When HHVM doesn't work quite right | Hacker News Reader

When HHVM doesn't work quite right

ma.rtin.so

87 pointsmartin_26 comments

You might check out http://circus.readthedocs.org/en/latest/ instead of supervisord; you can restart managed processes on a schedule, and it also publishes events to zmq and statsd for integration with any other tooling you might want to use for health checks.

Sweet thanks! I had to make a workaround because supervisord doesn't have a way to serially restart "groups". Will give this a shot!

Even that is not necessary. If the original service was running under systemd, why not just use service instances? (see description in https://www.digitalocean.com/community/tutorials/understandi... for example)

3.8 and 3.9-dev are way more stable than any previous version.

We used to have to restart hhvm weekly to prevent white screens of death but no longer with 3.8+

3.8 is also 10% faster than previous versions: http://hhvm.com/blog/9803/hhvm-3-8-0

I just started a new Laravel project and decided to try hhvm. Unfortunately, on 3.8.0 it segfaults under load just serving the default Laravel page. It's really unfortunate since I'd love to give hhvm a shot.

hhvm has tested compatibility with Laravel. Are you sure you were doing everything right?

A PHPUnit test suite passing and being stable under high load are two entirely unrelated things.

And having a local project segfaulting with zero extra code is something else altogether.

Are you sure you built it right?

Maybe try 3.9-dev (trunk)to see if it was a bug.

When did you try it? HHVM 3.8 was only released officially a week ago, perhaps you were on the unstable branch?

I just started the project 2 days ago, and it was installed from the official HHVM repository:

sudo apt-get install software-properties-common

sudo apt-key adv --recv-keys --keyserver hkp://keyserver.ubuntu.com:80 0x5a16e7281be7a449

sudo add-apt-repository "deb http://dl.hhvm.com/ubuntu $(lsb_release -sc) main"

sudo apt-get update

sudo apt-get install hhvm

$hhvm --version

HipHop VM 3.8.0 (rel)

Compiler: tags/HHVM-3.8.0-0-gbb32f8d645b28561d8491870c44730b9b3f08d35

Repo schema: 74bee5d310450724cfd49245fd6c2e1e4323b10a

if you can run it under gdb, repro, and get a "thread apply all bt" that'd be very helpful. The HHVM team is very responsive on GitHub issues!

That's cool I guess, but this article is more about setting up a load balancer than it is about HHVM. I was hoping for some juicy bit about it completely hosing up on some obscure PHP function.

"nginx does provide health_check in its commercial builds, but at this time I'm not too inclined to pay $1500/server/year for the pleasure."

But Varnish does it for free out of the box ! Quick example from the documentation (https://www.varnish-cache.org/trac/wiki/BackendPolling) :

backend b0 { .host = "phk.freebsd.dk"; .probe = { .url = "/probe.cgi"; .timeout = 34 ms; .interval = 1s; .window = 10; .threshold = 8; } }

Who goes "this software has stability problems" and decides the correct solution is to go with the even more unstable branch?

People that have identified leaks affecting them which are fixed on the nightlies?

People trading some known bugs for a whole heap of new ones, ones likely to change all the time? Doesn't sound like the smartest operations move.

Given they've gone and written stuff to automatically restart the processes under certain conditions anyway, might as well just catch the memory leak and restart, rather than gamble with "not for production use" software in production environments.

The build we're using was considered pretty safe and was performing better for us than 3.7.x - of course we aren't grabbing the new nightly constantly and plan to move to a "stable" branch soon :)

Pragmatists, who know that "stable" and "unstable" branch are just tags, and what's important is the actual stability of the software in each as it relates to their use case.

Even worse, I've been known to cherry-pick commits from different unstable branches to get a semblance of "what I need".

It's not pretty, but we do what we need to do.

Sweet, I am going to borrow your health check script. I can't wait until official packaged support for the ZMQ extension.

> First, I disabled HHVM via systemd and substituted in supervisord.

Why? what does supervisord have that systemd doesn't provide?

I'm not aware of a method to manage pools with systemd, also it requires processes to be daemonized which would make configuration a little messier. Also I'm simply more familiar with supervisord as systemd only recently became the default init system in debian.

Not sure about the other merits, but systemd definitely does not require daemonized processes.

http://homepage.ntlworld.com/jonathan.deboynepollard/FGA/sys...

> I'm not aware of a method to manage pools with systemd

I'm not sure what you mean by manage pools. Maybe you're asking how to make multiple processes? Check out unit templates, they're the ones with an '@' in them. see `man systemd.unit` for more info.

> also it requires processes to be daemonized which would make configuration a little messier.

It does not. infact, it prefers you don't. Check out the various service unit types. see `man systemd.service` or http://www.freedesktop.org/software/systemd/man/systemd.serv...

in the time you read this article, facebook developers have added another 1500 lines of php and pushed most of it to production.

Reply on news.ycombinator.com