Cylc 5.4.7 questions

[Annette Osprey]

Hi Hilary,

After a bit of a break from cylc we have just installed 5.4.7 and it
seems to be working fine. I just have a couple of questions.

1. I tried running the battery tests - however a few of them fail,
mainly with timeout errors. I was just wondering whether I should expect
these to succeed? I haven't had much time to work on cylc lately so
haven't really looked into it myself. I'm happy cylc works well enough
for our test cases which are remote submission/polling and interaction
with Rose, but I'd like to avoid any potential problems.

2. In the configuration settings, I understand that localhost is used as
the default for all hosts. In our case we want localhost to use pyro but
all other hosts to using polling, ideally without explicitly specifying
every host in the site.rc file. Is there a way to change the logic so
that would work? Obviously it's not a priority - there won't be that
many hosts in practice.

Thanks,
Annette

---

Test Summary Report
-------------------
./cylc-get-job-status/02-loadleveler.t (Wstat: 0 Tests: 2 Failed: 0)
Parse errors: No plan found in TAP output
./cycletime/01-iso_timeparser.t (Wstat: 0 Tests: 1 Failed: 1)
Failed test: 1
./cylc-get-global-config/00-basic.t (Wstat: 0 Tests: 9 Failed: 1)
Failed test: 9
./purge/00-purge.t (Wstat: 0 Tests: 2 Failed: 1)
Failed test: 2
./intercycle/01-future.t (Wstat: 0 Tests: 2 Failed: 1)
Failed test: 2
./restart/05-submit-failed.t (Wstat: 0 Tests: 13 Failed: 5)
Failed tests: 9-13
./satellite/00-basic.t (Wstat: 0 Tests: 2 Failed: 1)
Failed test: 2
./intercycle/02-asynch.t (Wstat: 0 Tests: 2 Failed: 1)
Failed test: 2
./restart/03-retrying.t (Wstat: 0 Tests: 13 Failed: 6)
Failed tests: 8-13
./restart/04-running.t (Wstat: 0 Tests: 13 Failed: 8)
Failed tests: 6-13
./restart/02-failed.t (Wstat: 0 Tests: 13 Failed: 8)


--

Annette Osprey
NCAS Computational Modelling Services
Department of Meteorology, University of Reading

+44 (0)118 378 5429 http://cms.ncas.ac.uk/

[Hilary Oliver]

Hi Annette,

1. There is not much in cylc that is platform-specific, and diagnosing test failures could be difficult for non-experts, so the automated tests are mainly for developers to check that code changes haven't broken anything. However, that information, and test battery usage is not too well documented, sorry (I'll improve this before the next release!). Additionally, we run 9 tests at a time, in parallel, by default, which is too many for most machines and will result some tests timing out (timeouts are used to catch unexpected errors of the sort that stop a test suite completing and shutting down automatically). This is almost certainly what you've experienced, in which case you've got nothing to worry about. That said, if you're concerned, you can reduce the amount of concurrency with the "-j N" option (passed through to "prove"; N=1 for sequential tests, which will be very slow). You can also run subsets of the tests ("cylc test-battery -j 2 SUBSET" for tests/SUBSET/); and you can run (most) individual tests directly with prove (e.g. "prove tests/combined/00-simple.t")

2. Ssh messaging reinvokes cylc messaging commands on the suite host, where the final communication between processes, to the suite daemon, is always via Pyro. So you can safely specify "task communiction method = ssh" under localhost as a default for all task hosts, to get the behaviour you want.

regards,
Hilary


_____
Dr Hilary James Oliver
National Institute of Water & Atmospheric Research
Email: hilary...@niwa.co.nz
Phone: +64 4 386 0461, +64 21 133 3318


________________________________________
From: cylc...@googlegroups.com [cylc...@googlegroups.com] on behalf of Annette Osprey [a.os...@reading.ac.uk]
Sent: Saturday, February 15, 2014 8:30 AM
To: cylc...@googlegroups.com
Subject: [cylc-dev] Cylc 5.4.7 questions

--

---
You received this message because you are subscribed to the Google Groups "cylc-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cylc-dev+u...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
--
Please consider the environment before printing this email.
NIWA is the trading name of the National Institute of Water & Atmospheric Research Ltd.