Re: [Taskforest-discuss] Looking for TaskForest USERS!!!
Brought to you by:
enoor
Menu
▾
▴
Re: [Taskforest-discuss] Looking for TaskForest USERS!!!
|
From: Aijaz A. <en...@us...> - 2009-04-18 18:48:28
|
[This email has some ASCII diagrams and is best read in a fixed-width font]
Lana:
I've been thinking about your email and the four (5, depending on how you
count them) great questions you raised:
1) Handling the midnight point
2) Calendars
3) Dependencies between Families
Dependencies between Families with different schedules
4) Backward Compatibility
In this email, I'll only focus on the Midnight point issue, because this
will be a very long email. I'll address the other issues in a subsequent
email.
THE SUMMARY OF EVERYTHING THAT FOLLOWS:
You can resolve the apparent problem where jobs need to run on both sides
of midnight by changing the timezone of the Family. This will only work
if all jobs in the family run within a 24 hour period (ignoring Daylight
Savings Time, when we have 23 hours in a day once a year).
THE DETAILS:
Having a midnight point gives us a convenient time interval within which
a run of a job is considered valid. For example, with the system as it is
now, saying "job J1 has run" means that it has run *today*. If J2 depends
on J1, we only have to look for successful runs of J1 today.
If we don't have a known time interval, then we have to resolve the
problem that occurs when jobs run for longer than 24 hours. Let's start
with a simple example first. Assume that our Family file looks like this:
+--------------------------------------------------------------------------------
|start => '00:00', tz => 'America/Chicago', days => 'Mon,Tue,Wed,Thu,Fri'
|
| J1( start=>22:00 )
| J2()
+--------------------------------------------------------------------------------
Example 1: J1's run time is 5 hours. This is what happens today
(Chicago)
Day Time Action(s)
Mon 00:00
01:00
02:00
03:00
04:00
05:00
06:00
07:00
08:00
09:00
11:00
12:00
13:00
14:00
15:00
16:00
17:00
18:00
19:00
20:00
21:00
22:00 J1 Starts
23:00 |
Tue 00:00 |
01:00 |
02:00 v
03:00 J1 Ends
04:00
05:00
06:00
07:00
08:00
09:00
11:00
12:00
13:00
14:00
15:00
16:00
17:00
18:00
19:00
20:00
21:00
22:00 J1 Starts
23:00 |
Wed 00:00 |
01:00 |
02:00 v
03:00 J1 Ends
04:00
So you can see that J2 never gets a chance to run! At 03:00 on Tuesday,
the system checks to see if J1 has run for that day, and it hasn't. So J2
can't run for that day.
Now, I could remedy this particular case with the following technique (or
hack):
Example 2: J1's run time is 5 hours. J2's run time is 1 hour.
Change the Family File to say:
+--------------------------------------------------------------------------------
| # changed the start time, time zone and days of week
|start => '00:00', tz => 'GMT', days => 'Tue,Wed,Thu,Fri,Sat'
|
| J1( start=>03:00 )
| J2()
+--------------------------------------------------------------------------------
What this does is *shift* the entire Family so that all jobs in it run
on the same day in the time zone specified. My crontab entry that invokes
taskforest does not need to be changed. That can stay the same. (Reply
to this email if it's not clear why)
Now, with the system we have today, this is what happens:
(Chicago) (GMT)
Day Time Day Time Action(s)
Mon 00:00 Mon 05:00
01:00 06:00
02:00 07:00
03:00 08:00
04:00 09:00
05:00 11:00
06:00 12:00
07:00 13:00
08:00 14:00
09:00 15:00
11:00 16:00
12:00 17:00
13:00 18:00
14:00 19:00
15:00 20:00
16:00 21:00
17:00 22:00
18:00 23:00
19:00 Tue 00:00
20:00 01:00
21:00 02:00
22:00 03:00 J1 Starts
23:00 04:00 |
Tue 00:00 05:00 |
01:00 06:00 |
02:00 07:00 v
03:00 08:00 J1 Ends, J2 Starts
04:00 09:00 J2 Ends
05:00 11:00
06:00 12:00
07:00 13:00
08:00 14:00
09:00 15:00
11:00 16:00
12:00 17:00
13:00 18:00
14:00 19:00
15:00 20:00
16:00 21:00
17:00 22:00
18:00 23:00
19:00 Wed 00:00
20:00 01:00
21:00 02:00
22:00 03:00 J1 Starts
23:00 04:00 |
Wed 00:00 05:00 |
01:00 06:00 |
02:00 07:00 v
03:00 08:00 J1 Ends, J2 Starts
04:00 09:00 J2 Ends
05:00 10:00
Shifting the start/end of the days by changing the timezones has fixed
this problem. J1 still starts at 22:00 Chicago time, and J2 will still
run if J1 runs successfully.
Lana, if the jobs in your Family files run in less than 24 hours, you
can use this method now.
However, this is not a universal solution. What happens if J1's run time
is 26 hours? Even with the family as shown in Example 2, J1 ends the day
after it started. This means that J2 will never run, no matter what we set
the timezone to be. This is because no matter what day it is, *today's*
run of J1 will never have completed that day.
So the current version of taskforest does not support Family files whose
jobs run for more than 24 hours from the start of the first job to the end
of the last job, even if each individual job runs for less than 24 hours.
Example 3: J1 runs for 25 hours
(Chicago) (GMT)
Day Time Day Time Action(s)
Mon 00:00 Mon 05:00
01:00 06:00
02:00 07:00
03:00 08:00
04:00 09:00
05:00 11:00
06:00 12:00
07:00 13:00
08:00 14:00
09:00 15:00
11:00 16:00
12:00 17:00
13:00 18:00
14:00 19:00
15:00 20:00
16:00 21:00
17:00 22:00
18:00 23:00
19:00 Tue 00:00
20:00 01:00
21:00 02:00
22:00 03:00 J1 Starts
23:00 04:00 |
Tue 00:00 05:00 |
01:00 06:00 |
02:00 07:00 |
03:00 08:00 |
04:00 09:00 |
05:00 11:00 |
06:00 12:00 |
07:00 13:00 |
08:00 14:00 |
09:00 15:00 |
11:00 16:00 |
12:00 17:00 |
13:00 18:00 |
14:00 19:00 |
15:00 20:00 |
16:00 21:00 |
17:00 22:00 |
18:00 23:00 |
19:00 Wed 00:00 |
20:00 01:00 |
21:00 02:00 |
22:00 03:00 | J1 Starts
23:00 04:00 v |
Wed 00:00 05:00 J1 Ends |
01:00 06:00 |
02:00 07:00 |
03:00 08:00 |
04:00 09:00 |
05:00 10:00 |
The other thing we're seeing here is that now we have 2 instances of J1
running simultaneously (for 2 hours), even though that is not the intent
of the Family file. This could be a major problem. J2 never runs.
I can't think of a foolproof way to resolve this.
If we implement rules that require a family to 'complete' all of the
previous day's jobs until it can run no more - because all jobs have
completed successfully. I don't know if this is a satisfactory approach
or not. This means that J1's start times will be as follows:
Mon: 22:00
Wed: 00:00
On Tuesday we don't run the Family, because Monday's run hasn't completed.
That's fine. But what about Wednesday? With this definition, Wednesday
will never run as well because Tuesday's run never completed (there wasn't
a run on Tuesday). So the system has to be smart about that and realize
that. And that's fine, until you change the family on Tuesday to run a
new job J3 after J2. Then, the system needs to know on Wednesday not to
wait for look for a successful run of J3 for Monday's run (because J3
didn't exist on Monday). Which means that the system needs to know your
change control history. Which is a whole other problem.
Do you folks see any elegant way to resolve this?
Aijaz
On Sat, Apr 18, 2009 at 12:15 PM, Aijaz Ansari
<en...@us...>wrote:
> Forwarded to the Taskforest-discuss mailing list with the original author's
> permission:
>
> ---------- Forwarded message ----------
> From: Aijaz Ansari <en...@us...>
> Date: Fri, Apr 17, 2009 at 1:48 PM
> Subject: Re: [Taskforest-discuss] Looking for TaskForest USERS!!!
> To: Svetlana Lemeshov <lem...@mi...>
>
>
> Lana:
>
> Thanks for your email. These are excellent points. I'll give you a quick
> response now (as I'll be leaving my desk soon), and a more detailed one
> tomorrow.
>
> In a nutshell -
>
> a) You're right about the midnight issue. I had some ideas about it, and
> was discussing it with a user, but haven't heard back from him. I'd love to
> discuss this with you. I'll send you a more detailed response about this
> tonight.
>
> b) This brings us back to another of your questions. Yes, the system
> changes quite often, but every effort has been taken to keep version n+1
> backward compatible with version n. The only exception I can think of is
> the --collapse option, which determines how status is displayed. I try to
> ensure that new development does not break any of the old test cases.
>
> c) Arbitrary job calendars is in the works. I know how it's going to be
> implemented and am waiting to complete the current version that deals with
> tokens before working on calendars. The main feature timetable is as
> follows: Tokens (currently in dev), Calendars, and then support for multiple
> machines. The multiple machine support might be version 2.0.
>
> d) Dependencies between families - honestly, I hadn't even thought of
> that. Your examples make sense, but there are some important design
> decisions that will need to be made - where each option has it's own
> merits. Because of issues like this, I prefer not making a decision before
> speaking to users about it. Their needs may cause one option to outweigh
> the other.
>
> e) I'm taking the development a little slower than I'd like because I
> really want to speak to users before making major design decisions. I use
> TaskForest in Production every day on multiple machines, and it meets my
> needs. I won't use some of the new features as much as I use the older
> ones, so I consider the input from users such as yourself (who rely on these
> features) as having more weight than my own input. Unfortunately, even
> though TaskForest is apparently being used by a lot of people, you are only
> the third person to email me about it in the last 12 months. So getting a
> dialog with users is kinda tough :)
>
> I'll send you some a more detailed email tomorrow morning. In the
> meantime, is it OK if I forward your email, and this response and future
> responses to the mailing list? It might help other users decide whether or
> not TaskForest is appropriate for them. If not, that's cool too.
>
> Thanks,
>
> Aijaz.
>
>
>
> On Fri, Apr 17, 2009 at 12:45 PM, Svetlana Lemeshov <lem...@mi...>wrote:
>
>> Hello Aijaz,
>>
>> Thank you very much for promptly responding to my inquiry!
>>
>> I would love to hear feedback from your users.
>>
>> Meanwhile, if you don't mind, I have a couple of questions for you
>> since I couldn't find any forum where I can post my questions and have
>> them
>> answered by other users. In addition, I thought you might be interested
>> to hear the feedback from the potential users of your product.
>>
>> Here is some background information:
>> I am a Data Warehouse developer/analyst designated to explore and analyze
>> various scheduling products available on the market today with the goal of
>> migrating our jobs to a more flexible scheduler. I've installed TaskForest
>> and spent some time exploring its features trying to figure out how well
>> it can handle various DW specific needs.
>>
>> MIT Data warehouse consists of data coming from the various source
>> systems:
>> Hr, Financial, Libraries, Student Systems, Facilities, etc.
>>
>> Our data gets refreshed nightly, starting after 10:00PM and finishing
>> around
>> 7:30 AM. The refresh consists of numerous extraction and load processes
>> and has
>> many inter--dependencies between them.
>>
>> For instance, many financial loads depend on HR loads, etc.
>>
>> While exploring TaskForest I stumbled on the following obvious
>> limitations:
>>
>> - Handling a 'midnight' point;
>> I restart taskforest at 00:01 from cron file, but since our loads run all
>> night
>> long, we need to come up with some custom scripts to handle scripts that
>> run
>> 'across' midnight point.
>> Specifically, the situation when a job that runs after midnight depends on
>> the
>> status of the one that finished before the midnight.
>>
>> - TaskForest Schedule based only on the days of the week;
>> Our current cron utility allows for various scheduling options (by months,
>> day of the month, etc.) Since TaskForest is only Days-of-the-Week driven,
>> and some of our jobs happen to run only on certain days of the month
>> (first of each month, or once in a quarter), we will have to handle this
>> with the custom scripts.
>>
>> - Absence of dependencies between the Families;
>> There is no mechanism that would allow to have a Family within a Family to
>> handle Jobs that obviously belong to the same family, but have slightly
>> different execution schedules.
>>
>> Sometimes, we can have a situation when jobs that obviously belong to the
>> same
>> family (same subject area, lots of dependencies between them) run on a
>> different days of the week (Ex. Job A runs everyday, but it depends on the
>> successful completion of Job B which runs only on weekdays).
>>
>> If I understand it correctly, in current TaskForest setting, this scenario
>> calls
>> for 2 different families. However, it seems that there is no mechanism to
>> establish dependencies between 2 families. So, it should be handled
>> through
>> custom scripts like many other things (handling Holidays, etc)
>>
>> Overall, it seems that our application will require a lot of custom
>> scripts to
>> help TaskForest handle our jobs.
>>
>> One more observation. It seems that the product is still evolving and
>> every time
>> I check there is a new version of it. Since I started working
>> with the tool a few months ago, I've already upgraded it once, but I am
>> still
>> about 2 versions behind. I am concerned about the on-going maintenance and
>> support of the product and possible frequent modifications to the
>> scheduler
>> once it is implemented in production (example: if/when a new feature like
>> calendar is implemented, it might require some modifications on our site.)
>>
>> Aijaz, I am sorry for my lengthy rambling, but these are some of the
>> issues I am
>> running into. I am not sure whether it is due to the fact that I just fail
>> to
>> use the product correctly, or these are real limitations.
>>
>> I was hoping to discuss these issues with other users, but, on the other
>> hand,
>> who can address these points better than the author of TaskForest?
>>
>> I am giving a TaskForest presentation to my team on Wednesday, April 22
>> and will be covering many of my 'findings'.
>>
>> Aijaz, it would be really very helpful if you could clarify some of the
>> points
>> mentioned above! I appreciate your time and attention!
>>
>> Thank you very much!
>>
>> Lana Lemeshov
>>
>>
>>
>>
>>
>> Quoting Aijaz Ansari <en...@us...>:
>>
>> Svetlana:
>>>
>>> I'm the author of TaskForest. Since I'm obviously biased, I won't give
>>> you
>>> my unsolicited opinion :), but feel free to ask me any questions - I'll
>>> be
>>> glad to answer them. I'm also taking feature requests. Many of the
>>> recent
>>> features are a result of user emails.
>>>
>>> ...
>>>
>>>
>>> Thanks,
>>>
>>> Aijaz.
>>>
>>> On Thu, Apr 16, 2009 at 3:53 PM, Svetlana Lemeshov <lem...@mi...>
>>> wrote:
>>>
>>> Hello,
>>>>
>>>> My name is Lana Lemeshov, I work for MIT DataWarehouse.
>>>> We are currently using UNIX cron utility for job scheduling, however
>>>> we would like to switch to a more flexible product.
>>>>
>>>> We are looking into a couple of schedulers and TaskForest is one of
>>>> the strong contenders (with Oracle 10g Scheduler as an alternative.)
>>>>
>>>> I am trying to find someone who has any experience using TaskForest and
>>>> wouldn't mind sharing the feedback us.
>>>> MIT Data Warehouse would love to hear honest reviews of the product
>>>> from the user's point of view: specific problems, limitations, support
>>>> issues,
>>>> etc.
>>>>
>>>> We are interested to compare how TaskForest stands against the Oracle
>>>> Sheduler.
>>>>
>>>> ANY feedback is greatly appreciated.
>>>>
>>>> Thank you very much!
>>>>
>>>> Lana
>>>>
>>>>
>>>>
>>>>
>>>> ------------------------------------------------------------------------------
>>>> Stay on top of everything new and different, both inside and
>>>> around Java (TM) technology - register by April 22, and save
>>>> $200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco.
>>>> 300 plus technical and hands-on sessions. Register today.
>>>> Use priority code J9JMT32. http://p.sf.net/sfu/p
>>>> _______________________________________________
>>>> Taskforest-discuss mailing list
>>>> Tas...@li...
>>>> https://lists.sourceforge.net/lists/listinfo/taskforest-discuss
>>>>
>>>>
>>>
>>
>>
>
>
|
