[DOCKER] Update docker protocol (#2793)

apache · Mar 12, 2019 · 5e3ceaa · 5e3ceaa
1 parent f8ac138
commit 5e3ceaa
Show file tree

Hide file tree

Showing 101 changed files with 203 additions and 31 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -3,6 +3,28 @@
 // Jenkins pipeline
 // See documents at https://jenkins.io/doc/book/pipeline/jenkinsfile/
 
+// Docker env used for testing
+// Different image may have different version tag
+// because some of them are more stable than anoter.
+//
+// Docker images are maintained by PMC, cached in dockerhub
+// and remains relatively stable over the time.
+// Flow for upgrading docker env(need commiter)
+//
+// - Send PR to upgrade build script in the repo
+// - Build the new docker image
+// - Tag the docker image with a new version and push to tvmai
+// - Update the version in the Jenkinsfile, send a PR
+// - Fix any issues wrt to the new image version in the PR
+// - Merge the PR and now we are in new version
+// - Tag the new version as the lates
+// - Periodically cleanup the old versions on local workers
+//
+ci_lint = "tvmai/ci-lint:v0.50"
+ci_gpu = "tvmai/ci-gpu:v0.50"
+ci_cpu = "tvmai/ci-cpu:v0.41"
+ci_i386 = "tvmai/ci-i386:v0.50"
+
 // tvm libraries
 tvm_runtime = "build/libtvm_runtime.so, build/config.cmake"
 tvm_lib = "build/libtvm.so, " + tvm_runtime
@@ -39,7 +61,7 @@ stage("Sanity Check") {
     node('CPU') {
       ws('workspace/tvm/sanity') {
         init_git()
-        sh "${docker_run} tvmai/ci-lint  ./tests/scripts/task_lint.sh"
+        sh "${docker_run} ${ci_lint}  ./tests/scripts/task_lint.sh"
       }
     }
   }
@@ -103,7 +125,7 @@ stage('Build') {
            echo set\\(CMAKE_CXX_COMPILER g++\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
            """
-        make('tvmai/ci-gpu', 'build', '-j2')
+        make(ci_gpu, 'build', '-j2')
         pack_lib('gpu', tvm_multilib)
         // compiler test
         sh """
@@ -117,7 +139,7 @@ stage('Build') {
            echo set\\(CMAKE_CXX_COMPILER clang-6.0\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
            """
-        make('tvmai/ci-gpu', 'build2', '-j2')
+        make(ci_gpu, 'build2', '-j2')
       }
     }
   },
@@ -138,15 +160,15 @@ stage('Build') {
            echo set\\(CMAKE_CXX_COMPILER g++\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
            """
-        make('tvmai/ci-cpu', 'build', '-j2')
+        make(ci_cpu, 'build', '-j2')
         pack_lib('cpu', tvm_lib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_cpp_unittest.sh"
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_python_vta.sh"
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_rust.sh"
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_golang.sh"
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_python_unittest.sh"
-          sh "${docker_run} tvmai/ci-cpu ./tests/scripts/task_python_integration.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_cpp_unittest.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_python_vta.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_rust.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_golang.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_python_unittest.sh"
+          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_python_integration.sh"
         }
       }
     }
@@ -166,7 +188,7 @@ stage('Build') {
            echo set\\(CMAKE_CXX_COMPILER g++\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
            """
-        make('tvmai/ci-i386', 'build', '-j2')
+        make(ci_i386, 'build', '-j2')
         pack_lib('i386', tvm_multilib)
       }
     }
@@ -180,8 +202,8 @@ stage('Unit Test') {
         init_git()
         unpack_lib('gpu', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_python_unittest.sh"
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_python_integration.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_unittest.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_integration.sh"
         }
       }
     }
@@ -192,9 +214,9 @@ stage('Unit Test') {
         init_git()
         unpack_lib('i386', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-i386 ./tests/scripts/task_python_unittest.sh"
-          sh "${docker_run} tvmai/ci-i386 ./tests/scripts/task_python_integration.sh"
-          sh "${docker_run} tvmai/ci-i386 ./tests/scripts/task_python_vta.sh"
+          sh "${docker_run} ${ci_i386} ./tests/scripts/task_python_unittest.sh"
+          sh "${docker_run} ${ci_i386} ./tests/scripts/task_python_integration.sh"
+          sh "${docker_run} ${ci_i386} ./tests/scripts/task_python_vta.sh"
         }
       }
     }
@@ -205,7 +227,7 @@ stage('Unit Test') {
         init_git()
         unpack_lib('gpu', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_java_unittest.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_java_unittest.sh"
         }
       }
     }
@@ -219,7 +241,7 @@ stage('Integration Test') {
         init_git()
         unpack_lib('gpu', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_python_topi.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_topi.sh"
         }
       }
     }
@@ -230,7 +252,7 @@ stage('Integration Test') {
         init_git()
         unpack_lib('gpu', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_python_frontend.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_frontend.sh"
         }
       }
     }
@@ -241,7 +263,7 @@ stage('Integration Test') {
         init_git()
         unpack_lib('gpu', tvm_multilib)
         timeout(time: max_time, unit: 'MINUTES') {
-          sh "${docker_run} tvmai/ci-gpu ./tests/scripts/task_python_docs.sh"
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_docs.sh"
         }
         pack_lib('mydocs', 'docs.tgz')
       }

diff --git a/docker/Dockerfile.ci_cpu b/docker/Dockerfile.ci_cpu
@@ -1,4 +1,5 @@
 # CI docker CPU env
+# tag: v0.41
 FROM ubuntu:16.04
 
 RUN apt-get update --fix-missing

diff --git a/docker/Dockerfile.ci_gpu b/docker/Dockerfile.ci_gpu
@@ -1,5 +1,5 @@
 # CI docker GPU env
-# tag: v0.41
+# tag: v0.50
 FROM nvidia/cuda:8.0-cudnn7-devel
 
 # Base scripts

diff --git a/docker/Dockerfile.ci_i386 b/docker/Dockerfile.ci_i386
@@ -1,4 +1,5 @@
 # CI docker i386 env
+# tag: v0.50
 
 FROM ioft/i386-ubuntu:16.04
 

diff --git a/docker/bash.sh b/docker/bash.sh
@@ -33,6 +33,12 @@ else
     DOCKER_BINARY="docker"
 fi
 
+if [[ ! -z $CUDA_VISIBLE_DEVICES ]]; then
+    CUDA_ENV="-e CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES}"
+else
+    CUDA_ENV=""
+fi
+
 # Print arguments.
 echo "WORKSPACE: ${WORKSPACE}"
 echo "DOCKER CONTAINER NAME: ${DOCKER_IMAGE_NAME}"
@@ -53,6 +59,8 @@ ${DOCKER_BINARY} run --rm --pid=host\
     -e "CI_BUILD_UID=$(id -u)" \
     -e "CI_BUILD_GROUP=$(id -g -n)" \
     -e "CI_BUILD_GID=$(id -g)" \
+    -e "PYTHONPATH=python:topi/python"\
+    ${CUDA_ENV}\
     ${CI_DOCKER_EXTRA_PARAMS[@]} \
     ${DOCKER_IMAGE_NAME}\
     bash --login /docker/with_the_same_user \

diff --git a/docker/install/ubuntu_install_onnx.sh b/docker/install/ubuntu_install_onnx.sh
@@ -5,10 +5,7 @@ set -u
 set -o pipefail
 
 # fix to certain version for now
-pip2 install onnx>=1.1.0
 pip3 install onnx>=1.1.0
 
-pip2 install http://download.pytorch.org/whl/cu75/torch-0.2.0.post3-cp27-cp27mu-manylinux1_x86_64.whl
-pip2 install torchvision
-pip3 install http://download.pytorch.org/whl/cu75/torch-0.2.0.post3-cp35-cp35m-manylinux1_x86_64.whl
+pip3 install https://download.pytorch.org/whl/cu80/torch-1.0.1.post2-cp36-cp36m-linux_x86_64.whl
 pip3 install torchvision
diff --git a/docker/with_the_same_user b/docker/with_the_same_user
@@ -26,8 +26,15 @@ getent passwd "${CI_BUILD_UID}" || adduser --gid "${CI_BUILD_GID}" --uid "${CI_B
 usermod -a -G sudo "${CI_BUILD_USER}"
 echo "${CI_BUILD_USER} ALL=(ALL) NOPASSWD:ALL" > /etc/sudoers.d/90-nopasswd-sudo
 
+if [[ ! -z $CUDA_VISIBLE_DEVICES ]]; then
+    CUDA_ENV="CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES}"
+else
+    CUDA_ENV=""
+fi
+
 HOME=${CI_BUILD_HOME}\
     sudo -u "#${CI_BUILD_UID}" --preserve-env\
+    ${CUDA_ENV}\
     PATH=${PATH}\
     JAVA_HOME=${JAVA_HOME}\
     LD_LIBRARY_PATH=${LD_LIBRARY_PATH}\

diff --git a/docs/contribute/pull_request.rst b/docs/contribute/pull_request.rst
@@ -12,7 +12,7 @@ This is a quick guide to submit a pull request, please also refer to the detaile
     git rebase upstream/master
 
 - Make sure code style check pass by typing the following command, and all the existing test-cases pass.
-  - ``docker/bash.sh tvmai/ci-lint ./tests/scripts/task_lint.sh``  
+  - ``docker/bash.sh tvmai/ci-lint ./tests/scripts/task_lint.sh``
      (Note: You must install docker beforehand so you can run a docker image.)
 - Add test-cases to cover the new features or bugfix the patch introduces.
 - Document the code you wrote, see more at :ref:`doc_guide`
@@ -27,6 +27,28 @@ This is a quick guide to submit a pull request, please also refer to the detaile
 
 - The patch can be merged after the reviewers approve the pull request.
 
+
+CI Environment
+--------------
+We use docker container to create stable CI environments
+that can be deployed to multiple machines.
+You can find the prebuilt images in `<https://hub.docker.com/r/tvmai/>`_ .
+Because we want a relatively stable CI environment and make use of pre-cached image,
+all of the CI images are built and maintained by comitters.
+
+Upgrade of CI images can cause problems and need fixes to accommodate the new env.
+Here is the protocol to update CI image:
+
+- Send PR to upgrade build script in the repo
+  - Can be done by a contributor, the following steps need committership.
+- Build the new docker image
+- Tag the docker image with a new version and push to tvmai
+- Update the version(most of the time increase the minor version) in the Jenkinsfile, send a PR.
+- Fix any issues wrt to the new image versions in the PR.
+- Merge the PR and now we are in new version.
+- Tag the new version as the latest.
+- Periodically cleanup the old versions on local workers
+
 Testing
 -------
 Even though we have hooks to run unit tests automatically for each pull request, It's always recommended to run unit tests

diff --git a/include/tvm/api_registry.h b/include/tvm/api_registry.h
@@ -8,6 +8,7 @@
 #define TVM_API_REGISTRY_H_
 
 #include <string>
+#include <utility>
 #include "base.h"
 #include "packed_func_ext.h"
 #include "runtime/registry.h"

diff --git a/include/tvm/attrs.h b/include/tvm/attrs.h
@@ -33,6 +33,7 @@
 #include <functional>
 #include <type_traits>
 #include <string>
+#include <utility>
 #include "ir.h"
 #include "base.h"
 #include "expr.h"

diff --git a/include/tvm/build_module.h b/include/tvm/build_module.h
@@ -9,6 +9,8 @@
 #include <string>
 #include <vector>
 #include <utility>
+#include <unordered_map>
+#include <unordered_set>
 #include "runtime/packed_func.h"
 #include "schedule_pass.h"
 #include "lowered_func.h"

diff --git a/include/tvm/expr.h b/include/tvm/expr.h
@@ -10,6 +10,7 @@
 #include <ir/IRPrinter.h>
 #include <string>
 #include <algorithm>
+#include <unordered_map>
 #include "base.h"
 #include "runtime/c_runtime_api.h"
 

diff --git a/include/tvm/ir_functor_ext.h b/include/tvm/ir_functor_ext.h
@@ -8,6 +8,7 @@
 
 #include "tvm/node/ir_functor.h"
 #include "ir.h"
+#include <utility>
 
 namespace tvm {
 namespace ir {

diff --git a/include/tvm/ir_pass.h b/include/tvm/ir_pass.h
@@ -11,6 +11,7 @@
 
 #include <arithmetic/Simplify.h>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
 #include <string>
 #include "expr.h"
@@ -336,7 +337,7 @@ Stmt RewriteUnsafeSelect(Stmt stmt);
 Stmt LowerStorageAccessInfo(Stmt stmt);
 
 /*!
- * \brief Decorate the stmt with a device scope, this is helpful for 
+ * \brief Decorate the stmt with a device scope, this is helpful for
  * hardware accelerator without thread blocks.
  *
  * \param stmt The stmt to be trasnformed

diff --git a/include/tvm/relay/error.h b/include/tvm/relay/error.h
@@ -9,6 +9,7 @@
 #include <string>
 #include <vector>
 #include <sstream>
+#include <unordered_map>
 #include "./base.h"
 #include "./expr.h"
 #include "./module.h"

diff --git a/include/tvm/relay/expr_functor.h b/include/tvm/relay/expr_functor.h
@@ -9,6 +9,8 @@
 
 #include <tvm/node/ir_functor.h>
 #include <string>
+#include <utility>
+#include <unordered_map>
 #include "./expr.h"
 #include "./adt.h"
 #include "./op.h"

diff --git a/include/tvm/relay/pattern_functor.h b/include/tvm/relay/pattern_functor.h
@@ -9,6 +9,8 @@
 
 #include <tvm/node/ir_functor.h>
 #include <string>
+#include <utility>
+#include <unordered_map>
 #include "./expr.h"
 #include "./op.h"
 #include "./error.h"

diff --git a/include/tvm/runtime/packed_func.h b/include/tvm/runtime/packed_func.h
@@ -16,6 +16,7 @@
 #include <string>
 #include <limits>
 #include <memory>
+#include <utility>
 #include <type_traits>
 #include "c_runtime_api.h"
 #include "module.h"

diff --git a/include/tvm/schedule.h b/include/tvm/schedule.h
@@ -7,6 +7,7 @@
 #define TVM_SCHEDULE_H_
 
 #include <string>
+#include <unordered_map>
 #include "base.h"
 #include "expr.h"
 #include "tensor.h"

diff --git a/include/tvm/tensor.h b/include/tvm/tensor.h
@@ -10,6 +10,7 @@
 #include <tvm/node/container.h>
 #include <string>
 #include <vector>
+#include <utility>
 #include <type_traits>
 
 #include "base.h"

diff --git a/nnvm/include/nnvm/graph.h b/nnvm/include/nnvm/graph.h
@@ -10,6 +10,7 @@
 #include <string>
 #include <utility>
 #include <algorithm>
+#include <memory>
 #include <unordered_map>
 #include <unordered_set>
 #include "base.h"

diff --git a/nnvm/include/nnvm/graph_attr_types.h b/nnvm/include/nnvm/graph_attr_types.h
@@ -8,6 +8,7 @@
 
 #include <vector>
 #include <string>
+#include <unordered_map>
 #include "tuple.h"
 #include "layout.h"
 

diff --git a/nnvm/include/nnvm/node.h b/nnvm/include/nnvm/node.h
@@ -9,6 +9,7 @@
 #include <memory>
 #include <string>
 #include <vector>
+#include <utility>
 #include <unordered_map>
 #include "base.h"
 #include "op.h"

diff --git a/nnvm/include/nnvm/op_attr_types.h b/nnvm/include/nnvm/op_attr_types.h
@@ -10,6 +10,7 @@
 #include <string>
 #include <utility>
 #include <functional>
+#include <unordered_map>
 #include "base.h"
 #include "node.h"
 #include "tuple.h"