can download the source from the usual place:

http://monkey.org/~provos/libevent-1.4.8-stable.tar.gz

You can download the source from the usual place:

http://monkey.org/~provos/libevent-1.4.7-stable.tar.gz

This release fixes a bug where headers arriving in multiple packets were not parsed. The bug fix is from Jiang Hong. Thank you.

With all the DNS stuff that is going on at the moment, it took longer than planned to release libevent 1.4.6-stable, but here it is. You can download the source from the usual place:

http://monkey.org/~provos/libevent-1.4.6-stable.tar.gz

This was meant to be a quick bug fix release, but a number of changes have crept in. Here's a brief summary:

• Several HTTP fixes
  • Corrected handling of trailing headers in chunked replies
  • Correctly deal with multi-line HTTP headers
• Improved signal handling for kqueue and epoll backends
• Various other bug fixes

See the changelog for full details.

We would like to thank the people who have reported bugs and patches including Scott Lamb, Moshe Litvin, Alexander Drozdov, Adam Langley, Ian Bell and others. To report a bug, make a feature request, or submit code, you can use our sourceforge interface.

I am happy to announce the release of libevent 1.4.5-stable. You can download the source here:

http://monkey.org/~provos/libevent-1.4.5-stable.tar.gz

There have been a few bug fixes since 1.4.4-stable. Here's a brief summary:

• Several HTTP fixes
• Fixed the Windows port
• event_rpcgen.py correctly generates fixed length entries now

See the changelog for full details.

We would like to thank the people who have reported bugs including Forest Wilkinson, liusifan and others. To report a bug, make a feature request, or submit code, you can use our sourceforge interface here.

I am happy to announce the release of libevent 1.4.4-stable. You can download the source here:

http://monkey.org/~provos/libevent-1.4.4-stable.tar.gz

There have been a few bug fixes since 1.4.3-stable. Here's a brief summary:

• Epoll fixes
  
  • Correctly handle timeouts larger than 35 minutes for older Linux kernels


• Tagging fixes:
  
  • Fixes a potential stack corruption on 64-bit architectures


• Bufferevent changes:
  
  • Fixes a corner cases for read watermarks
  
  
  • Expose the watermark functionality in bufferevent


• HTTP changes:
  
  • Fix a bug where it was not possible to accept on multiple sockets
  
  
  • Expose the evhttp_accept_socket() functionality which allows the HTTP server to listen on an already created socket


• Portability fixes:
  
  • Fix functionality and compile problems on Windows and IRIX
  
  
  • Provide a timercmp function that works on all platforms

See the changelog for full details.

We would like to thank the people who have reported bugs including Matt Domsch, Forest Wilkinson, Jon and several anonymous reporters. To report a bug, make a feature request, or submit code, you can use our sourceforge interface.