ci(sdk): build wheels on musllinux

.circleci/config.yml

@@ -639,7 +639,7 @@ jobs:
           name: Run wandb core's Go tests and collect coverage
           command: |
             cd core
-            go test -race -coverprofile=coverage.txt -covermode=atomic ./...
+            go test -tags=nomusl -race -coverprofile=coverage.txt -covermode=atomic ./...
       - upload_codecov: { flags: unit }
 
   store-local-testcontainer:
@@ -667,7 +667,7 @@ jobs:
           name:
           command: |
             go install github.com/google/go-containerregistry/cmd/gcrane@latest
-            pip install -U pip nox requests
+            pip install -U pip nox requests uv
             nox -s local-testcontainer-registry
           no_output_timeout: 5m
 

.github/workflows/release-sdk.yml

@@ -112,10 +112,8 @@ jobs:
           # In cp36-*, the wheel name sometimes includes additional dashes that
           # make it invalid, breaking the job.
           #
-          # Not sure why we skip PyPy and musllinux builds.
-          #
           # See https://cibuildwheel.readthedocs.io/en/stable/options/#build-skip
-          CIBW_SKIP: cp36-* pp* *musllinux*
+          CIBW_SKIP: pp* *manylinux*
           CIBW_ARCHS_LINUX: x86_64 aarch64
           CIBW_ARCHS_MACOS: x86_64 arm64  # arm64 == aarch64
 
@@ -141,9 +139,12 @@ jobs:
             esac &&
             export DOWNLOAD_GOVERSION=$( grep '^go' core/go.mod | cut -d' ' -f2 ) &&
             curl -L https://golang.org/dl/go$DOWNLOAD_GOVERSION.linux-$DOWNLOAD_GOARCH.tar.gz > go.tar.gz &&
-            tar -C /usr/local/ -xzf go.tar.gz
+            tar -C /usr/local/ -xzf go.tar.gz &&
+            env
           CIBW_ENVIRONMENT_LINUX:
             PATH=$PATH:/usr/local/go/bin
+          # TODO: fixme
+          CIBW_REPAIR_WHEEL_COMMAND_LINUX: ""
 
       - uses: actions/upload-artifact@v3
         with:

core/hatch.py

@@ -25,7 +25,15 @@ def build_wandb_core(
             is the https://github.com/wandb/wandb repository. Otherwise, an
             empty string.
     """
+    print(os.environ)
     coverage_flags = ["-cover"] if with_code_coverage else []
+    # TODO: the musl tag is used to skip building nvidia gpu monitoring on musl-based systems
+    musl_tag = [
+        "-tags",
+        "musl"
+        if os.environ.get("AUDITWHEEL_PLAT", "").startswith("musl")
+        else "nomusl",
+    ]
     output_flags = ["-o", str(".." / output_path)]
     ld_flags = [f"-ldflags={_go_linker_flags(wandb_commit_sha)}"]
 
@@ -35,6 +43,7 @@ def build_wandb_core(
         [
             str(go_binary),
             "build",
+            *musl_tag,
             *coverage_flags,
             *ld_flags,
             *output_flags,

core/pkg/monitor/noop.go → core/pkg/monitor/gpu_amd_noop.go

@@ -6,36 +6,6 @@ import (
 	"github.com/wandb/wandb/core/pkg/service"
 )
 
-type GPUNvidia struct {
-	name     string
-	settings *service.Settings
-}
-
-func NewGPUNvidia(settings *service.Settings) *GPUNvidia {
-	gpu := &GPUNvidia{
-		name:     "gpu",
-		settings: settings,
-	}
-
-	return gpu
-}
-
-func (g *GPUNvidia) Name() string { return g.name }
-
-func (g *GPUNvidia) SampleMetrics() {}
-
-func (g *GPUNvidia) AggregateMetrics() map[string]float64 {
-	return map[string]float64{}
-}
-
-func (g *GPUNvidia) ClearMetrics() {}
-
-func (g *GPUNvidia) IsAvailable() bool { return false }
-
-func (g *GPUNvidia) Probe() *service.MetadataRequest {
-	return nil
-}
-
 type GPUAMD struct {
 	name     string
 	settings *service.Settings

core/pkg/monitor/gpu_nvidia.go

@@ -1,4 +1,4 @@
-//go:build linux && !libwandb_core
+//go:build linux && nomusl && !libwandb_core
 
 package monitor
 

core/pkg/monitor/gpu_nvidia_noop.go

@@ -0,0 +1,37 @@
+//go:build !linux || (linux && musl) || libwandb_core
+
+package monitor
+
+import (
+	"github.com/wandb/wandb/core/pkg/service"
+)
+
+type GPUNvidia struct {
+	name     string
+	settings *service.Settings
+}
+
+func NewGPUNvidia(settings *service.Settings) *GPUNvidia {
+	gpu := &GPUNvidia{
+		name:     "gpu",
+		settings: settings,
+	}
+
+	return gpu
+}
+
+func (g *GPUNvidia) Name() string { return g.name }
+
+func (g *GPUNvidia) SampleMetrics() {}
+
+func (g *GPUNvidia) AggregateMetrics() map[string]float64 {
+	return map[string]float64{}
+}
+
+func (g *GPUNvidia) ClearMetrics() {}
+
+func (g *GPUNvidia) IsAvailable() bool { return false }
+
+func (g *GPUNvidia) Probe() *service.MetadataRequest {
+	return nil
+}