My complaint is that the job doesn't able to process any data as we gave
maximum 75 minutes for a job to process, after every 75 minutes a new
job starts it is having the same batch of ids as the previous job and
this behaviour of the job continues untill we restarts the DUCC.
Last time when this happened, that time one of our machine out of three
was down and disconnected from the cluster.
The machine which was down having all the UIMA-AS services deployed on
it. Uima-AS services are used to process the data.
When the machine is down then uimaas services should allocate to the
other two machine, may be DUCC fails to allocate the uimaas services to
the other two machine due to which the job may not able to process data.
The problem still is with the job but i am trying to explain you every
possible thing that happened to my cluster when the problem arise to
make it easy for you to help me.
Are my problem clear to you now???
Thanks and Regards
Priyank Sharma
On Wednesday 15 November 2017 03:57 PM, Lou DeGenaro wrote:
Please note that we make a clear distinction between "services" and
"jobs". Earlier e-mail from you suggested that your trouble was related to
jobs.
Here is my understanding of you situation. You use ducc_submit to submit a
job comprising several work items. DUCC starts three Job Processes all on
the same machine and these are successfully processing work items. At some
point before all work items are completed the connection to that machine is
lost. And at this point the trouble for you begins...is this correct?
DUCC should detect that the lost contact machine is down, and if there is
space on other machine(s) it should allocate new Job Processes to continue
the work. However, the disconnected machine may continue processing any
work items is was working on prior to losing connectivity, so it is
possible that the same work items may have overlapping processing. Is
overlapping processing of the same work items your complaint?
Lou.
On Tue, Nov 14, 2017 at 11:00 PM, priyank sharma <[email protected]>
wrote:
server down mean one out of three machine is disconnected from the cluster
of three and all the services were deployed on the machine which was
disconnect from the cluster.
Thanks and Regards
Priyank Sharma
On Tuesday 14 November 2017 04:08 PM, Lou DeGenaro wrote:
What do you mean by "server down", precisely? Since we have no logs to
look at we can only go by your descriptions. We're trying to help...
Lou.
On Mon, Nov 13, 2017 at 11:30 PM, priyank sharma <
[email protected]>
wrote:
When our job goes into infinite-loop that time uima analysis engine did
not start and one of the server out of three were down that server has
all
the service which is being used by the uima analysis engine.
Is the server down creates this issue?
is memory the problem?
Thanks and Regards
Priyank Sharma
On Monday 13 November 2017 07:38 PM, Eddie Epstein wrote:
Several different issues here. There is no "job completion cap", rather
there is a limit on how long an individual work item will be allowed to
process before it is labeled a timeout. The default number of such
errors
+
exceptions before a Job is stopped is 15. Please increase this cap if
you
expect a work item to go longer.
If a job process runs out of heap space it should go OOM at which point
unpredictable things will happen. Do you see OOM exceptions in the JP
logfiles?
As for a bug, it is still hard to understand what is happening. Newer
versions of DUCC include a ducc_gather_logs command that collects DUCC
daemon logfiles and state and makes it more likely we can understand
what
is happening. No user application logfiles are included in the captured
tar
file.
Regards,
Eddie
On Mon, Nov 13, 2017 at 12:33 AM, priyank sharma <
[email protected]>
wrote:
Yes, i am using DUCC v2.0.1 i have a three node cluster with 32gb ram,
40gb ram and 28gb ram. Job runs fine for 15-20 days after that it goes
into
the infinite loop with the same batch of the id's. We have a 75 minutes
cap
for a job to complete if not then its start again so after every 75
minutes
new job starts but with the same id batch as previous and not even a
single
document ingested in the data store it goes in the same state untill we
restarts the server.
Is this because of the DUCC v2.0.1, are this version of DUCC having
that
bug?
Is this problem occur because of the Java Heap Space?
Please suggest something as there are nothing in the logs regarding to
my
problem.
Thanks and Regards
Priyank Sharma
On Friday 10 November 2017 09:00 PM, Eddie Epstein wrote:
Hi Priyank,
Looks like you are running DUCC v2.0.x. There are so many bugs fixed
in
subsequent versions, the latest being v2.2.1. Newer versions have a
ducc_update command that will upgrade an existing install, but given
all
the changes since v2.0.x I suggest a clean install.
Eddie
On Fri, Nov 10, 2017 at 12:11 AM, priyank sharma <
[email protected]>
wrote:
There is nothing on the work item page and performance page on the web
server. There is only one log file for the main node, no log files for
other two nodes. Ducc job processes not able to pick the data from
the
data
source and no UIMA aggregator is working for that batches.
Are the issue because of the java heap space? We are giving 4gb ram
to
the
job-process.
Attaching the Log file.
Thanks and Regards
Priyank Sharma
On Thursday 09 November 2017 04:33 PM, Lou DeGenaro wrote:
The first place to look is in your job's logs. Visit the ducc-mon
jobs
page ducchost:42133/jobs.jsp then click on the id of your job.
Examine
the
logs by clicking on each log file name looking for any revealing
information.
Feel free to post non-confidential snippets here, or If you'd like
to
chat
in real time we can use hipchat.
Lou.
On Thu, Nov 9, 2017 at 5:19 AM, priyank sharma <
[email protected]
wrote:
All!
I have a problem regarding DUCC cluster in which a job process gets
stuck
and keeps on processing the same batch again and again due to
maximum
duration the batch gets reason or extraordinary status
*"**CanceledByUser"
*and then gets restarted with the same ID's. This usually happens
after
15
to 20 days and goes away after restarting the ducc cluster. While
going
through the data store that is being used by CAS consumer to ingest
data,
the data regarding this batch does never get ingested. So most
probably
this data is not being processed.
How to check if this data is being processed or not?
Are the resources the issue and why it is being processed after
restarting
the cluster?
We have three nodes cluster with 32gb ram, 40gb ram and 28 gb ram.
--
Thanks and Regards
Priyank Sharma
