Closed omus closed 1 year ago
omus
commented
1 year ago After doing a self-review I think there's something I can do to greatly simplify the code
Update: There was some room for improvement but not quite as much as I originally thought. We now no longer use a fixed port for workers and determining the pod IP address is a much simpler operation.
codecov[bot]
commented
1 year ago Merging #108 (aeec2e3) into main (3fc33a8) will increase coverage by
22.58%. The diff coverage is79.06%.
@@ Coverage Diff @@
## main #108 +/- ##
===========================================
+ Coverage 65.85% 88.44% +22.58%
===========================================
Files 4 4
Lines 164 199 +35
===========================================
+ Hits 108 176 +68
+ Misses 56 23 -33 | Impacted Files | Coverage Δ | |
|---|---|---|
| src/K8sClusterManagers.jl | 66.66% <ø> (ø) |
|
| src/native_driver.jl | 78.94% <79.06%> (+67.28%) |
:arrow_up: |
:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more
Replaces #94. Adds support for having the Julia cluster manager running externally from the workers running inside a cluster while retaining support for in-cluster managers.
Julia's Distributed package isn't well suited for the heterogeneous network connections we're using here so I was forced to extend the
Distributed.connectfunction and modify it's behaviour for this particular cluster manager. In particular manager-to-worker connections need to use the special port-forwarded addresses while workers can continue to use intra cluster network connections.