oci-hpc

oci-hpc copied to clipboard

improve error handling, double deletion

1

in case when terraform provisioning fails due to cluster network being only partially provisioned the terraform doesn't register it and thus is not removed during deletion adds double delete for...

hrvolapeter

Hello, the installation of oracle-cloud-agent fails because the bucket url is not accessible to public. ``` yum: name: "https://objectstorage.us-phoenix-1.oraclecloud.com/p/aV_mSl96KIiapAeZtsyo-SUcPCSurDfWaj06f4XVVoNKIsxvqlZ65guPTnMuNawR/n/imagegen/b/agent_test/o/1.37.0/3/oracle-cloud-agent-1.37.2-10459.el8.x86_64.rpm" state: present disable_gpg_check: yes ``` ``` {"code":"BucketNotFound","message":"Either the bucket named 'agent_test'...

hrvolapeter

enhancement to support ubuntu in computing node

6

for HPC field. sometimes customer will use ubuntu to deploy computing node, and I failed to test ubuntu as computing node and found the current version doesn't support it. do...

davinwei

Slurm enters provisioning loop even when resources are unavailable

1

Not sure if this can be resolved, but I wonder if it would be possible to check if the nodes are available before provisioning the cluster network rather than provisioning...

cbutakoff

Potential problem handling array jobs

2

I have an array job limited to 2 jobs at a time: ``` 2145_[4-190%2] compute EP_108 opc PD 0:00 10 (JobArrayTaskLimit) 2145_3 compute EP_108 opc R 48:42 10 compute-hpc-node-[100,373,397,421,425,429,455,457,813,896] 2145_2...

cbutakoff

By default it seems that the only pmi installed in the cluster is mpich. It would be helpful to add also intel compatible pmi. I assume installing slurm-libpmi will help....

cbutakoff

See [source](https://bugs.schedmd.com/show_bug.cgi?id=3941#c7) where they explain the default for UnkillableStepTimeout=60 which doesn't fit with the plugin.

xksteven

Fix typo

1

uthark

Explicitly specify providers that are used

1

To prevent accidental breaks because provider versions are not pinned.

uthark

If the monitoring variable is false the cron job will schedule to run an empty command very often. This is to comment out the entire line as is done with...

xksteven