Add support for Google Batch used specified boot images (#5268)

The default boot image is batch-cos, but it may not be desired in all situations. In particular, there is an ongoing issue in which the batch-cos image does not retry pulling a docker image if there was a network issue. The user may also have some some other configuration pre-configured in their custom boot image. This commit adds the ability to specify a custom boot disk image by using the configuration option ``` google.batch.bootDiskImage = '<NAME>' ``` Signed-off-by: Siddhartha Bagaria <[email protected]> Signed-off-by: Paolo Di Tommaso <[email protected]> Co-authored-by: Paolo Di Tommaso <[email protected]>
nextflow-io · Sep 3, 2024 · 0aaa648 · 0aaa648
1 parent b3ba2c2
commit 0aaa648
Show file tree

Hide file tree

Showing 5 changed files with 27 additions and 7 deletions.
diff --git a/docs/config.md b/docs/config.md
@@ -886,6 +886,11 @@ The following settings are available for Google Cloud Batch:
   :::
 : Define the set of allowed locations for VMs to be provisioned. See [Google documentation](https://cloud.google.com/batch/docs/reference/rest/v1/projects.locations.jobs#locationpolicy) for details (default: no restriction).
 
+`google.batch.bootDiskImage`
+: :::{versionadded} 24.08.0-edge
+  :::
+: Set the image URI of the virtual machine boot disk, e.g `batch-debian`. See [Google documentation](https://cloud.google.com/batch/docs/vm-os-environment-overview#vm-os-image-options) for details (default: none).
+
 `google.batch.bootDiskSize`
 : Set the size of the virtual machine boot disk, e.g `50.GB` (default: none).
 

diff --git a/plugins/nf-google/src/main/nextflow/cloud/google/batch/GoogleBatchTaskHandler.groovy b/plugins/nf-google/src/main/nextflow/cloud/google/batch/GoogleBatchTaskHandler.groovy
@@ -332,6 +332,9 @@ class GoogleBatchTaskHandler extends TaskHandler implements FusionAwareTask {
                 instancePolicyOrTemplate.setInstallGpuDrivers(true)
             }
 
+            if( executor.config.getBootDiskImage() )
+                instancePolicy.setBootDisk( AllocationPolicy.Disk.newBuilder().setImage( executor.config.getBootDiskImage() ) )
+
             if( fusionEnabled() && !disk ) {
                 disk = new DiskResource(request: '375 GB', type: 'local-ssd')
                 log.debug "[GOOGLE BATCH] Process `${task.lazyName()}` - adding local volume as fusion scratch: $disk"

diff --git a/plugins/nf-google/src/main/nextflow/cloud/google/batch/client/BatchConfig.groovy b/plugins/nf-google/src/main/nextflow/cloud/google/batch/client/BatchConfig.groovy
@@ -38,6 +38,7 @@ class BatchConfig {
     private GoogleOpts googleOpts
     private GoogleCredentials credentials
     private List<String> allowedLocations
+    private String bootDiskImage
     private MemoryUnit bootDiskSize
     private String cpuPlatform
     private int maxSpotAttempts
@@ -54,6 +55,7 @@ class BatchConfig {
     GoogleOpts getGoogleOpts() { return googleOpts }
     GoogleCredentials getCredentials() { return credentials }
     List<String> getAllowedLocations() { allowedLocations }
+    String getBootDiskImage() { bootDiskImage }
     MemoryUnit getBootDiskSize() { bootDiskSize }
     String getCpuPlatform() { cpuPlatform }
     int getMaxSpotAttempts() { maxSpotAttempts }
@@ -72,6 +74,7 @@ class BatchConfig {
         result.googleOpts = GoogleOpts.create(session)
         result.credentials = result.googleOpts.credentials
         result.allowedLocations = session.config.navigate('google.batch.allowedLocations', List.of()) as List<String>
+        result.bootDiskImage = session.config.navigate('google.batch.bootDiskImage')
         result.bootDiskSize = session.config.navigate('google.batch.bootDiskSize') as MemoryUnit
         result.cpuPlatform = session.config.navigate('google.batch.cpuPlatform')
         result.maxSpotAttempts = session.config.navigate('google.batch.maxSpotAttempts', DEFAULT_MAX_SPOT_ATTEMPTS) as int

diff --git a/plugins/nf-google/src/test/nextflow/cloud/google/batch/GoogleBatchTaskHandlerTest.groovy b/plugins/nf-google/src/test/nextflow/cloud/google/batch/GoogleBatchTaskHandlerTest.groovy
@@ -105,6 +105,7 @@ class GoogleBatchTaskHandlerTest extends Specification {
         !instancePolicy.getMachineType()
         !instancePolicy.getMinCpuPlatform()
         instancePolicy.getProvisioningModel().toString() == 'PROVISIONING_MODEL_UNSPECIFIED'
+        !instancePolicy.getBootDisk().getImage()
         and:
         allocationPolicy.getLocation().getAllowedLocationsCount() == 0
         allocationPolicy.getNetwork().getNetworkInterfacesCount() == 0
@@ -121,6 +122,7 @@ class GoogleBatchTaskHandlerTest extends Specification {
         and:
         def ACCELERATOR = new AcceleratorResource(request: 1, type: 'nvidia-tesla-v100')
         def BOOT_DISK = MemoryUnit.of('10 GB')
+        def BOOT_IMAGE = 'batch-debian'
         def CONTAINER_IMAGE = 'ubuntu:22.1'
         def CONTAINER_OPTS = '--this --that'
         def CPU_PLATFORM = 'Intel Skylake'
@@ -134,6 +136,7 @@ class GoogleBatchTaskHandlerTest extends Specification {
             getConfig() >> Mock(BatchConfig) {
                 getAllowedLocations() >> ['zones/us-central1-a', 'zones/us-central1-c']
                 getBootDiskSize() >> BOOT_DISK
+                getBootDiskImage() >> BOOT_IMAGE
                 getCpuPlatform() >> CPU_PLATFORM
                 getMaxSpotAttempts() >> 5
                 getSpot() >> true
@@ -211,6 +214,7 @@ class GoogleBatchTaskHandlerTest extends Specification {
         and:
         instancePolicy.getAccelerators(0).getCount() == 1
         instancePolicy.getAccelerators(0).getType() == ACCELERATOR.type
+        instancePolicy.getBootDisk().getImage() == BOOT_IMAGE
         instancePolicy.getDisks(0).getNewDisk().getSizeGb() == DISK.request.toGiga()
         instancePolicy.getDisks(0).getNewDisk().getType() == DISK.type
         instancePolicy.getMachineType() == MACHINE_TYPE

diff --git a/plugins/nf-google/src/test/nextflow/cloud/google/batch/client/BatchConfigTest.groovy b/plugins/nf-google/src/test/nextflow/cloud/google/batch/client/BatchConfigTest.groovy
@@ -18,6 +18,7 @@
 package nextflow.cloud.google.batch.client
 
 import nextflow.Session
+import nextflow.util.MemoryUnit
 import spock.lang.Requires
 import spock.lang.Specification
 /**
@@ -29,21 +30,20 @@ class BatchConfigTest extends Specification {
     @Requires({System.getenv('GOOGLE_APPLICATION_CREDENTIALS')})
     def 'should create batch config' () {
         given:
-        def CONFIG = [google: [
-            batch: [
-                spot: true
-            ]
-        ] ]
+        def CONFIG = [:]
         def session = Mock(Session) { getConfig()>>CONFIG }
 
         when:
         def config = BatchConfig.create(session)
         then:
-        config.getSpot()
+        !config.getSpot()
         and:
         config.retryConfig.maxAttempts == 5
         config.maxSpotAttempts == 0
         config.autoRetryExitCodes == [50001]
+        and:
+        !config.bootDiskImage
+        !config.bootDiskSize
     }
 
     @Requires({System.getenv('GOOGLE_APPLICATION_CREDENTIALS')})
@@ -54,7 +54,9 @@ class BatchConfigTest extends Specification {
                 spot: true,
                 maxSpotAttempts: 8,
                 autoRetryExitCodes: [50001, 50003, 50005],
-                retryPolicy: [maxAttempts: 10]
+                retryPolicy: [maxAttempts: 10],
+                bootDiskImage: 'batch-foo',
+                bootDiskSize: '100GB'
             ]
         ] ]
         def session = Mock(Session) { getConfig()>>CONFIG }
@@ -67,6 +69,9 @@ class BatchConfigTest extends Specification {
         config.retryConfig.maxAttempts == 10
         config.maxSpotAttempts == 8
         config.autoRetryExitCodes == [50001, 50003, 50005]
+        and:
+        config.bootDiskImage == 'batch-foo'
+        config.bootDiskSize == MemoryUnit.of('100GB')
     }
 
 }