fixes

Author: kleinhenz

ci/Dockerfile

@@ -1,5 +1,4 @@
-FROM ghcr.io/kleinhenz/docker-slurm-cluster@sha256:ed26ab967e84c955f75499d50ef7a4f844e48d7812de5ef6b64423728d9d10c2 as base
-MAINTAINER Joseph Kleinhenz <jkleinh@umich.edu>
+FROM ghcr.io/kleinhenz/docker-slurm-cluster@sha256:c62b169970eaab879898a7df5950f888b0417147e06f8c0e32fd15087b22d9f2
 
 ARG JULIA_VERSION=1.11.2
 

ci/docker-compose.yml

@@ -1,8 +1,9 @@
-version: "3.3"
-
 services:
   slurmctld:
     image: slurm-cluster-julia
+    build:
+      context: ..
+      dockerfile: ci/Dockerfile
     command: ["slurmctld"]
     container_name: slurmctld
     hostname: slurmctld

test/runtests.jl

@@ -10,15 +10,16 @@ using Distributed, Test, SlurmClusterManager
 project_path = abspath(joinpath(@__DIR__, ".."))
 println("project_path = $project_path")
 jobid = withenv("JULIA_PROJECT"=>project_path) do
-  read(`sbatch --export=ALL --parsable -n 4 -o test.out script.jl`, String)
+  strip(read(`sbatch --export=ALL --parsable -n 4 -o test.out script.jl`, String))
 end
 println("jobid = $jobid")
 
 # get job state from jobid
 getjobstate = jobid -> begin
-  info = read(Cmd(`scontrol show jobid=$jobid`, ignorestatus=true), String)
+  cmd = Cmd(`scontrol show jobid=$jobid`, ignorestatus=true)
+  info = read(cmd, String)
   state = match(r"JobState=(\S*)", info)
-  return state
+  return isnothing(state) ? nothing : state.captures[1]
 end
 
 # wait for job to complete
@@ -29,17 +30,13 @@ status = timedwait(60.0, pollint=1.0) do
   return state == "COMPLETED" || state == "FAILED"
 end
 
+state = getjobstate(jobid)
+
 # check that job finished running within timelimit (either completed or failed)
 @test status == :ok
+@test state == "COMPLETED"
 
 # print job output
 output = read("test.out", String)
 println("script output:")
 println(output)
-
-state = getjobstate(jobid) |> split
-# length should be two because creating the workers creates a job step
-@test length(state) == 2
-
-# check that everything exited without errors
-@test all(state .== "COMPLETED")

test/script.jl

@@ -9,5 +9,6 @@ hosts = map(workers()) do id
   remotecall_fetch(() -> gethostname(), id)
 end
 sort!(hosts)
+println(hosts)
 
 @assert hosts == ["c1", "c1", "c2", "c2"]