Closed omus closed 3 years ago
Really glad I added those Julia 1.3 tests in. I used a syntax that's only available on Julia 1.5+
I'll work through the CI failures tomorrow as I'm out of time for today. I'll note that I'm not seeing them locally.
I should have fixed the CI issue now. I seem to be unable to re-use the KuberContext
in environments where I can't successfully connect to the Kubernetes API: https://github.com/JuliaComputing/Kuber.jl/issues/27
Merging #31 (87ef696) into main (8ede3ab) will decrease coverage by
13.02%
. The diff coverage is12.50%
.
@@ Coverage Diff @@
## main #31 +/- ##
===========================================
- Coverage 40.70% 27.67% -13.03%
===========================================
Files 4 4
Lines 113 112 -1
===========================================
- Hits 46 31 -15
- Misses 67 81 +14
Impacted Files | Coverage Δ | |
---|---|---|
src/K8sClusterManagers.jl | 81.81% <ø> (-1.52%) |
:arrow_down: |
src/native_driver.jl | 4.61% <11.53%> (-3.45%) |
:arrow_down: |
src/pod.jl | 15.00% <14.28%> (-50.22%) |
:arrow_down: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update 8ede3ab...87ef696. Read the comment docs.
I've updated codecov to use main
as the origin/HEAD
.
I'm going to tackle test coverage in another PR
Fixes https://github.com/beacon-biosignals/K8sClusterManagers.jl/issues/13
Overview of the changes:
K8sNativeManager
toK8sClusterManager
.addprocs
function typically handles specifying defaults forexename
andexeflags
I needed to move the worker pod spec generation into thelaunch
method. Doing this required changes to what was stored within theK8sClusterManager
struct.image
while running the code outside of a pod an exception will be thrown. This check is now done in theK8sClusterManager
constructor to avoid having the exception raised inside of a task (howlaunch
is called).K8sClusterManager
constructor that takes number of worker to launchNamedTuple
. Should allow for additions without breaking the code.KuberContext
instance while testing (reduces noise)kill
andmanage
methods for our cluster manager as we're not adding value. The defaultkill
will remotely callexit
on the workers which allows the worker pods to complete instead of always error.