Index: scripts/slave/recipe_modules/gpu/api.py |
diff --git a/scripts/slave/recipe_modules/gpu/api.py b/scripts/slave/recipe_modules/gpu/api.py |
index 15f101716a78f731ef707897c00480963f6be455..83146419a78bab34a772fcfe36e7a223ee4e05b9 100644 |
--- a/scripts/slave/recipe_modules/gpu/api.py |
+++ b/scripts/slave/recipe_modules/gpu/api.py |
@@ -137,11 +137,32 @@ class GpuApi(recipe_api.RecipeApi): |
]) |
# Pixel tests. |
+ # Try servers pull their results from cloud storage; the other |
+ # tester bots send their results to cloud storage. |
+ # |
+ # NOTE that ALL of the bots need to share a bucket. They can't be split |
+ # by mastername/waterfall, because the try servers are on a different |
+ # waterfall (tryserver.chromium) than the other test bots (chromium.gpu |
+ # and chromium.webkit, as of this writing). This means there will be |
+ # races between bots with identical OS/GPU combinations, on different |
+ # waterfalls, attempting to upload results for new versions of each |
+ # pixel test. If this is a significant problem in practice then we will |
+ # have to rethink the cloud storage code in the pixel tests. |
+ ref_img_arg = '--upload-refimg-to-cloud-storage' |
+ if 'rietveld' in self.m.properties: |
+ ref_img_arg = '--download-refimg-from-cloud-storage' |
+ cloud_storage_bucket = 'chromium-gpu-archive/reference-images' |
yield self.run_telemetry_gpu_test('pixel_test', |
args=[ |
- '--generated-dir=%s' % self._generated_dir, |
- '--reference-dir=%s' % self._reference_dir, |
- '--build-revision=%s' % self._build_revision, |
+ '--build-revision', |
+ str(self._build_revision), |
+ ref_img_arg, |
+ '--refimg-cloud-storage-bucket', |
+ cloud_storage_bucket, |
+ '--os-type', |
+ self.m.chromium.c.TARGET_PLATFORM, |
+ '--test-machine-name', |
+ self.m.properties['buildername'] |
]) |
# Archive telemetry pixel test results |