Runtime code modification. Erlang? No, Python | Hacker News Reader

Runtime code modification. Erlang? No, Python

blog.ducksboard.com

41 pointswulczer17 comments

It isn't that Erlang is anywhere near the only language that can do code reloading, it is that Erlang is unusual in the scope of reloading it permits and the care that it takes, to the point that reloading can be used in Erlang not as a clever hack to bridge gaps, but used as a planned update mechanism that causes zero downtime.

Since data in Erlang is immutable, you don't have to worry about what references will be left dangling. Since Erlang has no user-defined data types, you don't have to worry about redefining a class. (Big problem with Python reloading, all existing objects still belong to the old class instance.) Since Erlang is structured in terms of processes, and OTP can track which process is which thing, OTP has a defined mechanism for taking running server processes, examining their state, and allowing you to upgrade from old states to new states cleanly, and since in Erlang a process is basically nothing but its call stack and its current state, that upgrade is complete and relatively safe. You can literally upgrade a server mid-TCP connection without the client knowing.

Most languages, if poked properly, can change code at runtime. Even C can unload a dynamic library and load a new version of it at runtime. It's just that mutable values and poorly-defined boundaries between systems make it range from effectively impossible through "usable only as a desparate hack" (as seen here), Erlang makes it production-quality.

I say this only for educational purposes, not advocacy per se. I'd like to see further understanding of how Erlang works so that other language communities are more likely to pick up on the essential features of Erlang, instead of getting stuck on the superficial ones which is the usual case. (OTP is what you need to port, not "message passing"...)

> OTP is what you need to port, not "message passing"...

So true.

This reminded me of that story about Lisp at the JPL, when the author debugged and hotswapped code midway through a mission thanks to a built-in REPL.

"Having a read-eval-print loop running on the spacecraft proved invaluable in finding and fixing the problem."

http://www.flownet.com/gat/jpl-lisp.html

Since then, I always make sure I have a (safe) REPL backdoor into my long-running Perl apps.

>>(safe) REPL backdoor into my long-running Perl apps.

Is it on CPAN?

Not that I know of. But it's so simple to hook input to an eval in Perl (or any dynamic language for that matter), that I've always wrote it myself. It can also be implementation specific (preforking server, event-loop, dispatcher daemon...). I've used different approaches throughout the years:

- a controller action in Catalyst that listens at http://host/repl which evals the POST parameters. As a client, I use a simple admin-repl html page.

- in the main loop of a daemon script, a quick check for a row in a DB table or a special file in the server. When the server detects a new repl event, it evals the contents and writes the response back (either to the DB or to a file log).

- a 3 line AnyEvent::JSONRPC::Lite server that listens on a given local port for a string to eval. A matching JSONRPC client that calls that port with the repl string. I've used it to run scripts, and recently with a Corona-based Dancer app.

Maybe it could become a Plack::Middleware someday.

Poke the word REPL into search.cpan.org and start looking around. Since you frequently want it integrated with some network you might be able to find one for whatever network thing you are doing.

We used this technique once in one of our twisted servers to quickly fix in-memory data structures that had been loaded from old sources.

Recently, my team has been developing most of our newer network services using gevent. It's probably the closest thing to erlang that python has when it comes to programming asynchronous servers.

One of the things I always missed was this manhole feature, but while rummaging around the other day, I found this:

http://www.gevent.org/gevent.backdoor.html

Haven't tried it yet, but it looks like it will achieve the same task :)

Mix gevent and zeromq and you're even closer to Erlang.

> Soon enough I found the issue:

I am just wondering how long would it take if the author and the Bad Engineer were different people.

Btw, it's still not the same as Erlang. So, I don't think the title is fare.

I haven't tried this but I wonder if hot loading code could be done using the manhole and the reload() BIF.

Probably not, Python's reload has its fair share of caveats. There was an effort in Twisted to provide robust reloading support for Python modules[1], but I think it's a bit abandoned.

That being said, we are using reload() with manhole to update code. You just have to be extra careful and know what you're doing.

[1] http://twistedmatrix.com/documents/11.0.0/api/twisted.python...

Sorry, but to my understanding that's not even close to Erlang's code hot-swapping.

In Erlang one could easily update any module, and re-importing modules is quite complicated in Python. Update will be seamless, too - old code will be running its last reductions (so already connected clients won't experience downtime) while new messages will be processed with a new code.

Surely, it is possible to achieve the same in Python, but that'd require a lot more code than monkey-patching several values.

I did some googling and it looks like this only works if you're running twisted. Is anyone aware of any options that will work with Django or other long running python processes?

That's one advantage of scripting languages :)

Depends on how you define a "scripting" language. If you mean an interpreted language, then it's not true; you can do hot code swapping on native code. Erlang's HiPE does this.

You could do the same on compiled C by attaching gdb to the process and changing the timeout value so this has not much to do with "scripting". Just saying.

True but this was only about values. Changing values is not the most you can do with dynamic languages at runtime. In Python it is possible to create a completely new method implementation at runtime.

Not to say that dynamically changing code at runtime is totally impossible in C - buffer overflow attacks do exactly that.

Reply on news.ycombinator.com