Closed garlick closed 5 years ago
Partially fixed with #1875, but we need to figure out if/how to make "nokz" default to close this one I think.
Just ran with the following change as suggested by @grondo in #1875, and tried running with and without -onokz=false
, and got the expected results. All the tests are passing, although the wreck tests are using t/rc/rc1-wreck
so they are not affected by this change to the installed rc1
file.
@trws seemed enthusiastic about it. Should we push this into our 0.11.0 release, and then make a 0.11.1 if we need to back it out or make other changes, after the full team is back in the new year?
I guess flux-submit.adoc
and flux-wreckrun.adoc
might need a little tweak to inform users how to defeat the default with -onokz=false
.
diff --git a/etc/rc1 b/etc/rc1
index b5872f0f..e5a4130a 100755
--- a/etc/rc1
+++ b/etc/rc1
@@ -15,6 +15,9 @@ flux module load -r 0 cron sync=hb
flux module load -r 0 userdb ${FLUX_USERDB_OPTIONS}
+# make flux wreckrun/submit -onokz the default (override w/ -onokz=false)
+flux wreck setopt nokz
+
wait $pids
core_dir=$(cd ${0%/*} && pwd -P)
Perhaps we could make the option to force use of kvs for output more user friendly, e.g. add a -o kz
option to force kz when it is disabled or something. . Let me know if you'd like me to make a pass at the man pages.
That seems like a good idea. Please feel free (or I'm happy to have a go and just have you review - how is typing these days?)
On Thu, Dec 27, 2018, 7:49 AM Mark Grondona <notifications@github.com wrote:
Perhaps we could make the option to force use of kvs for output more user friendly, e.g. add a -o kz option to force kz when it is disabled or something. . Let me know if you'd like me to make a pass at the man pages.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/flux-framework/flux-core/issues/1865#issuecomment-450176309, or mute the thread https://github.com/notifications/unsubscribe-auth/AAKX20n8EfamR5BZcw2eQFYGmk5TWPrAks5u9OwIgaJpZM4ZP6Fy .
Ok. We need this done today right? (I'm not going to have time tomorrow). I'll try to get something up that is at least a starting point, if not good enough.
Typing is fine, just don't have much stamina! ;-)
It's kind of an arbitrary deadline to enable us to move on, but I would like to get the release out this week. If you can put something together today that would be great, but if it becomes painful, LMK and @chu11 and I can finish the job.
As a workaround for poor KVS/content storage efficiency for some stdio patterns, implement a path for stdio and stderr that bypasses the KVS entirely.
(This is already a WIP, just wanted to capture it in an issue for 0.11.0 milestone)