Roman Shaposhnik created GUACAMOLE-663:
------------------------------------------
Summary: guacd SEGVs intermittently on systems with small(er)
thread stack sizes
Key: GUACAMOLE-663
URL: https://issues.apache.org/jira/browse/GUACAMOLE-663
Project: Guacamole
Issue Type: Bug
Components: guacd
Affects Versions: 1.0.0
Reporter: Roman Shaposhnik
I've observed at least guacd VNC plugin to run out of stack space when deployed
on Alpine Linux with musl libc. This is not limited to Alpine since quite a few
distributions are using musl these days (Void, Bedrock, etc.) and the list is
growing. musl libc is especially popular for distributions meant to be the base
for Docker containers. Hence, I believe addressing this will be quite important
for enabling use of guacd in modern containerized environments.
The issue itself is well know to Alpine folks:
[https://github.com/voidlinux/void-packages/issues/4147] and they feel (perhaps
for right reasons) that applications requiring large(r) thread stacksizes
should explicitly declare their requirement. There are currently two ways of
accomplishing this: either via a global pthread_setattr_default_np call
[https://www.openwall.com/lists/musl/2017/01/03/1] or via per-pthread_create
setting of defaults e.g.
[https://github.com/voidlinux/void-packages/commit/5d65448221af12296f7ef2339ced92771bb465b4]
I feel that the first option is a better choice for guacd. Attaching a
preliminary patch that makes guacd work fine on Alpine 3.8.x
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
