IRC, freenode, #hurd, 2011-08-25
< frhodes> how can I replace an existing running server with a new one
without rebooting?
< antrik> frhodes: depends. if other critical things depend on it, you
can't. there is no mechanism to serialize and pass on the open sessions
< antrik> in some situations, you can orphan the old translator while
starting a new one, so the previous clients will stay with the old one
while new one will get the new one
< antrik> obviously that only works for things that aren't exclusive by
nature
< antrik> in some cases, you might even be able simply to remove the old
translator... but obviously only for non-critical stuff :-)
IRC, freenode, #hurd, 2013-10-21
<braunr> mhmm, there is a problem with thread destruction
?fix have kernel resources.
<braunr> actually, translator self destruction
<braunr> if a request arrives after the last thread servicing a port set
returns from mach_msg because of a timeout, but before the translator is
detached from its parent, the client will get an error
<braunr> it should very rarely happen, but if it does, we could face the
same kind of issues we have when a server crashes
<braunr> e.g. sshd looping over select() returning EBADF, consuming all cpu
<braunr> not sure we want to introduce such new issues
<braunr> i don't think i'll be able to make translators disappear reliably
..
<braunr> but at least, thread consumption will correctly decrease with
inactivity
IRC, freenode, #hurd, 2014-01-30
<sjbalaji> can any one exmplain me hello translator ? I am new to hurd
<teythoon> sjbalaji: sure, what do you want to know ?
<teythoon> how to use it ?
<sjbalaji> No I mean what is the main reason of that translator. I am
familiar with Linux.
<gnu_srs> sjbalaji: start with:
https://www.gnu.org/software/hurd/hurd/documentation/translator_primer.html
<sjbalaji> I ran that example but I am still clueless about the actual
reason behind the translators and this simple hello world translator
example.
<teythoon> sjbalaji: the Hurd is a multiserver os, almost all functionality
lives in userspace servers called 'translators'
<teythoon> sjbalaji: the Hurd uses the file system as namespace to lookup
these servers
<teythoon> e.g. /servers/socket/1 is /hurd/pflocal and handles pipes and
unix socket communication
<sjbalaji> I can see from the example that a hello file is associated with
a /hurd/hello translator
<teythoon> yes
<teythoon> think of translators like FUSE-based filesystems, only more
general
<teythoon> b/c translators are not restricted to provide filesystem-like
services
<sjbalaji> So this example hello translator just adds hello world in the
associated file, am I correct ?
<teythoon> it's not adding stuff to a file
<teythoon> say you did settrans -ac /tmp/hi /hurd/hello, if you do cat
/tmp/hi, cat does some rpc calls to the started /hurd/hello program that
returns 'hello world' as the file contents
<teythoon> in a sense /hurd/hello is a 'filesystem' that provides a single
file
<sjbalaji> So is it like hello is the mount moint for that hello server ?
<teythoon> sjbalaji: yes, kind of that, but in a more general sense
<sjbalaji> teythoon: How can I see the different servers that are running
in the system ? I tried top in the terminal but it returned cannot find
/proc/version
<teythoon> sjbalaji: so it seems your procfs is not running, try as root:
settrans /proc /hurd/procfs -c
<sjbalaji> teythoon: But how does one differentiate between a server and a
normal process ?
<teythoon> one does not
<teythoon> for a rule of thumb: anything from /hurd is a translator
<teythoon> you can view a nodes passive translator record using showtrans,
e.g. showtrans /dev/hd0
<sjbalaji> Is there something like a man page for translators ? Like how to
work with them or to figure out what services are offered by them ?
<teythoon> well, there is --help
<teythoon> also, go to /dev and /servers and look around using showtrans or
fsysopts
<sjbalaji> teythoon: What is the difference between a nodes active and
passive translator ?
<teythoon> a passive translator record is stored in the file system for the
node
<teythoon> if the node is accessed, and no translator is currently running,
it is started on demand
<teythoon> we call a running translator an active one
<sjbalaji> So the hello translator in the example is a passive one ?
<teythoon> if you used settrans foo /hurd/hello, a node foo is created with
an passive translator record
<teythoon> if you used settrans -a foo /hurd/hello, the translator is
started immediately
<sjbalaji> teythoon: What do you mean by a passive translator record ?
<teythoon> sjbalaji: it's an argv-vector encoded in the filesystem
(currently, only ext2 supports this)
<teythoon> in ext2, it is stored in a block and a os-specific field in the
inode points to that block
<sjbalaji> teythoon: I can't understand the logic behind that :(
<teythoon> this way, the servers are started on demand
<sjbalaji> But once they are invoked they are always online after that.
<teythoon> yes
<sjbalaji> I thought that the server goes down once its used
<gnu_srs> teythoon: shouldn't the passive ones time out if unused?
<teythoon> yes, that's how it was intented to be, but that has been
patched-out in debian/hurd
<gnu_srs> reason?
<teythoon> i don't know the details, but there is a race condition
(libports_stability.patch.)
IRC, freenode, #hurd, 2014-01-31
<sjbalaji> How can I see the complete control flow when I run the hello
translator example ?
IRC, freenode, #archhurd, 2014-02-05
<CalimeroTeknik> plus I discussed quickly that idea with Richard Stallman
and he told me translators had a conception flaw that would forbid such a
system to be usable
IRC, freenode, #archhurd, 2014-02-06
<antrik_> CalimeroTeknik: the "conceptal problem" rms told you about was
probably the simple issue that translators are always followed, even if
they are run by another user
<antrik> CalimeroTeknik: the conceptal problem is only in that the original
designers believed that would be safe, which it isn't. changing that
default policy (to be more like FUSE) wouldn't do much harm to the Hurd's
features, and it should be easy to do
<antrik> it's just nobody bothered so far, because it's not a big deal for
typical use cases
<antrik> rms isn't really in touch with Hurd development. he was made to
believe it was a fundamental issue by a former Hurd developer who got
carried away; and he didn't know enough to realise that it's really not a
big deal at all
Candidates for Google Summer of Code Project Ideas
Extend ls et al. for Translators
IRC, freenode, #hurd, 2014-02-08
<youpi> heh
<youpi> I was wondering what that incoming/ directory was in my home
<youpi> ls gave me hundreds of packages
<youpi> then I remembered I had /hurd/httpfs http://incoming.debian.org/ on
it :)
<cluck> if only there were an easy and automated way to make ls and file
managers (like dired!) aware of links, mounts and translators :)
<youpi> cluck: what do you mean by "awaree"?
<cluck> someting like: lrwxrwxrwx 1 foo foo 31 Aug 21 18:01
my_translator-23.0 -> ../some/fakefs /some_parameters*
<cluck> (yes, i realize it goes against some security practices but maybe
there could be a distinction like with soft/hard links that made it
opaque for some use cases)
Passive Translators
IRC, freenode, #hurd, 2014-02-12
<braunr> well don't expect rsync to save passive translator records ..
<braunr> i recommend you save either the entire disk image or the partition
<gg0> should i expect it from tar/cp ?
<braunr> no
<braunr> i'm not even sure dumpe2fs does it
<braunr> the only reliable way is to save the block device
<azeem> might be a worthwhile GSOC
<azeem> "implement Hurd specific features in GNU utilities"
<azeem> there were some patches floating around for some things in the past
IIRC
<antrik> azeem: the plan for supporting Hurd features in FS utilities was
exposing them as xattrs, like Roland's Linux patch did... cascardos once
did some work on this, but don't remember how far he got
<antrik> you are right though that it would make for a good GSoC project...
<antrik> of course, *some* utilities would still benefit from explicit Hurd
support -- most notably ls
<azeem> IIRC there were also ls patches at one point
<antrik> can't recall that... but maybe it was befor my time ;-)