Recent changes to bugs

DNS dependencies

Aaron Seth — Wed, 16 Apr 2003 01:13:01 -0000

We are too dependent on a functional DNS.

The Wiki RM module keeps lists of host communications
by hostname, but prolly should use IP address for all
internal needs.

The Node Daemon uses the hostname:

1) To determine the task ID of the process to launch.

2) For starting MPI jobs, only launching the process
if the current node is the "head node" of the job.
(This relies on the executed "mpirun" to correctly
distribute the job itself).

For TCP (sched --> node) communication, the scheduler
(Wiki RM module) could handle #1&2 above by simply
telling the appropriate nodes what their taskid is and
start or alloc command.

For Multicasting (sched --> nodes) communication,
handling #1 is complicated since each node is a
different ID, but each node gets the exact same
multicast message! In fact I'm not sure if there is a
better way to handle it than by hostname or IP address
as it is already #2 could be handled by a hybrid
approach: TCP to head node and multicast to the rest of
them.

...

If the scheduler internals used IP addresses instead of
hostnames, the only real gotcha would be how to handle
multi-homed hosts. But I think we would still want to
use DNS or /etc/hosts for "pretty printing" in user output.

XML data needs to quote bad chars

Aaron Seth — Wed, 16 Apr 2003 00:35:43 -0000

Our client <-> server communications needs to make sure
to quote XML "dangerous" characters, in particular '
and " because a lot of data is sent back inside tag
attributes. (Certainly frowned upon by most real xML
hackers, I'm sure).

This problem is most noticeable when the "checknode.sh"
script returns a message with the abbreviation [can't]
where the apostrophe screws up the quoting.

Feature or trojan?

Anonymous — Fri, 03 Jan 2003 20:54:59 -0000

I downloaded maui-3.2.5p1, and whenever I start maui,
it tries to connect to an IP address in Utah (63.90.74.90).

root@vmpwww maui]# netstat -ptuna | grep maui
tcp 5 0 129.59.197.18:57190
63.90.74.90:80 CLOSE_WAIT 23202/maui

This connection vanishes whenever maui is killed, and
reappears the instant it is restarted.

Is this some feature that I just don't know about, or
has maui been trojaned some how? I downloaded the
source tarball from the website, so if it's trojaned
there may be others with the same problem.

Mathew Binkley
Mathew.Binkley@vanderbilt.edu

thread deadlock in 1.3

Anonymous — Sat, 15 Jun 2002 01:59:03 -0000

It seems that the server and scheduler threads can
deadlock if client requests come too quickly. This
seems to happen mostly with multiple concurrent
mauisubmits. I think it's fixed in CVS. Need to test...

--
(_mauisched_, just not logged in 'cause I'm lazy)

Checkpoint data memory leak?

Aaron Seth — Mon, 22 Apr 2002 23:42:47 -0000

Serialized checkpoint data grows over time. Just try
the following MySQL commands every few days or so:

SELECT LENGTH(data) FROM checkpoint;

This could either be a memory leak in the scheduler
code or in the Java VM itself.

The VM I'm seeing this problem on Linux:

[root@dahui mauisched]# /usr/java/jdk/bin/java -version
java version "1.3.1_02"
Java(TM) 2 Runtime Environment, Standard Edition (build
1.3.1_02-b02)
Java HotSpot(TM) Client VM (build 1.3.1_02-b02, mixed mode)

Please chime in if you're seeing problems like this.

XML Parser doesn't like \0

Aaron Seth — Sun, 14 Apr 2002 00:13:53 -0000

XML Parser is looking for an EOF. Fix this wanky bug!

relying on setuid in FreeBSD fails

Tad Kollar — Fri, 12 Apr 2002 13:31:06 -0000

Maui version: 1.2
OS: FreeBSD 4.5

The setuid/setreuid calls in client_wrap.c have no
effect under FreeBSD (there's even a comment in there
to that effect). The scheduler is unusable since jobs
are submitted with the real uid of the user instead of
mauid (User=xxxx is not allowed to run Maui!).

The fix would probably be to have maui check the euid
instead of the uid. It looks like this may have been
addressed in the CVS, but a patch to 1.2 would be great.

ghost reservations when paused

Aaron Seth — Wed, 10 Apr 2002 23:27:45 -0000

Setting reservations or submitting jobs while the
scheduler is paused can make "ghost" reservations occur.

Suggested fix: refactor code to eliminate multiple
containers for data (which causes this kinda mismatch).
Basically have just one canonical resource tree from
which we can extract a list of jobs, reservations,
nodes, etc...

MD5.txt file overwrites previous!

Aaron Seth — Sun, 10 Mar 2002 02:23:58 -0000

The "MD5.txt" file that we're supplying for download
overwrites previous incarnations of itself! Thus we
need to remember to give it a unique releasename.

In addition, we may want to do the GPG thing.

deadlock in database glue code

Anonymous — Fri, 08 Mar 2002 23:42:05 -0000

The server and sched threads can deadlock over the
database object. Enforcing synchronization is prolly
unnecessary since MySQL does it already. Suggested fix
should be to remove synchronization primitives from the
glue code.