osc/pt2pt: fix infinite frag allocation loop

[hjelmn]

Signed-off-by: Nathan Hjelm [email protected]
(cherry picked from commit 12b52b2)
Signed-off-by: Nathan Hjelm [email protected]

[ibm-ompi]

The IBM CI (GNU Compiler) build failed! Please review the log, linked below.

Gist: https://gist.github.com/51b6e8e91673586a73efff49b7b3f3de

[jjhursey]

The IBM CI failure is valid, but not related to this PR. @rhc54 it's a failure in PMIx (stack below)

#0  0x00003fff7c99593c in pmix_server_init ()
   from /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/libopen-rte.so.0
#1  0x00003fff7c296630 in rte_init () from /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/openmpi/mca_ess_hnp.so
#2  0x00003fff7c95121c in orte_init () from /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/libopen-rte.so.0
#3  0x00003fff7c98ed84 in orte_submit_init ()
   from /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/libopen-rte.so.0
#4  0x00000000100012e8 in orterun (argc=7, argv=0x3fffff3a0328) at orterun.c:133
#5  0x0000000010000fc0 in main (argc=7, argv=0x3fffff3a0328) at main.c:13

It looks like this is just on the v3.x branch (heads up @bwbarrett @hppritcha ). I'm going to disable the IBM testing against the v3.x branch until we can get a fix.

[rhc54]

@jjhursey I don't know what I can do with that info - is there any way to tell us what failed in that function?

[jsquyres]

@jjhursey What's with these warning messages in the IBM CI gist:

[c656f6n04:147737] mca_base_component_repository_open: unable to open mca_state_staged_hnp: /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/openmpi/mca_state_staged_hnp.so: undefined symbol: orte_staged_execution (ignored)
[c656f6n04:147737] mca_base_component_repository_open: unable to open mca_state_staged_orted: /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/openmpi/mca_state_staged_orted.so: undefined symbol: orte_staged_execution (ignored)
[c656f6n04:147737] mca_base_component_repository_open: unable to open mca_oob_usock: /home/mpiczar/jenkins/workspace/ompi_public_pr_release_gnu/ompi-install/lib/openmpi/mca_oob_usock.so: undefined symbol: opal_sec (ignored)

[rhc54]

Those are all removed symbols, and those components should also have been removed. Maybe some mistiming of PR's?

[jjhursey]

Humm. I'm investigating. But I'm wondering if Jenkins merged this into v2.x (which has those symbols) instead of v3.x (which doesn't).

[jjhursey]

bot:ibm:gnu:retest

[jjhursey]

Ok so I found the problem in the IBM CI setup and it's fixed now. We were picking up an old install which was throwing off our release build. This looks clean now. Sorry for the noise... 😞

[hjelmn]

@jsquyres Looks like the typo is indeed here. The other pt2pt commit is needed.

[jsquyres]

Ah, ok. I thought somehow it didn't matter over here.

Will you add c72fb30 to this PR?

[hjelmn]

@jsquyres Ready to review.