orte: Expand use of !orte_keep_fqdn_hostnames MCA parameter #2015
Conversation
|
Related to Issue #1614 but a more general solution. |
|
@rhc54 I don't know if you want to take a look at this PR before I merge it in. I'll let it sit here a for a little while, but it would be nice to get into 2.0.2 as it fixes a problem found on an internal test environment. |
|
scratching my head...it looks to me like this does exactly what happens when you set |
|
You made a comment in Issue #1614 about some environments where the difference between short versus long hostnames is important. The That being said, if the intention of So I'm good with just reusing I think I got all of the places in the code that needed to be modified to make this work. Certainly it worked for our use cases with LSF, rankfiles, hostfiles, and |
|
Hmmm...perhaps I should clarify that comment. What I was saying was that we don't support mixing short and long hostnames because we've had problems in the past with host confusion when we tried to do so. Thus, the param dictates that you use either all short names, or all long names. The user was asking us to support mixing the two, and I refused. 😄 |
|
Ah. In that case then:
|
|
Yep, that is correct! |
|
OK. I'll update the PR tomorrow to bring those options together. |
jjhursey
changed the title
|
@rhc54 I updated the PR with a new commit (I'll squash them together before merge). This should match what we discussed. |
|
Build Failed with XL compiler! Please review the log, and get in touch if you have questions. Gist: https://gist.github.com/c34c2ef0c667c7d2968dc69e75dfa48e |
|
Build Failed with GNU compiler! Please review the log, and get in touch if you have questions. Gist: https://gist.github.com/50fb5793fcad3937828ba301bf17ccd7 |
|
I always find it funny when the IBM CI fails on one of my PRs 😄 This is due to a temporary file system issue on the test machine. I've deactivated our CI until it comes back. |
|
👍 |
|
bot:ibm:retest |
* Expand the use of the `orte_keep_fqdn_hostnames` MCA parameter when
it is set to false.
* If that parameter is set to false (default) then short hostnames
(e.g., `node01`) will match with the long hostnames (e.g.,
`node01.mycluster.org`). This allows a user (or resource manager)
to mix the use of short and long hostnames.
- Note that this mechanism does _not_ perform a DNS lookup, but
instead strips off the FQDN by truncating the hostname string at
the first `.` character (when not an IP address).
- By default (`false`) the following is true:
`node01 == node01.mycluster.org == node01.bogus.com`
since we use `node01` as the hostname.
jjhursey
force-pushed
the
topic/mixed-hostnames
branch
from
cff8fae to
d26dd2c
Compare
* Related to c0038eded3544db94f68f3d5b58c89739834eb96 * See discussion on Open MPI community PR: - open-mpi#2015 * After broader discussion it was decided to expand the use of the !orte_keep_fqdn_hostnames MCA parameter to shorten all hostnames. This was exactly what the orte_use_mixed_hostnames MCA parameter was doing. - This also means that the LSF folks will get the behavior they want by default in Open MPI. * Upstream will see one commit that combines this commit and c0038eded3544db94f68f3d5b58c89739834eb96 (cherry picked from commit a33a2308ca80766fe6cf1f217b5a467687669603)
|
Hi, I know it is a way old thread here, but is there any way to specify orte_keep_fqdn_hostnames value globally in a system instead of doing it from the command line? |
|
Put the MCA param in the default param file where OMPI was installed. |
4 participants
orte_keep_fqdn_hostnamesMCA parameter when it is set to false.node01) will match with the long hostnames (e.g.,node01.mycluster.org). This allows a user (or resource manager) to mix the use of short and long hostnames..character (when not an IP address).false) the following is true:node01 == node01.mycluster.org == node01.bogus.comwe usenode01as the hostname.