One of my nodes did not have /etc/environment pointing to the bin directory where the "erl" command was installed. When that node was configured to accept jobs on the master, the entire job would die.

It be much better if the master would just redirect to another node.

In some cases tracebacks from failed tasks are formatted incorrectly on the status page.

Filter doesn’t work correctly on the job status page.

In the git source, the homedisco example (at the bottom of util/homedisco.py) fails to run with the following error:

sqs2 ~/src/disco: python util/homedisco.py
[09/01/02 00:47:19 none ()] Received a new map job! [09/01/02 00:47:19 none ()] Done: 3 entries mapped in total [09/01/02 00:47:19 none ()] 0 chunk://localhost/homedisco@1230878839/map-chunk-0 [09/01/02 00:47:19 none ()] Received a new reduce job! **[09/01/02 00:47:19 none ()] Starting reduce connect_input(fname=chunkfile://data/homedisco@1230878839/map-chunk-0)
Traceback (most recent call last):
File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 39, in open_local

f = file(fname)

IOError: [Errno 2] No such file or directory: 'data/homedisco@1230878839/map-chunk-0'
None
**[09/01/02 00:47:19 none (chunkfile://data/homedisco@1230878839/map-chunk-0)] Can't access a local input file: chunkfile://data/homedisco@1230878839/map-chunk-0 Traceback (most recent call last):
File "util/homedisco.py", line 78, in

reduce = fun_reduce)

File "util/homedisco.py", line 44, in new_job

disco_worker.op_reduce(req)

File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 430, in op_reduce

fun_reduce(red_in.iter(), red_out, red_params)

File "util/homedisco.py", line 60, in fun_reduce

for k, v in iter:

File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 285, in multi_file_iterator

sze, fd = connect_input(fname)

File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 131, in connect_input

return open_local(input, local_file, is_chunk)

File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 50, in open_local

% input, input)

File "/Users/sqs/src/disco/node/disconode/disco_worker.py", line 39, in open_local

f = file(fname)

IOError: [Errno 2] No such file or directory: 'data/homedisco@1230878839/map-chunk-0'

It appears that the open_local path is incorrectly determining the filename from the chunkfile:// URI it is given. It does not prepend the value of the DISCO_ROOT environment variable as it should.

Result_iterator also tries to load the result from a relative path when it should be applying DISCO_ROOT to the beginning. It fails with this error if only the open_local issue is fixed:

[09/01/02 00:55:54 none ()] Received a new map job! [09/01/02 00:55:54 none ()] Done: 3 entries mapped in total [09/01/02 00:55:54 none ()] 0 chunk://localhost/homedisco@1230879354/map-chunk-0 [09/01/02 00:55:54 none ()] Received a new reduce job! [09/01/02 00:55:54 none ()] Starting reduce connect_input(fname=chunkfile://data/homedisco@1230879354/map-chunk-0)
[09/01/02 00:55:54 none ()] Reduce done: 3 entries reduced in total [09/01/02 00:55:54 none ()] Reduce done [09/01/02 00:55:54 none ()] 0 disco://localhost/homedisco@1230879354/reduce-disco-0 ['file://data/homedisco@1230879354/reduce-disco-0'] Traceback (most recent call last):
File "util/homedisco.py", line 80, in

for k, v in result_iterator(res):

File "build/bdist.macosx-10.5-i386/egg/disco/core.py", line 261, in result_iterator IOError: [Errno 2] No such file or directory: 'data/homedisco@1230879354/reduce-disco-0'

After applying this patch, the correct output is returned:

sqs2 ~/src/disco: python util/homedisco.py
[09/01/02 00:57:57 none ()] Received a new map job! [09/01/02 00:57:57 none ()] Done: 3 entries mapped in total [09/01/02 00:57:57 none ()] 0 chunk://localhost/homedisco@1230879477/map-chunk-0 [09/01/02 00:57:57 none ()] Received a new reduce job! [09/01/02 00:57:57 none ()] Starting reduce [09/01/02 00:57:57 none ()] Reduce done: 3 entries reduced in total [09/01/02 00:57:57 none ()] Reduce done [09/01/02 00:57:57 none ()] 0 disco://localhost/homedisco@1230879477/reduce-disco-0 KEY red:dog VALUE dog
KEY red:cat VALUE cat
KEY red:possum VALUE possum

The patch also fixes the problem for a custom HomeDisco job I wrote, but there's no test suite for me to determine whether it is correct in all cases. Specifically, it does not appear to introduces issues when running remote jobs (i.e., not through HomeDisco), but I can't guarantee anything. Also, there may be a better way of doing this. (I saw that the LOCAL_PATH env var exists, but it already has "/data" at the end, and the filenames we are appending to $DISCO_ROOT have "/data" at the beginning, so using LOCAL_PATH would result in an incorrect "/data/data".)

Instead of having status & configuration on the same webpage, it would be useful to separate them into 2 separate pages. The configuration can then be password protected and accessible only by the cluster administrator.

balls.png is missing from the git repository master/www/ directory

A few scripts use #!/usr/bin/python (i.e. disco-worker) instead of #!/usr/bin/env python. When the scripts are executed directly, non-system python installations cannot be used.

'nr_reduces = 5000' fails due to the map task trying to merge 5000 files together. This may also happen with many input files, in which case nr_reduces = nr_maps / 2.

22:54 < neal> hi
22:55 < neal> when i try to run homedisco i get the following error
22:55 < neal> Traceback (most recent call last):
22:55 < neal> File "wordcount.py", line 1, in
22:55 < neal> from homedisco import HomeDisco
22:55 < neal> File "/home/neal/school/seng474/frequent_itemsets/cfim/homedisco.py", line 4, in

           <module>

22:55 < neal> from disconode import disco_worker
22:55 < neal> File "/usr/local/lib/python2.6/dist-packages/disconode/disco_worker.py", line 504, in

           <module>

22:55 < neal> init()
22:55 < neal> File "/usr/local/lib/python2.6/dist-packages/disconode/disco_worker.py", line 64, in

           init

22:55 < neal> OOB_URL = ("http://%s/disco/ctrl/oob_get?" % this_master())<br/> 22:55 < neal> File "/usr/local/lib/python2.6/dist-packages/disconode/disco_worker.py", line 46, in

           this_master

22:55 < neal_> return sys.argv[4].split("/")[2]

Lighttpd has the default setting

server.max-write-idle = 360

which closes the connection after 360 seconds of client inactivity. The default value is too low for many reduce tasks that keep many connections open simultaneously.

Error while trying out the experimental deb-packages on a fresh Deb 5 VM. There is no /
etc/disco/disco.conf found anywhere. See

http://groups.google.com/group/disco-dev/browse_thread/thread/1ee969fb18aa61d3?hl=en

We have commented out lines of host names and IPs in our /etc/hosts that we'd rather not remove at this time. However, make-lighttpd-proxyconf.py still parses those old and invalid lines. Modifying the script to the code below fixes the problem.

Thanks!

    #!/usr/bin/python
    import os, re

    port = os.environ["DISCO_PORT"]
    print "proxy.server = ("
    r = re.compile("^(\d+\.\d+\.\d+\.\d+)\s+(.*)", re.MULTILINE)
    for x in re.finditer(r, file("/etc/hosts").read()):
            ip, host = x.groups()
            print '"/disco/node/%s/" => (("host" => "%s", "port" => %s)),' %\
                    (host, ip, port)
    print ")"

Investigate if dis.dis() disassembler could be used to infer dependant modules automatically for the m/r functions. See

http://groups.google.com/group/disco-dev/browse_thread/thread/7e63e6da57bcd05b?hl=en

Disco can copy required files and libraries to nodes when using the external interface. It should be possible to use this feature also with native Python code so that simple modules etc. could packaged in a job request and distributed automatically to nodes.

archwild reported on the IRC channel that Java processes lauched via the external interface sometimes don't die when the job is killed. They don't seem to die with pkill either.

Test Java with external interface.

archwild reported on IRC that a Java process can cause Erlang to consume 100% of CPU, if a misbehaving external process produces masses of output to stderr.

This should be easy to fix by applying rate limit to all types of messages, not just .

If a node dies (e.g. "ERROR: Worker crashed in map:51 @ disco-21: noconnection") when the job is running, the job isn't rescheduled properly to another node.

A misplaced "print" in a job can cause disco_worker.erl consume loads of CPU and memory. Apply rate limit or limit the buffer size for any output from processes.

When using base64 encoded values, external sort seems to fail and reports: "Could not match all bytes"

The current scheduler is suboptimal when the cluster is fully loaded. Whenever a task finishes on any node, the next task is scheduled on its place, regardless where its input data is located.

A node-specific task queue would solve this problem by pre-sorting tasks to nodes according to locations of input files.

when nr_maps < len(input), the web ui shows a negative number of waiters in the job info box

On macosx, disco-worker gets a master_url in sys.argv[4] that on macosx does not use localhost if you're using a single host but rather the name part of 'foo.local' (ie 'foo'), the alternative host naming scheme. Since this doesn't resolve disco fails with "WARN [map:0] Failed to get http://foo:7000/disco/master/_disco_4441/2c/wordcount@1251260554//params: (2, 'Temporary failure in name resolution')"

To fix I hack node/disco-worker with:
master_url = master_url.replace('foo', 'localhost')

If you use the sample count_words.py code in the documentation and you wrap the new job call in a loop (sleeping 30 seconds after completion), then the 2nd call will always fail (at least on osx) with a socket.error: (54, 'Connection reset by peer')

If you don't use the loop and call the script manually repeatedly, no such error occurs. Seems to me that disco is not closing connections properly after finishing a job.

While experimenting with 0.2.3, I found I couldn't run both

disco master start

and

disco worker start

at the same time. In particular, when starting the worker, I got a message to the effect of "lighttpd is already running with worker config". However, the only lighttpd process running was that with the master config file.

I also noticed that the MASTER_DISCO_PORT config parameter has been removed in 0.2.3, which I suspect is related. Does the master automatically start a worker as well? Or is there some other way of running both a master and slave on the same node?

External sort uses the LC* environment variables to determine sort order. To guarantee consistent results, LC_ALL=C should be set for sort.

OK, I have sovled this problem

Hi, I have configure a cluster with two computers, xiliu-fedora (master) and xiliu-public (worker). Both are running Fedora 11, python 2.6.3. However, I did alway get following timeout errors, and on the worker node, erl is crash, and throws erl_crash.dump

2009/10/08 19:20:36master

map:0 added to waitlist

2009/10/08 19:20:36xiliu-public

WARN: [map:0] Node failure: "Couldn't connect to xiliu-public (timeout). Node blacklisted temporarily."

2009/10/08 19:20:04master

map:0 assigned to xiliu-public

2009/10/08 19:20:04master

map:0 added to waitlist

2009/10/08 19:20:04master

Map phase

2009/10/08 19:20:04master

Starting job

2009/10/08 19:20:04master

New job!

[xiliu@xiliu-public ~]$ tail -f erl_crash.dump timeout
infinity
fun
'' '$end_of_table' 'nonode@nohost' '_' true
false
=end ^C

When running tests map tasks execute ok even if lighttpd is down on a node. However, reduce fails since it can't access the map results. At least with comm_httplib this results to a cryptic error message - at least it could say nicely what's wrong.

It should be possible to combine various low-level readers, such as gzip decoding, with higher level custom readers. This requires that you can specify a list (stack) of readers / writers instead of just one. See

http://groups.google.com/group/disco-dev/browse_thread/thread/7e63e6da57bcd05b?hl=en

Merge Eran Sandler's patches to add support for Gzipped inputs. See

http://groups.google.com/group/disco-dev/browse_thread/thread/7e63e6da57bcd05b?hl=en

http://groups.google.com/group/disco-dev/browse_thread/thread/8c0970c82587fa5d?hl=en

Job fails fatally due to a glusterfs error (timeout), even though the error should be recoverable.

Gluster logs (client):

[2009-08-12 15:32:09] E [client-protocol.c:437:client_ping_timer_expired] dx26-vol1: Server 172.16.1.27:9900 has not responded in the last 60 seconds, disconnecting.

Server:
[2009-08-12 18:44:41] E [server-protocol.c:3903:server_readv] vol1: invalid argument: state->fd [2009-08-12 18:44:41] W [server-protocol.c:4210:server_fstat] server: fd - 1: unresolved fd [2009-08-12 18:44:41] E [server-protocol.c:5763:server_finodelk] server: fd - 1: unresolved fd

Disco:

Worker failed. Last words:
Traceback (most recent call last):
File "/usr/bin/disco-worker", line 69, in

method(m)

File "/var/lib/python-support/python2.5/disconode/disco_worker.py", line 392, in op_map

run_map(job_input[0], partitions, map_params)

File "/var/lib/python-support/python2.5/disconode/disco_worker.py", line 327, in run_map

for entry in reader:

File "/home/tuulos/src/disco/pydisco/disco/func.py", line 60, in netstr_reader File "/home/tuulos/src/disco/pydisco/disco/func.py", line 40, in read_netstr IOError: [Errno 107] Transport endpoint is not connected
close failed: [Errno 107] Transport endpoint is not connected

If Glusterfs fails on a single node, it is detected only when results are being moved to resultfs, which leads to a fatal error. However, the error should be recoverable by executing the task on another node.

Disco 0.2.3 depends on erlang-base 13b, which Debian Lenny doesn't supply. The image "ami-e69d798f" is a lenny install, so it doesn't have the required packages.

Furthermore, setup-instances.py overrides /etc/apt/sources.list to only use discoproject and lenny repos, so it reinforces the same issue.

Running Disco 0.2.3 source distribution on ami-2946a740 (Debian Squeeze AMI from http://alestic.com/), with one master node and one remote slave, I was unable to run a job successfully. I went through all of the steps on the Troubleshooting Disco page, and found that the slave:start test failed with an {error, timeout} result. I tried a number of different flavors of node names (ec2 internal name, short names defined in /etc/hosts), and met with the same results. I was successfully able to run slave:start on the slave node to start a slave locally, but the same command from the master failed.

http://discoproject.org/doc/start/troubleshoot.html#troubleshooting contains suggestions that don't work for 0.2.3. For example, erl -boot master/disco cannot be done. I'm trying to figure out why I cannot start the master on a Linux box

/usr/local/lib/erlang/erts-5.7.2/bin/epmd -daemon This one caused me a headache as I wasn't able to restart the master. The epmd process was still running and prevented master from starting up.

Integrate MochiWeb as a default web server to Disco, to make setting up the system easier. Lighttpd should be still provided as a high-performace option (Lighty could be supervised by Disco).

Master should store job records persistently, so master could be restarted without losing any records.

You can easily make the worker run out of file descriptors e.g. by setting nr_reduces = 1000. This may result e.g. to spurious DNS failures, since no new sockets can be opened.

Disco should limit the number of fds used in a worker process, at least to make the error messages more descriptive.

The bin/disco script could include a troubleshoot mode which executes steps at

http://discoproject.org/doc/start/troubleshoot.html

automatically and shows suggestions how to fix detected issues.

If a task fails due to e.g. a temporarily unavailable input, it should wait for some time before re-trying. Otherwise it can easily reach the maximum number of failed tasks in a few seconds.

Adding new lines to the config table seems to drop old lines sometimes. Correspondingly, adding new nodes to the cluster on the fly doesn't always produce expected results.

Raising the number of reduces to thousands is problematic because it requires a huge number of open file descriptors per process.

If a job leaks memory / consumes vast amounts of memory otherwise, it can take nodes down easily. Master should monitor per-task/job/node memory consumption and kill rogue processes.

disco-worker catches exceptions with the following code:

    try:
            run(method, mode, part, m)
    except comm.CommException, x:
            util.data_err("HTTP error: %s" % x)
    except IOError, x:
            util.data_err("IO error: %s" % x)

The calls to util.data_err fail because it requires 2 parameters.

If hundreds (>400) of jobs are running (and finishing) concurrently, gen_server:call(disco_server) calls can start to timeout.

On FreeBSD 7.2, the inet:gethostname() call returns the short hostname without the domain, e.g., "example" instead of "example.com." This causes the master node to fail with the error:

Error{EXIT,{{badmatch,{error,nxdomain}},[{handle_job,gethostname,0},{handle_job,set_disco_url,2},{handle_ctrl,handle,2},{scgi_server,handle_request,2},{scgi_server,scgi_worker,1}]}}

The solution is to patch handle_job.erl to account for incomplete hostnames and append the domain when it isn't present.

--- master/src/handle_job.erl.orig  2009-12-05 22:09:09.811589122 -0500
+++ master/src/handle_job.erl   2009-12-05 21:54:09.920556045 -0500
@@ -114,10 +114,14 @@
         end.
 
 gethostname() ->
-        {ok, Hostname} = inet:gethostname(),
-        {ok, Hostent}  = inet:gethostbyname(Hostname),
+        {ok, Host} = inet:gethostname(),
+        case inet_db:res_option(domain) of
+             "" -> Hostname = Host; 
+             Domain -> Hostname = Host ++ "." ++ Domain
+        end,
+        {ok, Hostent}  = inet:gethostbyname(Hostname),
         Hostent#hostent.h_name.
-        
+
 set_disco_url(undefined, Msg) ->
         {value, {_, SPort}} =
                 lists:keysearch(<<"SERVER_PORT">>, 1, Msg),

This code was taken from http://web.mit.edu/erlang_vR11B-5/arch/i386_linux26/lib/erlang/lib/kernel-2.11.5/src/net_adm.erl

Basically, one part passes a unicode string to setopt in pycurl. I recommend removing pycurl as dependency for disco. Traceback is below:

URL: u'http://localhost:8989/localhost/77/disco_tut@1260921574/reduce-index.txt'
Traceback (most recent call last):
  File "count_words.py", line 24, in <module>
    for word, total in result_iterator(results):
  File "...disco-0.2.3/pydisco/disco/core.py", line 466, in result_iterator
    sze, fd = comm.open_remote(url)
  File ".../disco-0.2.3/pydisco/disco/comm_curl.py", line 149, in open_remote
    conn = CurlConn(url, handle = c, expect = expect)
  File ".../disco-0.2.3/pydisco/disco/comm_curl.py", line 66, in __init__
    self.init_handle(url)
  File ".../disco-0.2.3/pydisco/disco/comm_curl.py", line 89, in init_handle
    self.handle.setopt(URL, url)
TypeError: invalid arguments to setopt