spanner

🔧 spanner 🔧

spanner is a tool for easy job managment on HPC like referencing by name, logging, logging into nodes.

Works across Slurm and PBSPro.

Easy installation using Mamba

Install MambaForge on your cluster. In case you don't have internet access there, you can use mitten (https://github.com/unkaktus/mitten).

Then, install spanner itself:

mamba install -c https://mamba.unkaktus.art spanner

Installation using Go

1. Install MambaForge on your cluster. In case you don't have internet access there, you can use mitten (https://github.com/unkaktus/mitten).

2. Install Go

mamba install go

1. Install spanner:

go install github.com/unkaktus/spanner/cmd/spanner@latest

1. Add $HOME/go/bin into your $PATH into your .bashrc:

export PATH="$HOME/go/bin:$PATH"

Manual building

1. Install Go (https://go.dev)

2. Build spanner for Linux:

git clone https://github.com/unkaktus/spanner
cd spanner/cmd/spanner
env GOOS=linux GOARCH=amd64 go build

1. scp the spanner binary to your favorite supercomp and add it to your $PATH.

Example uses

List jobs

$ spanner list
╭───────────────────────┬─────────┬───────┬─────────────────────────┬───────╮
│         NAME          │  STATE  │ QUEUE │          TIME           │ NODES │
├───────────────────────┼─────────┼───────┼─────────────────────────┼───────┤
│ Compare_Apples        │ R [0]   │ small │ [8%] 2h0m41s/24h0m0s    │     8 │
│ Compare_Oranges       │ Q [0]   │ small │ [0%] 0s/20h0m0s         │     2 │
│ Compare_Bananas       │ F [9]   │ small │ [0%] 0s/20h0m0s         │    16 │
╰───────────────────────┴─────────┴───────┴─────────────────────────┴───────╯

Logs

Open full logs in $EDITOR (defauts to vim):

$ spanner logs Compare_Apples

Follow the log tail of a job:

$ spanner logs -f Compare_Apples

Shell

To connect to the shell on the job nodes, you first need to start your job binary via spanner tent:

[mpirun -n 16] spanner tent ./exe/binary

Then, to connect to the shell of the node 1 of running job Compare_Apples:

$ spanner shell Compare_Apples 1
node123$

Stopping jobs

Cancel job Compare_Apples:

$ spanner cancel Compare_Apples

Portable jobs

Start a portable job using compare_apples.begin file and configuration file data.csv for the run:

$ spanner begin -f compare_apples.begin data.csv

Remote commands

Do all above without logging manually to the cluster:

$ spanner on supercomp list

$ spanner on supercomp shell compare_apples

This requires to have spanner to be installed and added to the PATH there.

Port forwarding

Forward a port to the node of a job:

$ spanner port-forward -p 11111 -m supercomp compare_apples